Computational Art Therapy 9780398091774, 9780398091781

Citation preview

Seong-in Kim

By

SEONG-IN KIM, Ph.D. Professor Emeritus School of Industrial Management Engineering Korea University Seoul, South Korea

Published and Distributed Throughout the World by CHARLES C THOMAS • PUBLISHER, LTD. 2600 South First Street Springfield, Illinois 62704

This book is protected by copyright. No part of it may be reproduced in any manner without written permission from the publisher. All rights reserved. © 2017 by CHARLES C THOMAS • PUBLISHER, LTD. ISBN 978-0-398-09177-4 (Hard) ISBN 978-0-398-09178-1 (Ebook)

With THOMAS BOOKS careful attention is given to all details of manufacturing and design. It is the Publisher’s desire to present books that are satisfactory as to their physical qualities and artistic possibilities and appropriate for their particular use. THOMAS BOOKS will be true to those laws of quality that assure a good name and good will. Printed in the United States of America SK-C-1 Library of Congress Cataloging-in-Publication Data Names: Kim, Seong-in, author. Title: Computational art therapy / by Seong-in Kim, PH.D., professor emeritus, School of Industrial Management Engineering, Korea Univer sity, Seoul, South Korea. Description: Springfield, Illinois : Charles C Thomas, Publisher, LTD., [2017] | Includes bibliographical references and index. Identifiers: LCCN 2017015868 (print) | LCCN 2017018568 (ebook) | ISBN 9780398091781 (ebook) | ISBN 9780398091774 (hard) Subjects: LCSH: Art therapy--Methodology. | Arts--Therapeutic use. | Computer art. | Psychotherapy. Classification: LCC RC489.A7 (ebook) | LCC RC489.A7 K485 2017 (print) | DDC 616.89/1656--dc23 LC record available at https://lccn.loc.gov/2017015868

To my parents who were ever so proud of me, Honorable Yun-Haeng Kim, Justice of the Supreme Court of Korea, Eung-Kyu Jung, and my wife who loves me so much, Myeong-hee.

About the Author

Seong-in Kim, Ph.D. Professor Emeritus School of Industrial Management Engineering Korea University Lover of wine and dogs [email protected] www.kimscatscoms.com Seong-in Kim, professor of Industrial Engineering, at Korea University since 1979. He received his B.S. degrees in Economics (1970) and Applied Mathematics (1973) from Seoul National University, and his M.S. (1975) and Ph.D. (1979) in Industrial Engineering from the Korea Advanced Institute of Science and Technology (KAIST). He was a post-doctoral researcher (1981-1982) at Virginia Polytechnic Institute and State University in Blacksburg, Virginia. He served as the Dean of Academic Affairs, the Dean of Admissions Office, and the Director of Computing and Information Center at Korea University. He was Outside Director of Young-Poong Co. and President of the Korean Institute of Industrial Engineers (KIIE). He has produced 17 PhDs and 77 MAs. His research interests include applied statistics, artificial intelligence, and quality control. He developed a computer sentencing system in criminal cases, and recently he has focused on developing computer systems in art therapy. He is the author of several books on statistics and statistical quality control, Statistical Inference on Bernoulli Trials, DaeWoo Series, 1995, Statistical Quality Control, Parkyoung-sa, 1990, and Quality Control in Service Industry, Chong-moongak, 1995. His articles have appeared in journals including Technometrics, Communications in Statistics, Journal of Quality Technology, Statistics and Probability Letter, Transportation Science, Operations Research Letters, Computer and Operations Research, Naval Research Logistics, Discrete Applied Mathematics, Expert Systems with Applications, Photonic Network vii

viii

About the Author

Communications, Telecommunication Systems, Networks, Industrial Engineering Transactions, Global Optimization in Engineering Design, Industrial Engineering, Rutgers Journal of Computers and Law, The Arts in Psychotherapy, Expert Systems Research Trends, Hydrometallurgy, Metallurgical and Materials Transactions B, Art Therapy: Journal of American Art Therapy Association, and New Ideas in Psychology. He wrote two book chapters in Expert systems research trends and The Wiley handbook of art therapy. He holds 14 patents on computer sentencing and computational art therapy registered in Korea, Japan, and the U.S.A. He has published several newspaper articles on criticizing the current Korean college entrance system including articles in Joong-ang-il-bo (9, July, 2005), Dong-a-il-bo (25, November, 2000), Cho-sun-il-bo (14, November, 1998), etc. He is the recipient of the Best Paper Award from KIIE (1996), the Award for Excellent Paper in Science and Technology from the Korean Federation of Science and Technology Societies (1992), and the Best Research Award from the Korean Art Therapy Association (2007). He developed commercial software packages such as Statistical Process Control and Acceptance Sampling, which have been integrated as modules of Samsung’s UniERP. His laboratory, the AAALab (Artificial intelligence, Applied statistics, and Art therapy Lab), conducts research on interdisciplinary studies incorporating various fields such as computer science, law, statistics, art therapy, etc. Currently he heads Kims_CATS_COMS, a corporation that has developed the Computer Art Therapy apps as well as Computer Sentencing System web.

Acknowledgments

My training and experience in interdisciplinary subjects - economics, mathematics, and engineering - are what paved the way for me to write this book. It all began with the support and encouragement from my late parents. In the early 1970s, my father, who was then a Justice of the Supreme Court of South Korea, and I together developed a model for computer-aided sentencing guidelines, which awaits to be further developed. I had defied his wish for me to follow in his footsteps in the field of law but it was not all lost. His wish skipped a generation: my younger son entered the legal field and as if by fate, my older son became a professor in artificial intelligence. My greatest wish for my two sons, whom I am deeply proud of and love very much, is that they together improve another area of interdisciplinary work, the computerized sentencing system on criminal cases which their father and grandfather had developed. It will be the most honorable project, linking three generations, that belatedly fulfills their grandfather’s wish that their father fell short of. Soon after I started working on this book, I had to battle cancer, which is now in remission. My wife, Myeonghee, stood beside me as she had done for the past 44 years. This book would not have been possible without her. Thank you and I love you more than all the wine and dogs in the universe. Michael Payne Thomas, the president of my publisher, Charles C Thomas Publisher, allowed me many extensions in the past six years with continuous encouragement and sincere wishes for my speedy recovery. Thank you again, and although this specific topic text might not be widely popular among various readers, I hope this book can become a bestseller in the field of art therapy for opening a new horizon of emerging academic areas. To Youngho Lee, his continuous push for me to finish the book is what kept me going in the final chapters. Without his inspiring this book would still be in drafts, collecting dust. Seok Yoo’s occasional progress in developing app and web systems for art therapy and computer sentencing replenished my energy when I was in lack of it. ix

x

Acknowledgments

Wae Sun Choi, formal president of Korean Arts Therapy Association, and Rubin Cruz, editor of The Arts in Psychotherapy, gave me the opportunity to venture into the world of art therapy. They encouraged me along the way and published my research papers overriding the resistance from the pessimists. I still cannot forget the thrill I felt when I presented this new concept of Computational Art Therapy (CAT) at conferences in the U.S.A. as well as in Korea. Linda Gantt and Donna Betts came to every presentation of mine at many conferences with strong encouragement. It was a great pleasure that, together with Sarah Deaver, the four of us co-authored and presented our common research interests. Cathy Malchiodi, among many art therapists, showed great interest and anticipated further development of the CAT. All the attention they gave to my work almost made me think of myself as a burgeoning art therapist. Subsequently, constructive critiques and invaluable suggestions from Lynn Kapitan, editor of Art Therapy: Journal of American Art Therapy Association, and Dave Gussak, editor of The Wiley Handbook of Art Therapy, on my manuscripts gave me a sort of reality check, reminding me that I am an engineer interested in the CAT; after all, I am only a novice in the field of art therapy. This book is based on many published research papers. Hyung-Seok Kang, Ji-Ho Ghil, and Jong-Hoon Kim, the three musketeers, continued to work with me beyond my retirement from Korea University, participating in computerized analysis as well as software development. Especially, Jong-Hoon, the last doctorate under my guidance and my companion of kimscatscoms (Kim’s Computational Art Therapy System and COMputer Sentencing) co., has given his all in the whole process of books, design, illustrations, tables and even making InDesign files, and now he is thinking about extending the company to kimscatscomspubs (PUBlication System). Also, I thank his fiancée, So-Hwa Son, for the beautiful cover. Hyuk Kwon, a winner of world hackers’ competition, finally wrapped up the software package. Jun Bae, Youn-Hee Kim, Youn-Joo Oh, Jeonghee Han, Eun-joo Hong, Hyun Kyung Kim, and Eun-Jin Kim are the co-authors of some of the published research papers included in this book. Marie Seong-hak, your outstanding ideas in reviewing this book helped it catapult to the level of the high quality of your own book. Seun and Esther Hong, your talent in English proofreading is beyond any professional services. Thank you for your delightful insight and magical touch in editing this book with all your might. Joel, Christopher, and Seoyeon, thank you for your continued support. Lastly, I must thank my most faithful writing companions, wine and Julie. Wine may be the biggest contributor to my illness, and without it this book just might have been completed a few years earlier. My dog Julie, who came to me after my illness as sort of a reward for surviving cancer, has kept me company and entertained me during countless hours of writing and proofreading at my new research lab, a tree house-like-space where you can

Acknowledgments

always expect the company of good wine and an animal friend. This space is aptly named WA-GAE-YEON, WA for wa-in (wine in Korean), GAE for gae (dog in Korean), YEON for yeon-gu-shil (lab in Korean). I guarantee it is the best hide out to drink and write with my dog and you are invited anytime for beer or wine, or both. I thank all of you, and countless others, from the bottom of my heart. I am just grateful to complete this long term project which took unexpected 7 years, during which I even doubted that I would finish.

xi

Preface

The invaluableness of art as a media for expressing oneself is well conveyed Art therapy and in the widely quoted words of Georgia O’Keefe (1887-1986), “I found that computer science I could say things with colors and shapes that I couldn’t any other way Things I had no words for” (Tripp, 2016). As an engineer and an art therapy researcher, I believe that an interdisciplinary approach incorporating computer science with art therapy using art as a media can significantly expand the potential and value of art therapy. Lord Kelvin (1824-1907) pointedly stated: “When you can measure what you are speaking about, and express it in numbers, you know something about it; but when you cannot express it in numbers, your knowledge is of a meager and unsatisfactory kind; it may be the beginning of knowledge, but you have scarcely, in your thoughts, advanced to the stage of science” (Thomson, 1889). Art therapy, accompanied by scientific thought, can achieve its greatest effect in its purpose of improving and enhancing the physical, mental, and emotional well-being of individuals of all ages. The purpose of this book is to explain, in a reader-friendly format, the ways in which computer technologies can help art therapists improve the quality of their practice and advance the theory of art therapy. Computer science research, especially in the field of artificial intelligence, has developed methods for computers even to think, learn, and improve by themselves. Of course, art therapists already use computer technologies for various purposes, such as word processing, business-related packages, computer graphics, digital storage, retrieval of client artwork, e-mail, website, etc. An interdisciplinary approach to art therapy can critically benefit from artificial intelligence, however, the purpose of which is to make a computer that can imitate intellectual actions of human beings. Knowledge in art therapy is largely empirical, heuristic, and subjective, Ill-structured based on an art therapist’s individual professional expertise and experience, paradigm which are not always amenable to organized algorithms. Thus, the field of xiii

xiv

Preface

art therapy can be classified as a so-called “ill-structured paradigm,” which is referred to in the field of artificial intelligence as a vague and unclear problem domain in which there are few means for finding a solution and in certain cases the solutions found are inevitably contradictory. This nature of art therapy makes decision-making in art evaluation and art interpretation complex and difficult for art therapists. In this sense, art therapy is analogous to such areas as economic demand forecasting, weather forecasting, diagnosis of diseases, or judicial sentencing. The use of artificial intelligence in these areas is already in progress and has yielded significant results. In the search for a solution to the problem dealing with the subjective, ambiguous, inconsistent, and sometimes contradictory nature of the huge volume of knowledge involved, art therapists should be made aware of new computer technologies applicable to the improvement of practice and theory in their field. In this era of rapidly expanding computer technologies, discourses of art therapists with computer technologies could be made livelier and more productive, creating new perspectives on which to build knowledge. Interdisciplinary This book is concerned with the interdisciplinary studies applying computer study technologies to the theory and practice of art therapy. The contents consist of the author’s sixteen papers published, twelve patents in Korea, Japan, and the U.S.A., and other relevant materials, all organized in a logical sequence. This book is intended for art therapy courses at upper undergraduate and graduate levels. No prior computer knowledge is assumed. The difficult concept of computer science is explained in a simple and concrete way with illustrations, sample drawings, and case studies. This book explains statistical methods, various functions of a computer, technologies in digital image processing, computer algorithms, methodologies in expert systems, and the Bayesian network. All these elements can be used to improve the practice and theory in the evaluation of art and the interpretation of art. The readers do not need to worry about unfamiliar terms such as digital image, algorithm, expert system, and Bayesian network which appear here. Neither should they be concerned about pixel, cluster, edge, blurring, convex hull, regression, etc., the terms which appear later in the book. These terms will be explained with illustrations and drawings for easy understanding. Art therapists are often asked by parents to diagnose their child’s psychological state by evaluating and interpreting a drawing created by the child. Identifying a person’s state of mind by evaluating only what they have drawn is just a small part of art therapy. Unfortunately, this small part is all what this book is about. Interpretation of drawings on drawer’s psychological state based on the results of elements evaluation, however, is more than just a small part of art therapy and remains one of the most important and fundamental tasks for art therapists to perform. Performing that task no longer needs to be done manually by the therapists themselves because, as this book argues, computerized systems can perform the steps of evaluation

Preface

xv

and interpretation. The author strongly believes that the proposed computer systems will contribute to the advancement of art therapy. As the title, Computational Art Therapy (CAT), indicates, this book is a Organization of the study of art therapy using computer technologies. Part One, Art Evaluation, book is concerned with the computerized art evaluation of elements of drawings, and Part Two, Art Interpretation, is concerned with the computerized art interpretation of the drawer’s psychological state. Before the two parts, in the Prologue, definitions, the need for, potential contributions, and the pros and cons of CAT are introduced. The Prologue introduces common material integral to two parts, including the Computer Color-Related Elements Art Therapy Evaluation System (C_CREATES) which is a computerized system for the evaluation of elements of a drawing. Each chapter begins with an “abstract,” “summary,” “keynote,” or “key point(s)” section or a combination of any two of these which is appropriate for the chapter. When readers come across unfamiliar materials or contents, they don’t need to fear. Rather, the reader is encouraged to imagine freely. Whether the imagination is correct or not will become clear as the reader progresses through the chapters and to the CAT. The first two chapters in Part One develop methods for the evaluation of 12 basic elements in the C_CREATES. These elements are basic in the sense that the evaluation results are perfectly accurate, thus their validation is not necessary. The next 5 chapters develop methods for the evaluation of 7 elements, which are applied ones in the sense that the evaluations are based on the basic elements. The evaluation results are consistent but can be somewhat variable even if the computerized methods are applied, and thus the reliability of human raters and the validity of the system need to be proved. The last chapter applies the various evaluation methods to an art therapy tool and demonstrates how the evaluation results of Part One can be used for the interpretation of Part Two. The first four chapters in Part Two deal with the expert systems for the interpretation of a drawing. The chapters cover, respectively, the concept, the reasoning process, and two applications to art therapy tools. The following three chapters develop systems for, respectively, the decision on whether a drawing is made by a patient or a non-patient, the estimation of the level of a psychological state, and the comparison of the efficiencies of various tools in estimating the level of the psychological state. The last chapter proposes the probabilistic interpretation instead of traditional deterministic one. See the prologue for a more detailed flow among the chapters. The epilogue discusses the contributions of the CAT and suggests its potential for further developments. Although this book is self-contained, CD or companion website, http://www.ccthomas.com/cat is available to provide readers with software packages of the C_CREATES with program files, original drawings, samples, etc., which art therapy practitioners can

xvi

Preface

easily use via their PC to access technologies to help them obtain a more reliable analysis. Practitioners can thus save time and effort in their practice, easily enhancing the quality of their practice. The manual of the software package is given in the Appendix. Author’s hope The author hopes this book will not only promote the use of various art therapy tools but also provide a foundation for new methodologies through which art therapy researchers can develop their own methodologies to improve the practice and theory of art therapy. The author wishes that interdisciplinary studies incorporating art therapy, psychology, psychiatry, art, computer science, applied statistics, etc. will be performed under the umbrella of the CAT. Finally, I admit, as an engineer, that my interest in the CAT lies mainly on the technical side. I am, after all, a novice in art therapy. I acknowledge there remains a significant amount of further study. It is my hope that any oversights and errors found in this work will serve as a stepping stone for the betterment of future CAT. Seong-in Kim

Contents

About the Author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

vii ix xiii

Prologue Introduction to the Computational Art Therapy (CAT) Keynote

P.1 The need of computer technologies in the field of art therapy P.1.1 Present status of computer technologies in art therapy . . . . P.1.2 Problems and difficulties in art therapy . . . . . . . . . . . . . . . . . . P.1.3 Computer technologies as a solution . . . . . . . . . . . . . . . . . . . . P.1.4 The definition of Computational Art Therapy (CAT) . . . . . . . P.2 Computer technologies relevant to art therapy P.2.1 Built-in functions of a computer . . . . . . . . . . . . . . . . . . . . . . . . P.2.2 Techniques of digital image processing . . . . . . . . . . . . . . . . . . P.2.3 Computer algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . P.2.4 Expert system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . P.2.5 Statistical methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . P.2.6 Bayesian network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . P.3 Taxonomy of color-related elements in the C_CREATES for art evaluation P.3.1 Classification of evaluation elements in art therapy tools . . P.3.2 Reliability of art evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . P.4 The computer systems for art interpretation P.4.1 Traditional methods and computer systems for art interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . P.4.2 Validity of art interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . P.5 Organization of the book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . P.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii

1 2 3 4 4 5 6 6 7 8 10 13 15 17 18 22

xviii Contents

PART ONE

ART EVALUATION Chapter 1 Evaluation of Basic Elements in the Computerized Color-Related Elements Art Therapy Evaluation System (C_CREATES) (I) Abstract and Summary

1.1 Built-in functions of a computer 1.1.1 Color in art therapy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.2 Color recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.3 Color classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.4 Edge detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Techniques in digital image processing 1.2.1 Noise removing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.2 Blurring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.3 Clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.4 Edge detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Evaluation of basic elements in the C_CREATES 1.3.1 Procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.2 Verifications of the C_CREATES . . . . . . . . . . . . . . . . . . . . . . 1.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

29 30 32 34 34 34 36 36 38 40 46

Chapter 2 Evaluation of Basic Elements in the C_CREATES (II) Summary

2.1 Elements related with color definitions, space colored, and .................................... pattern coloring 2.2 Primary / secondary, warm / cool, and complementary colors 2.2.1 Elements related with color definitions . . . . . . . . . . . . . . . . . 2.2.2 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Number of colored grids and area of colored convex hull 2.3.1 Elements related with space colored . . . . . . . . . . . . . . . . . . . 2.3.2 Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Completeness and accuracy 2.4.1 Elements related with pattern coloring . . . . . . . . . . . . . . . . . 2.4.2 Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

48 50 51 52 53 55 56 58

Chapter 3 A Computer System for Rating Variety of Colors Summary

3.1 Importance of color-related elements . . . . . . . . . . . . . . . . . . . . . 3.2 Method 3.2.1 Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

59 61

Contents

3.2.2 Sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Results 3.3.1 Inter-rater reliability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.2 System validity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

61 64 66 69

Chapter 4 Judgment of Main Color Using Computer Algorithm Keynote

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Procedure of main color judgment 4.2.1 Case examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.2 Computer algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 System verification 4.3.1 Inter-rater reliability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.2 System validity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

70 73 73 76 79 80

Chapter 5 Determination of Placement Using Techniques of Digital Image Processing Key points

5.1 The element of placement in art therapy tools . . . . . . . . . . . . . . 5.2 Methods for edge detection and definition of placement category 5.2.1 Methods for edge detection . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.2 Definitions of placement categories . . . . . . . . . . . . . . . . . . . 5.3 Determination of placement category 5.3.1 Information on placement category . . . . . . . . . . . . . . . . . . . 5.3.2 Information on other elements of drawings . . . . . . . . . . . . . 5.4 System verification 5.4.1 Sample examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.2 Inter-rater reliability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.3 System validity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.4 Other useful information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

81 83 85 86 87 87 91 91 91 92

Chapter 6 Grading and Ranking Prominence of Color and Details of Drawing Using Regression Models Key point

6.1 Regression analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Method and samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 Evaluations by human raters and their inter-rater reliabilities

93 95

xix

xx Contents 6.3.1 Grade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.2 Rank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4 Evaluation by regression models 6.4.1 Grade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4.2 Rank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5 System validities 6.5.1 Grade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5.2 Rank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.6 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

95 96 100 102 104 105 106

Chapter 7 Evaluation of Space Usage in a Drawing and Degree of Concentration in a Pattern Coloring Key points

7.1 Importance of space usage and degree of concentration . . . . 7.2 Regression models for the evaluation of space usage in grade and rank 7.2.1 Possible independent variables . . . . . . . . . . . . . . . . . . . . . . 7.2.2 Inter-rater reliabilities in the evaluation of dependent variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.3 Regression models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.4 System validity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3 Regression model for the evaluation of concentration in rank 7.3.1 Sample pattern colorings . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.2 Inter-rater reliability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.3 A regression model and its validity . . . . . . . . . . . . . . . . . . . 7.4 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

108 111 114 114 116 117 119 119 121

Chapter 8 A Bridge from Part One to Part Two: Computerization of Art Evaluation and Its Application to Art Interpretation Abstract

8.1 An approach to developing a computerized evaluation system and its connection to art interpretation . . . . . . . . . . . 8.2 Computerization of the Face Stimulus Assessment (FSA) 8.2.1 The FSA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2.2 Algorithms and criteria for each element . . . . . . . . . . . . . . 8.2.3 Elements in the Computerized Face Stimulus Assessment (c_FSA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2.4 Reliability and validity of the c_FSA . . . . . . . . . . . . . . . . . . 8.3 Application of the evaluation results in Part One to the interpretation in Part Two 8.3.1 Relationships between the space usage in the PPAT and severity, and degree of dementia . . . . . . . . . . . . . . . . . . . . .

123 124 125 126 129

130

Contents

8.3.2 Relationships of the five elements in the c_FSA . . . . . . . . . 8.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

132 133

PART TWO

ART INTERPRETATION Chapter 9 An Expert System Approach to Art Interpretation Abstract

9.1 Various factors considered in art interpretation . . . . . . . . . . 9.2 An expert system for art interpretation 9.2.1 System facilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2.2 Knowledge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2.3 Reasoning process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2.4 Advantages of the system . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2.5 System features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.3 Case study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

137 139 139 142 146 147 149 150

Chapter 10 Reasoning Process of an Expert System for Art Therapy Abstract

10.1 Modeling human decision process . . . . . . . . . . . . . . . . . . . . . . 10.2 Process of diagnosis consisting of nine sub-processes 10.2.1 Requirements of art interpretation process in expert system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.2.2 Model of reasoning process . . . . . . . . . . . . . . . . . . . . . . . 10.3 Reliability, consistency, and learning abilities . . . . . . . . . . . 10.4 Knowledge base for each stage . . . . . . . . . . . . . . . . . . . . . . . . . 10.5 Case study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

152 154 155 158 161 163 167

Chapter 11 An Expert System for Interpreting the Structured Mandala Coloring (SMC) Drawings Abstract and summary

11.1 The Structured Mandala Coloring (SMC) as a subject of expert system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Knowledge base 11.2.1 Knowledge expression . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2.2 Structure of knowledge base . . . . . . . . . . . . . . . . . . . . . . . 11.3 An expert system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.4 Case study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

168 171 172 176 178

xxi

xxii Contents 11.5 Discussion

..........................................

181

Chapter 12 Computerized Kinetic Family Drawing Using Patterns (p_KFD) Summary

12.1 The Kinetic Family Drawing (KFD) as a subject of computerizing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.2 Questionnaires with fact base . . . . . . . . . . . . . . . . . . . . . . . . . . 12.3 Composition and coloring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.4 Evaluation of elements and detection of changes 12.4.1 Evaluation of elements in the KFD . . . . . . . . . . . . . . . . . 12.4.2 Detection of changes in two KFD drawings . . . . . . . . . . 12.5 Interpretation with knowledge base . . . . . . . . . . . . . . . . . . . . 12.6 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

184 187 188 191 193 197 201

Chapter 13 Computerized Structured Mandala Coloring (c_SMC) for Differentiation and Identification of Psychological States Using Statistical Methods Key Points

13.1 The Structured Mandala Coloring (SMC) as a subject of computerization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.2 Methods 13.2.1 Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.2.2 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.2.3 Data analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.2.4 System validity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.3 Results 13.3.1 Differentiation of groups . . . . . . . . . . . . . . . . . . . . . . . . . . 13.3.2 Identification of a group . . . . . . . . . . . . . . . . . . . . . . . . . . 13.3.3 System validity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.4 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

204 207 207 208 209 209 210 212 214

Chapter 14 Statistical Models for Estimating Level of Psychological Disorder Abstract

14.1 Regression model to estimate degree of dementia using structured mandala . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14.2 Methodology 14.2.1 Application to dementia . . . . . . . . . . . . . . . . . . . . . . . . . . 14.2.2 Model I: Estimation of levels of dementia . . . . . . . . . . . 14.2.3 Model II: Probability of severe dementia . . . . . . . . . . . . 14.3 Results and system validity 14.3.1 Selection of independent variables and their effects in

216 219 220 220

Contents

Model I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14.3.2 Selection of independent variables and their effects in Model II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14.4 Case studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14.5 Discussion, conclusion, and further study . . . . . . . . . . . . . . .

221 225 227 228

Chapter 15 A Statistical Approach to Comparing the Effectiveness of Several Art Therapy Tools in Estimating Level of a Psychological State Abstract

15.1 A generalized approach to compare effectiveness of several art therapy tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15.2 Approach: Regression model . . . . . . . . . . . . . . . . . . . . . . . . . . 15.3 Case study 15.3.1 Subjects of psychological disorder and art therapy tools . 15.3.2 Independent variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15.4 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

232 234 235 237 238 243

Chapter 16 Probabilistic Art Interpretation Using Bayesian Network Abstract and keynote

16.1 Probabilistic interpretation vs. deterministic interpretation . 16.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16.3 A Bayesian network-based art interpretation . . . . . . . . . . . . 16.4 System verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16.5 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

245 247 250 252 253

Epilogue Searching for the Advancement of Art Therapy

..............

255

Appendix: Companion S/W . . . . . . . . . . . . . . . . . . . . . . . . . . Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Copyright Permissions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

259 270 272 275 289

xxiii

Introduction to the Computational Art Therapy (CAT)

Prologue Keynote In this section, we introduce the main theme of the book. We define the Computational Art Therapy (CAT) as an art therapy that actively uses various computer technologies including quantitative statistical methods. The CAT is proposed as an approach to solve the “Ill-Structured Paradigm (ISP)” problems in the field of art therapy. The art evaluation of Part One and the art interpretation of Part Two in this book demonstrate the difference in methodology between the CAT and conventional art therapy, by focusing on the computerized evaluation of elements of drawings using the Computer Color-Related Elements Art Therapy Evaluation System (C_CREATES) (Kim, Bae, & Lee, 2007; Kim, 2010) and the computerized interpretation of a drawing regarding the psychological state of the creator of the drawing. We describe the fundamental computer technologies and methods commonly used in the following 16 chapters of this book. In addition, we summarize the flow and the organization of the chapters. The primary goal of the CAT is to provide objective and quantitative evaluation of elements and systematic interpretation of art, computerize art therapy tools, develop new methodologies for the improvement of practice and theory, and thus establish art therapy as a science.

P.1 The need of computer technologies in the field of art therapy P.1.1 Present status of computer technologies in art therapy Art therapists are proud of using art in their profession. The invaluableness of art as a medium for expressing oneself is well conveyed in the famous quote of Georgia O’Keefe (Tripp, 2016), “I found that I could say things with colors and shapes that I couldn’t any other way - Things I had no words for.” However, people and art are so complex (Cohen & Mills, 1994) that scientific analysis of such subjects is seemingly impossible. This book aims 1

2 Prologue Introduction to the Computational Art Therapy (CAT) to help change this conventional belief. In relation to art evaluation and art interpretation, this Prologue delineates the strength and weakness as well as the potential of using computer technology in art therapy. Since the advent of mankind’s first computer in 1946, computer technology has made remarkable progress and greatly influenced human society. There are numerous fields which employ computer technology, such as the Computer Aided Design (CAD), Computer Aided Manufacturing (CAM), and Computer Aided Education (CAE). Various fields including manufacturing, design, education, and law have progressed remarkably by adopting rapidly changing computer technologies. In medicine, an expert system of artificial intelligence, MYCIN, developed in 1972 to diagnose epidemic diseases and prescribe anti-biotics, has had a tremendous effect (Shortliffe, 1972). IBM’s Watson, a super computer powered with artificial intelligence, is changing health care, from diagnosing disease to treating it (Smith, 2015) - correctly diagnosing a patient within minutes, something doctors failed to do after months (Billington, 2016). P.1.2 Problems and difficulties in art therapy Compared with other fields, art therapy has been relatively slow in adopting computer technology. Art therapists of course have been using PCs and e-mails every day, and have tried remote sessions on line to improve the quality of art therapy practices (Peterson, Stovall, Elkins, & Parker-Bell, 2005; McNiff, 1999; Parker-Bell, 1999; Malchiodi, 2000; Betts, 2006; Hartwich & Brandecker, 1997; Orr, 2006a). Thong (2007) combined traditional art therapy tools such as drawings, pictures, and collages with computer software of Photoshop (Adobe Creative Team, 2004) and Flying Colors (de Jong, 1993). Im, Oh, Lee, Chang, and Park (2010) developed a computer module that provides a familiar drawing board so that the client can easily sketch and draw online and the art therapist can review the whole sketching process in a time-efficient manner. On the other hand, Belkofer (2011) discussed the ethics in the context of using participatory social media such as Facebook, YouTube, Twitter, Skype, Blogging, and online video games, as well as e-mail for art therapy sessions. However, most of art therapy research so far has been confined to using the computer as a tool for art making (McLeod, 1999), and as a medium for distance art therapy of remote sessions with clients (Malchiodi, 1999). Several studies have gone beyond the above practice, using a computer system for evaluating drawings (Kim, Bae, & Lee, 2007; Kim, 2008a; Kim, Kang, & Kim, 2008; Kim, Han, & Oh, 2012; Kim, 2010; Mattson, 2009, 2011, 2012a, 2012b) and interpreting drawings (Kim, Ryu, Hwang, & Kim, 2006; Kim, Kim, Lee, Lee, & Yoo, 2006; Kim, Kim, & Kim, 2008; Kim, Han, Kim, & Oh, 2011). This line of research is still scarce, however. One of the reasons for this limited state of research is the lack of art therapists’ knowledge about state-of-the-art computer technology. Some practicing art

Prologue Introduction to the Computational Art Therapy (CAT)

therapists were initially skeptical about the effectiveness of technological tools (Fryrear & Corbit, 1992; Hartwich & Brandecker, 1997), just like in other fields that depend on human expertise and skill. Asawa (2009) reported that most art therapists’ responses to technology were anxiety and fear. It should be stressed here that the main purpose of using computers is not to replace human experts but rather to aid them by providing relevant information, reducing the time and effort in their work, and providing convenience (Kim, Kang, & Kim, 2008). It is now generally recognized in the art therapy field that computer technology should be actively understood and leveraged. We note this trend from a number of developments: In 1999, the American Art Therapy Association (AATA) published a special journal issue on “Digital Art Therapy.” In 2001, the title of the Psychotherapy Networker journal was “Our Technology Ourselves: How Digital Revolution is Changing Psychotherapy.” In 2007, Kapitan (2007) emphasized the use of computers for managing and practicing art therapy, and stressed the importance of multi-disciplinary research for new media. In 2008, the AATA conference held in Cleveland, US, featured the theme, “Art Therapy on the Cutting Edge: Invention and Innovation.” In 2009, the AATA published a special journal issue on “Art Therapy’s Response to Techno-Digital Culture.” Another reason for the slow adoption of computer technology in the field is that knowledge in art therapy is mostly empirical, heuristic, subjective, inconsistent, and even contradictory, relying on the art therapist’s professional expertise and experience. The field of art therapy could be classified as a so-called “Ill-Structured Paradigm (ISP),” the terminology often used in expert systems, which refers to a vague and unclear problem domain in which well-defined algorithms or an objective means for finding a solution hardly exist. In certain cases the solutions found are inevitably contradictory (Giarratano & Riley, 2005). The nature of art therapy itself makes the decision-making complex and difficult for art therapists. In this sense, art therapy is analogous to areas such as economic demand forecasting (Chang, Wang, & Liu, 2007), weather forecasting, diagnosis of diseases (Shortliffe, 1972), or judicial sentencing in criminal cases (Kim, Kim, Lee, Kim, & Baik, 1992). Gussak and Nyce (1999) pointed out that art therapy is eclectic in theory and practice and not reducible to a single set of algorithms. P.1.3 Computer technologies as a solution Solving problems with computers require scientific methods based on quantification. Since it is seemingly difficult, if not impossible, to quantify the process of art therapy, the field has not paid enough attention to using computers to its fullest extent. It has also been widely believed in the field that the ability to reason qualitatively, which is essential in art therapy, cannot be specified as a computer program. However, art therapists should now break out of this conservative view. Lord Kelvin, early on in history, emphasized

3

4 Prologue Introduction to the Computational Art Therapy (CAT) the importance of quantification, stating that “measuring whatever it is and expressing it in numbers may be the beginning of knowledge advanced to the stage of science (Thomson, 1889).” Many art therapists still claim that their evaluation of elements of drawings and interpretation regarding drawers’ psychological state are complex and not amenable for quantification. However, computer algorithms such as color analysis and edge extraction can provide basic building blocks for objectification and quantification of art evaluation. Furthermore, recent progress in artificial intelligence research allows computers to solve problems where objectification and quantification are difficult or even impossible. Artificial intelligence enables computers to reason with expert knowledge like humans. In the past, we used computers only for problems that were obviously quantifiable so as to write a sequence of instructions (i.e. an algorithm) to solve them. However, we can now make computers use human expert’s knowledge and experience to solve problems with uncertain or imperfect specifications. Kim (2008b) claimed that art therapists should embrace the techniques in artificial intelligence to assist art evaluation and art interpretation. Art therapy, supported by state-of-theart computer technology, can make significant contribution to improving the physical, mental, and emotional well-being of individuals of all ages, which is ultimate goal of art therapy. P.1.4 The definition of Computational Art Therapy (CAT) In this book, we define the Computational Art Therapy (CAT) as an art therapy that actively utilizes various computer technologies including quantitative statistical methods. This is different from the limited use of computers as a tool for generating drawings or communication media for remote sessions. Furthermore, it applies computer technology such as artificial intelligence to evaluate and interpret drawings.

P.2 Computer technologies relevant to art therapy P.2.1 Built-in functions of a computer First generation computers were able to process 102 instructions per second. Current PCs can process 1012 instructions per second. The ability to process a huge number of instructions per second makes software faster, saving art therapists’ time and effort in many aspects. The first generation computers were able to store several thousand characters, compared with more than 1012 characters nowadays. The ability to store a huge amount of information allows art therapists not only to store the drawings from clients as well as video recording of the art drawing process and evaluation results, but also to retrieve them any time. Today, the size of the population with access to the internet is estimated to be bigger than one billion. The progress in

Prologue Introduction to the Computational Art Therapy (CAT)

5

Figure P.1 Color recognition and edge detection by a computer after blurring and clustering.

communication technology facilitates opinion and information sharing with clients as well as remote drawing of art and its evaluation and interpretation, bringing remote art therapy one step closer to realization. Computers represent a drawing by splitting it into pixels, e.g. using 1,280 pixels horizontally and 960 pixels vertically, a total of 1,228,800 pixels. The color of a pixel can be any of 256 x 256 x 256 = 16,777,216 colors. There are a number of software tools for drawing on the display, e.g. Corel Painter, Tux Paint, InkScape, and CADian to name only a few. Photoshop (Adobe Creative Team, 2004) is one of the most well-known software tools for drawing. Such advances in computer graphics provide a novel way to generate drawings and a finer scale of information in drawings, computerize art evaluation and interpretation, facilitate new art therapy assessments, and improve the reliability and validity of art therapy tools. P.2.2 Techniques of digital image processing Digital image processing in pattern recognition is a subfield of cognitive science, which is about recognizing characters, speech, or shapes using computational machines. Blurring and clustering which are techniques of digital image processing can be used to evaluate the colored areas of each color and the length of edges of a drawing. We note that the elements in art therapy assessments are mentioned in italics throughout the book. Figure

6 Prologue Introduction to the Computational Art Therapy (CAT)

Figure P.2 Computer algorithm for judgment of main color.

P.1 shows how the computer recognizes colors and extracts edges between colors through blurring and clustering. Detailed explanation can be found in Chapter 1. P.2.3 Computer algorithm A computer algorithm refers to a sequence of instructions or computational methods to solve a given problem. Figure P.2 shows the flow of a computer algorithm for removing background color to determine a main color of a drawing. Detailed explanation can be found in Chapter 4. P.2.4 Expert system Artificial intelligence is a field about making machines such as computers or robots, able to complete complex tasks such as adapting to new situations, learning from past experiences, and recognizing things that are unique to human intelligence. Expert systems is one of the most successful subfields of artificial intelligence, where the focus is on capturing human expert knowledge for making computer reason and make decisions. Some expert systems are capable of learning as well, automatically accumulating

Prologue Introduction to the Computational Art Therapy (CAT)

7

Figure P.3 The common architecture of expert system.

and improving knowledge. Examples include the expert systems for deciding sentence in law (Kim et al., 1992) and medical diagnosis (de Clercq, Hasman, Blom, & Korsten, 2001). In expert systems, the expert knowledge is stored into the knowledge base, which is continuously accumulated and updated. For example, it can store, accumulate, and update newly discovered knowledge on the relationship between the psychological state and the elements of a drawing in art therapy assessments such as the Diagnostic Drawing Series (DDS) (Cohen, 1986/1994) or the Formal Elements Art Therapy Scale (FEATS) (Gantt & Tabone, 2003) rating system of the Person Picking an Apple from a Tree (PPAT) (Gantt, 1990). The MATLab (MATrix Laboratory) (MathWorks, 2016) can be used for the development of expert systems. Figure P.3 shows the architecture of the expert system. Detailed explanation can be found in Chapter 9. P.2.5 Statistical methods Various statistical methods are applied to art evaluation and art interpretation. The analysis of variance is used to examine the relationship between an element and an interpretation. The Pearson Correlation Coefficient

8 Prologue Introduction to the Computational Art Therapy (CAT)

Figure P.4 Illustration of regression.

(PCC), Rp, measures the relationship. Cohen’s kappa value, κC, Fleiss’ kappa value, κF, or the Quadratic Weighted Kappa (QWK) value, κ2, measures the inter-rater reliability or the consistency of human raters’ evaluations of element in grades, and the Rank (Spearman) Correlation Coefficient (RCC), Rs, measures evaluations in ranks. Regression analysis is applied to develop systems for evaluation of an element from other elements, and systems for predicting a psychological state from the evaluation results of drawings. The Coefficient Of Determination (COD), R2, is a measure for the appropriateness of regression model. Figure P.4 shows two cases where the element of prominence of color is estimated from either only the element of number of used colors or both the elements of number of used colors and area colored in number of pixels. Detailed explanation can be found in Chapter 6. Any textbooks on basic statistics such as Probability and statistics for engineers and scientists (Walpole & Myers, 2006) can be recommended for understanding the concept of these methods. Also any statistical software packages such as Statistical Package for the Social Science (SPSS) (IBM Corporation, 2015) or Statistical Analysis System (SAS) (SAS Institute, 2014) can be used to replace time-consuming calculations. P.2.6 Bayesian network The Bayesian network is a technique for artificial intelligence. It graphically represents the probabilistic cause-and-effect relationship using a set of random variables as nodes and their conditional probabilities as edges in a directed acyclic graph (Charniak, 1991). The Bayesian network allows us to encode knowledge that is not deterministic. For example, the traditional deterministic interpretation of art is that “children suffering from Attention Deficit and Hyperactivity Disorder (ADHD) tend to draw a full picture using more than 90% of the paper.” Using the Bayesian network, we can make a more quantitative expression such as “children with ADHD draw a full

Prologue Introduction to the Computational Art Therapy (CAT)

9

Figure P.5 Diagram of a Bayesian network.

picture with a probability of 0.75, whereas normal children draw such a picture with a probability of 0.1.” Figure P.5 shows a simple example of the Bayesian network (Ghil & Kim, 2010). Detailed explanation can be found in Chapter 16. In summary, we can make the computer perform art evaluation by combining the built-in functions of a computer, techniques of digital image processing, computer algorithms, statistical methods, and expert systems. Here, art evaluation refers to analyzing, measuring, or rating elements of drawings. Furthermore, we can computerize existing art therapy assessments (tools) of evaluation or develop new elements. Again, the statistical methods, expert systems, and Bayesian networks can be used to interpret drawings. Here, art interpretation refers to inferring the information on the drawer’s psychological states from the formal and generic elements of his or her drawing, as in the DDS and the PPAT. Through active application of the various computer technologies, we can make the art therapist’s knowledge more structured and systematic, which has tended to remain diversified and subjective, and sometimes even inconsistent and self-contradictory.

10 Prologue Introduction to the Computational Art Therapy (CAT)

P.3 Taxonomy of color-related elements in the C_CREATES for art evaluation This section introduces some representative art therapy assessments such as the DDS, the PPAT, and several computer systems including the Computer Color-Related Elements Art Therapy Evaluation System (C_CREATES) (Kim, Bae, & Lee, 2007; Kim, 2010), classifies their evaluation elements, and examines their inter-rater reliabilities. P.3.1 Classification of evaluation elements in art therapy tools The House-Tree-Person (HTP) (Buck, 1949), the Draw A Person (DAP) (Goodenough, 1926), and the Kinetic Family Drawings (KFD) (Burns & Kaufman, 1972) are known as psychological projective tests developed in the field of psychology. Creative art therapists have developed their own tools for use in art therapy practice, known as “art therapy assessments,” such as the DDS, the PPAT, the Silver assessments (Silver, 2002), the Levick assessment (Levick, 2001), the Face Stimulus Assessments (FSA) (Betts, 2003), and the Bird’s Nest Drawing (Kaiser & Deaver, 2009). The DDS is designed to gather clinical information about a client in a single session (Brooke, 2004). It evaluates the following 23 elements: (1) color types. (2) blending. (3) idiosyncratic color. (4) line / shape. (5) integration. (6) abstraction. (7) representation. (8) image. (9) enclosure. (10) groundline. (11) people. (12) animals. (13) inanimate objects. (14) abstract symbols. (15) word inclusion. (16) landscape. (17) line quality / pressure. (18) line length. (19) movement. (20) space usage. (21) tree. (22) tilt. (23) unusual placement. Element of drawings will be mentioned throughout the book in italics. The PPAT intends to provide a method for understanding and examining the non-symbolic aspects of art and to demonstrate how structural characteristics of a drawing furnishes information on a drawer’s clinical state and his or her psychiatric diagnosis (Brooke, 2004). The FEATS, its rating system, has the following 14 elements: (1) prominence of color. (2) color fit. (3) implied energy. (4) space. (5) integration. (6) logic. (7) realism. (8) problem-solving. (9) developmental level. (10) details of objects & environment. (11) line quality. (12) person. (13) rotation. (14) perseveration. The Descriptive Assessment of Psychiatric Artwork (DAPA) developed by Hacking (1999) includes the following 5 elements: (1) number of used colors. (2) list of colors. (3) space usage. (4) emotional tone. (5) line. As additional elements, the first picture of FSA considers the following 5 elements: (1) accuracy of coloring. (2) color fit. (3) form recognition. (4) precision of description. (5) space usage. The elements listed above concern global and formal attributes rather than the contents of a drawing. The reason is that if there are differences between two or more groups, we contend that these differences will result

Prologue Introduction to the Computational Art Therapy (CAT)

11

from the way how the group members draw rather than what they draw. Given that the possible subject matter for art is vast, one would have considerable difficulty in testing for measurable differences in content between groups (Gantt & Tabone, 2003). The DDS rating system is comprised of 23 elements, whereas the DAPA consists of 5 elements, the FEATS 14 elements, and the FSA 5 elements. Among them, the elements with different names actually have the same or a very similar meaning. Noting that many elements of evaluation is concerned directly or indirectly with the color in a drawing and also that evaluation of some elements can be computerized, Kim, Bae, and Lee (2007) and Kim (2010) developed the C_CREATES for art evaluation. The C_CREATES also proposes new elements that are color-related and able to be computerized. All of its 19 elements are evaluated automatically and quantitatively by computer technologies. We note that “color-related elements” are not limited to elements directly related to colors such as number of colors, but include elements indirectly related to colors - elements which are formed during the process of color analysis such as length of edges. We also note that the C_CREATES considers only colored drawings, not drawings with only lines. A computer system for the latter needs a separate study. The evaluation of these elements in the C_CREATES is the topic of Part One in this book. These elements are classified into two categories depending on the presence or absence of absolute objectivity. The first category includes elements which can be evaluated so objectively that the results of evaluation are not disputed. We call them basic elements. The second category includes elements whose evaluation results are consistent but can be somewhat subjective depending on the way evaluation methods are constructed. We call them applied elements. These elements can be evaluated based on the evaluation results of basic elements. Regression analysis, computer algorithm, and expert system can be applied. Currently the system incorporates 12 basic elements and 7 applied elements. Note this new taxonomy of color-related elements has modified from ones previously presented in Kim, Bae, and Lee (2007) and Kim (2010). This category is analyzed by the built-in functions of a computer, techniques Basic elements of digital image processing, computer algorithms, and expert systems such as color recognition and edge detection. It includes (1) number of used colors, (2) list of colors, (3) number of clusters, (4) length of the edges, (5) area of each color, (6) number of colored grids, (7) area of colored convex hull, (8) completeness, (9) accuracy, (10) primary / secondary colors, (11) warm / cool colors, and (12) complementary colors. A cluster refers to a contiguous area colored with the same color, not separated by different colors. The edge consists of pixels of which the color is different from that of its neighboring pixels. The convex hull refers to the area which includes all pixels on the line between its two points. Figure P.6

12 Prologue Introduction to the Computational Art Therapy (CAT)

Figure P.6 Number of colored grids and area of colored convex hull.

shows the computer technology and its mechanism to evaluate elements (6) and (7). We note that elements (3), (4), (5), (7), (10), (11), and (12) are evaluated in number of pixels and in percentage of whole area colored or total length of edges. See Chapter 7 for details. Chapter 1 deals with the first 5 elements, and Chapter 2 the remaining 7 elements. Applied elements This category can be determined by applying statistical models to the basic elements or identified by constructing an appropriate knowledge base, based on the basic elements. It includes (13) variety of colors, (14) main color, (15) placement, (16) prominence of color, (17) details, (18) space usage, and (19) degree of concentration. For example, variety of colors is ranked higher when there are more colors, and, when two drawings are found to have the same number of used colors, the one with the longer length of edges is ranked higher. Figure P.2 shows a computer algorithm regarding how main / subsidiary colors is identified. Prominence of color is, for example, statistically rated by the following regression function: Prominence of color (grade) = 1.201 + 0.247 x Number of colors + 0.002 x Number of clusters + 0.376 x 10-6 x Area of colored convex hull. See Chapter 6 for details. Chapter 3 deals with variety of colors, Chapter 4 main color, Chapter 5 placement, Chapter 6 prominence of color and details, and Chapter 7 space usage and degree of concentration. The total number of elements considered in traditional art therapy assessments, the DDS, the PPAT, the DAPA, and the FSA is 47. Some of them, such as integration, line quality, and space usage, are listed in both the DDS and the PPAT. Also, some are the same elements under different names, such as idiosyncratic color in the DDS and color fit in the PPAT. Thus, there are actually 30 elements. Meanwhile the C_CREATES evaluates 19 elements. For example, number of used colors, classified into a quantitative element in traditional assessments and classified into a basic element in the C_CRE-

Prologue Introduction to the Computational Art Therapy (CAT)

________________________________________________________________________ Basic elements (1) number of used colors (DAPA, DDS, FEATS). (2) list of colors (DAPA, DDS, FEATS). (3) number of clusters (new). (4) length of edges (new). (5) area of each color (new). (6) number of colored grids (new). (7) area of colored convex hull (new). (8) completeness, (new). (9) accuracy (new). (10) primary / secondary colors (new). (11) warm / cool colors (new). (12) complementary colors (new). ________________________________________________________________________ Applied elements (13) variety of colors (new). (14) main / subsidiary / background colors (new). (15) placement (DDS). (16) prominence of colors (FEATS). (17) details (FEATS). (18) space usage (DAPA, DDS, FEATS). (19) degree of concentration (new).

________________________________________________________________________

ATES is counted by the pixels colored by each color. The information is more detailed because it identifies both the existence of each color and area of each color. As another example, space usage is evaluated not only in the quantitative number of colored area but also in grade, rank, or percentile. The comparison of elements in conventional assessments and the C_ CREATES is summarized in Table P.1. We often use simply number of colors instead of number of used colors, space instead of space usage, concentration instead of degree of concentration. On the contrary, we often use main / subsidiary colors or main / subsidiary / background colors instead of main color. We will use them interchangeably where there is no possibility of confusion. Table P.1 indicates in parenthesis if the listed elements also exist in the DAPA, the DDS, and the FEATS, or if they are newly proposed in the C_CREATES. As a reference, the elements considered in conventional assessments, but not in the C_CREATES include (1) line quality, (2) abstract symbols, (3) person, and (4) implied energy. On the contrary, the elements considered in the C_CREATES, but not in conventional assessments include (1) area of each color, (2) main / subsidiary colors, (3) number of clusters, (4) accuracy, and (5) concentration. We note that the number of elements in the C_CREATES is 19, which can be increased if we divide some of the elements in detail. For example, the area of each color includes actually 15 elements according to the number of different colors such as red, blue, yellow, and so on. Likewise, the main color can also be divided into 15 elements according to the number of different colors. P.3.2 Reliability of art evaluation All ratings of elements in the DDS and the FEATS are more or less subjective and their evaluation results may differ depending on the raters. Even when raters were provided with concrete descriptors for rating all levels of each of the scales (elements), they tend to rate aspects of drawings differently simply because they liked certain drawings better than the others (White,

13 Table P.1 Taxonomy of the C_CREATES

14 Prologue Introduction to the Computational Art Therapy (CAT) Wallace, & Huffman, 2004). Even for some elements using certain selected objective criteria, such as rulers or grids made of tracing paper marked off in millimeters, this process is time consuming and there still remains a probability of some inaccuracy (Kim, 1990). Thus, we always face the problem of low inter-rater reliability or inconsistency between raters. High reliability requires special training, especially for the rating of qualitative elements. For example, the diagnostic interpretation of the DDS is difficult without the recommended training (Brooke, 2004). Inter-rater reliability in the DDS study was reported to be 95.7% (Mills, Cohen, & Meneses, 1993). The lowest agreement was 77% with the category of representation. The remainder of the categories indicated the agreement reached in the 90s %. Inter-rater reliability was found unsatisfactory by Fowler and Ardon (2002) as they felt that ratings were too heavily dependent on experience. Although they agreed that the handbook and training were clear and precise, their work suggested that clarification and precision were insufficient in practice. Brooke (2004) reported on the reliability of the PPAT: Aside from the perseveration with moderate correlation of 0.57, 0.74, and 0.52, the correlation for other scales were 0.74 and higher. In most cases, the computer systems including the C_CREATES automatically provide information that is ultimately accurate and detailed, which is something human raters cannot do. For the 12 basic elements, the C_CREATES provides perfectly accurate measurement. Examples are, number of used colors, list of colors, area of each color, number of clusters, area of colored convex hull, length of edges, etc. Also, the C_CREATES can provide not only the measurement but also the grade, rank, or percentile. Moreover, there are elements that can be provided only by the C_CREATES, not by human raters. For a simple example, raters cannot measure the area of each color. Now, for the 7 applied elements we examine reliabilities between human raters and the C_CREATES as well as the reliabilities among human raters. The C_CREATES evaluates the variety of colors so that the drawing with the largest number of used colors is ranked the highest (Kim & Hameed, 2009). When two drawings are found to have the same number of used colors, the one with the longer length of edges is ranked higher. Kim and Hameed (2009) reported high reliability between two human raters and moderate reliability between the raters and the computer system for the sample size of 52: The Rank (Spearman) Correlation Coefficient (RCC) (Walpole & Myers, 2006) between the raters was Rs = 0.825. The RCC between the computer system and Rater-1 (Rs = 0.759) was found to be higher than between the computer system and Rater-2 (Rs = 0.662). See Chapter 3 for more information. All the reliability measures in evaluation of other applied elements show the usability of the C_CREATES. See Chapters 4 - 7 for detailed information.

Prologue Introduction to the Computational Art Therapy (CAT)

15

Further studies on using this computer approach are expected to form the basis for the quantification and objectification of the currently subjective decisions made by expert raters based on individual experience. This does not imply that the computer system can replace human experts. Rather, the system can free human experts from the mundane tasks of gathering mechanical information about the drawings so that they can concentrate more on nuanced aspects of evaluation that require professional judgments by humans. In doing this, the CAT system never shirks its duty, is never fatigued with its responsibility, and never makes a mistake of omission. In conclusion, the relative merits of the CAT system compared to conven- Merits of the CAT tional art therapy assessments are as follows: (1) Objective - it provides objective information acceptable by art therapists. (2) Consistent - it provides always the same results for the same drawing. (3) Quantitative - it provides a result in quantitative terms and also in grade or percentile when desired. (4) Detail - it provides more detailed information than conventional assessments do. (5) Automatic - it provides information automatically so as to save the time and effort of art therapists. (6) Instantaneous - it provides information in a short time so as to save time of art therapists.

P.4 The computer systems for art interpretation This section surveys the studies of the DDS, the PPAT, and computer systems on their interpretation of drawings regarding drawers’ psychological states or disorders from the evaluation results of elements of drawings by the C_CREATES. Also, we discuss their validities in these tasks. P.4.1 Traditional methods and computer systems for art interpretation The purpose of the DDS and the PPAT is to gather clinical information on a person’s clinical state and psychiatric diagnosis. The validity of the DDS has been tested on various psychiatric symptoms. For example, light pressure, included animals in the free drawing, and disintegrated trees differentiated, at the significance level below 0.05, the dysthymia patients group. Unusual placement of images differentiated depressed patients. In addition, the lack of landscape in the tree picture and water images in the feeling picture differentiated depressed patients. Monochrome feeling pictures lacked integration in the free picture, and depicted short tree trunks are the characteristic of schizophrenic patients. Thus the DDS differentiated the patient groups (Cohen, Hammer, & Singer, 1988). Alzheimer patients used fewer number of used colors and less space usage compared with the control sample (Knapp, 1994). In the PPAT, energy, integration, logic, realism, problem-solving, developmental level, details, and person differentiated at least one group from

16 Prologue Introduction to the Computational Art Therapy (CAT) another at the significance level of 0.05 (Gantt, 2004). Prominence of color, details of objects and environment, and line quality differentiated children with ADHD from children with no learning or behavioral problems (Munley, 2002). Space usage, integration, realism, developmental level, and person have the strongest correlations with the Silver Drawing Test (SDT) scores (Kopytin, 2002). There have been several attempts to computerize art interpretation. Part Two introduces these eight studies of decision-making on the psychological status or disorders from the evaluation results of elements in art therapy assessments by the C_CREATES using computer technologies. Here, computer technologies include the expert system, statistical methods, and the Bayesian network, and art therapy assessments include the HFD (Human Figure Drawing) (Machover, 1949), the KFD (Burns & Kaufman, 1972), and the Structured Mandala Coloring (SMC) (Curry & Kasser, 2005). It is difficult for a computer to interpret a free drawing of vast subjects, because present computer technology cannot identify its forms. However, the C_CREATES can interpret a free drawing as far as its colors are concerned. The SMC and the first picture of the FSA are given some patterns of cluster lines. It involves only coloring work, not drawing, so there is no need for identifying forms. These are typical examples of art therapy assessments to which the C_CREATES can be fully applied. Kim, Betts, Kim, and Kang (2009) reported that the principal colors of brown, light-green, and green, number of clusters, and accuracy were selected as important elements for estimating the Mini-Mental State Examination (MMSE) score (Folstein, Folstein, & McHugh, 1975). They also estimated the probability of severe dementia by stepwise regression. Detailed explanation is given in Chapter 14. Chapter 15 compares the relative efficiencies of the PPAT, the FSA, and the SMC in doing this interpretation. Chapter 13 concerns determining whether the person who colored a SMC has a certain psychological disorder or not using statistical methods. Chapter 9 and Chapter 10 deal with the decision on the psychological states from the DAP by the expert system approach, while Chapter 11 is about the same process using the SMC, and Chapter 12 is about the same process using the p_KFD. Chapter 16 discusses a probabilistic art interpretation instead of a deterministic type. There is even an app for computer systems for the free, SMC, and KFD drawings, developed by the AAALab (which stands for Artificial intelligence, Applied statistics, and Art therapy) at Korea University (Malchiodi, 2012). This project is led by the author and a team of his colleagues and students who are interested in applying technology to art therapy and measuring graphic aspects of drawings. The app analyzes the elements related to colors in the free drawing and the SMC. Some manual intervention is needed for the KFD. At present it is in its pilot stage, but will be developed along with the progress of CAT.

Prologue Introduction to the Computational Art Therapy (CAT)

P.4.2 Validity of art interpretation Swenson (1957) has argued that Machover’s hypotheses dealing with the placement variable (element) was neither supported by later studies, and that in Roback (1968) nor was “generally” supported. Swenson (1968) reported nine positive studies supporting the hypotheses and six negatives, and concluded that such findings should be accepted with caution. There is still a heated debate among a number of art therapists about whether art interpretation can be objectively measured and whether personal judgments by non-art therapists are allowed for art interpretation (Gantt, 1998). Main difficulties in the art interpretation lie in the fact that a drawer’s personal and cultural background influences his or her choice of shapes, colors, and styles of drawing. This implies that even seemingly identical drawings may be subjected to diverse, inconsistent, and sometimes even contradictory interpretations. However, this phenomenon seems to be natural because the domain of art therapy refers to ISP in the artificial intelligence domain (Kim, Ryu, et al., 2006). For example, one study found that persons in a depressed state were likely to draw human figures that are smaller than would persons who are not in such a state (Lewinsohn, 1964), whereas another study reported there was no relationship between depression and the size of figure drawings (Salzman & Harway, 1967). In one study, omission of the mouth in HFDs indicated that a child tended to be shy and lacking of leadership (Lewinson, 1964), whereas Koppitz (1968) reported that omission of the mouth revealed fear, anxiety, and depression. King, Gullon, and Ollendick (1992) found that girls were more likely to feel fear than boys, whereas Carroll and Ryan-Wenger (1999) reported that there was no difference between boys and girls in the degree of experiencing fear. One-to-one equivalency of specific drawing elements (signs and symbols) with particular clinical features was arguably doomed from the start (Gantt, 2004). Before interpreting the meaning of signs and symbols, the context is the key to understanding how the signs and symbols function (Gantt, 2004). Kim’s interpretation methodology (Kim, Ryu, et al., 2006), having the features of several feedbacks, is not based on the one-to-one equivalence, but a process of comprehensive decision-making that considers drawers’ various circumstances including the social and cultural backgrounds in which an individual grew up, with a view to diversity and uncertainty. See Chapter 9 for a detailed explanation. As another solution for diversity and uncertainty, Ghil and Kim (2010) adopted the concept of probability in the interpretation of art. They applied the Bayesian network, an artificial intelligence technique widely applied in areas where uncertainty prevails (Burnside, 2005), which visually represents the cause-and-effect relationships among a set of random variables (elements) and their conditional independencies via a directed acyclic graph (Charniak, 1991). The traditional deterministic interpretation of art is that,

17

18 Prologue Introduction to the Computational Art Therapy (CAT) for example, children suffering from ADHD tend to draw a full-picture (more than 90% of paper). However, the probabilistic interpretation that children suffering from ADHD draw a full-picture with a probability of 0.75, and normal children with a probability of 0.1, will elevate the logical interpretation to a higher level. In particular, through a system, advantages of humans and computers can be integrated and collaborated. In this context, the Bayesian network can be a useful method to systemize and materialize knowledge in fields where knowledge itself is not organized. In addition, the Bayesian network enables art therapists to comprehend the cause-and-effect relationships between mental diseases, psychological symptoms, environment, and art elements and to integrate individual knowledge into a shared one. Moreover, the Bayesian network helps to make more objective, logical decisions by reasoning interactions of variables based on the probability value. In summary, one of the major advantages of the Bayesian network is that it enables art therapists to conduct better communications regarding theories and results, by showing probabilistic values along with the cause-and-effect relationship in a graph (Uusitalo, 2007). See Chapter 16 for a detailed explanation. Merits of the CAT In conclusion, the relative merits of the CAT system using the methodologies of the expert system, the statistical method, and Bayesian network are as follows: (1) Professional - it makes decisions derived from the expertise and experience of experts. (2) Comprehensive - it makes decisions through comprehensive knowledge. (3) Statistical - it makes decisions probabilistically and objectively based on statistical theory. (4) Systematic - it makes decisions based on the systematically classified knowledge. (5) Experimental - it knows in advance the results of applying various knowledge by experimentation.

P.5 Organization of the book Though the field of art therapy is immense, the most fundamental work is to judge the psychological state of a person by the evaluation and interpretation of the drawing created by the person. Art therapists have already recognized the significant inconsistency and subjectivity in evaluation and interpretation results depending on their subjective matter. Part One of this book, Art Evaluation, consists of 8 chapters, describing the 19 elements of C_CREATES in an objective approach to develop an objective and quantitative evaluation of elements of drawings using computer technologies. Part Two, Art Interpretation, consists of 8 chapters, describing development of computer systems which can objectively interpret a drawing based on the evaluation results obtained in Part One.

Prologue Introduction to the Computational Art Therapy (CAT)

19

This book features an edited version of the author’s 16 papers and each Part One chapter can be read independently. However, all papers are thoroughly reconstructed and rearranged. In Part One, the main drivers of the proposed methodologies are computer algorithms and statistical methods. Chapter 1 and Chapter 2 are concerned with the evaluation of 12 basic elements among the 19 elements of the C_CREATES using the techniques of digital image processing, whose accurate evaluation results cannot be disputed. Note that the 19 elements of the C_CREATES as mentioned above, are divided into 12 basic elements and 7 applied elements, depending on the presence or absence of the absolute objective evaluation methods. Chapter 1 evaluates (1) number of used colors, (2) list of colors, (3) area of each color, (4) number of clusters, and (5) length of edges. Built-in functions of the computer such as recognizing colors and counting the number of pixels and basic methods of digital image processing such as blurring and clustering are used. The evaluation results of these elements can be used to estimate the elements of Chapter 2. Chapter 2 evaluates (6) primary / secondary colors, (7) warm / cool colors, (8) complementary colors, (9) number of colored grids, (10) area of colored convex hull, (11) completeness, and (12) accuracy. Computer algorithms as well as digital image processing techniques are applied. These evaluation results obtained in Chapter 1 and Chapter 2 can be used to evaluate the elements in the following chapters of Part One. Chapters 3 - 7 are concerned with the evaluation of 7 applied elements using computer algorithms and statistical methods, whose evaluation results are consistent but can be somewhat different depending on the criteria of methods, and thus are in need of review by human raters to reassure the evaluating results. Chapter 3 devises a computer algorithm for the evaluation of (13) variety of colors. The computer determines a ranking of a drawing according to the detected number of used colors. In the case of a tie, the drawing with longer length of edges was given a higher rank. The subjective ranking by human raters has high reliability and the computer algorithm reveals high correlation with the human raters. Chapter 4 devises a computer algorithm to decide (14) main / subsidiary / background colors of a drawing. The computer system can automatically judge what main color is in a drawing. The algorithm focuses on the fact that although the main color usually occupies a large area, the main color is not simply the color with the largest area. When the drawing has a wide background, which is common in drawings, the color with the largest area is often the background color. Chapter 5 devises a computer algorithm to determine (15) placement of a drawing. The computer divides the entire sheet into several regions and considers the distribution of edge pixels in each region as the criteria for the corresponding placement category.

20 Prologue Introduction to the Computational Art Therapy (CAT) Chapter 6 develops a regression model to determine (16) prominence of color and (17) details. The computer grades and ranks the prominence and the details as a regression function of the number of used colors, the number of clusters, and the area of colored convex hull, which are selected among all of the elements in the C_CREATES evaluated in Chapter 1 and Chapter 2. Chapter 7 develops two regression models for the evaluation of (18) space usage and (19) degree of concentration. The computer grades and ranks the space usage of a drawing as a regression function of the area of each color and the area of colored convex hull and measures the degree of concentration as a regression function of number of used colors, number of clusters, and accuracy. Chapter 8, the last chapter of Part One, shows how the evaluation methods can be applied to an art therapy assessment with an example of the FSA. It also shows how the art evaluation of Part One is inter-related to the art interpretation in Part Two. Part Two In Part Two, the main drivers of the proposed methodologies are expert system, statistical regression model, and Bayesian network. In Chapters 9 - 12, we establish a small-scale prototype model of expert systems for art therapy. The knowledge base requires extensive knowledge of the characteristics (elements) of a drawing, the psychological symptoms of a drawer, the relationships between the characteristics of the drawing and the drawer’s symptoms, the diagnosis process, and the methods of knowledge acquisition. Collaboration among experts from many different fields is needed to provide such knowledge, including art therapists, psychiatrists, psychologists, artists, and knowledge engineers. We also provide details of an implementation strategy. In Chapters 9 - 12, respectively, we introduce an expert system approach to art therapy, devise an expert system’s reasoning process for art interpretation, develop an expert system for art therapy with an example of the SMC, and develop another expert system with the KFD. In Chapter 9, we develop a prototype model of the expert system to demonstrate the possibility of applying an artificial intelligence system to art therapy diagnosis. The model shows how similarities and contradictions of a mass of knowledge in art therapy are thoroughly reviewed, evaluated, classified, and organized to improve the quality of the knowledge base. In Chapter 10, we implement an expert system for the diagnosis process of art therapists. We model the complicated mechanism of this process as several procedural stages and feedbacks. We devise a suitable method of maintaining consistency among numerous decisions derived from the system. We also provide the system with a learning facility to improve its intelligence. In Chapter 11, we develop an expert system for the SMC which is selected as an art therapy tool. The system shows how the knowledge base is con-

Prologue Introduction to the Computational Art Therapy (CAT)

structed and evokes the relevant knowledge corresponding to the evaluation of the elements in the SMC and other drawer’s situations. In Chapter 12, we present an expert system knowledge base to facilitate interpretation of art and its associated knowledge acquisition method. We provide details of an implementation strategy for the KFD as a case study. We delineate the development of a computer system for the Kinetic Family Drawing with patterns (p_KFD) (Kim et al., 2011) which uses given patterns instead of free drawings. The p_KFD consists of four stages of providing clients with questionnaires, providing the client with various patterns of family members and backgrounds which possibly compose a KFD and client’s selecting a few among them, arranging, expanding or contracting, and coloring them, evaluating elements and detecting changes of evaluations in the KFD drawings, and interpreting one or several KFD drawings by invoking the knowledge in a knowledge base corresponding to the facts in the fact base. It is expected that the p_KFD with patterns would inherit and complement the validity of the traditional KFD with free drawings. In Chapters 13 - 15, we develop three regression models. The first model differentiates from a drawing whether or not the person who drew it is clinically ill. There have been many researches of this type, but most of them failed. The second model estimates the level of psychological disorder. The third model compares the efficiencies of several art therapy assessments (e.g., the PPAT, FSA, and SMC) when they are used for estimation in the second model. In Chapter 13, we computerize the SMC which finds statistically significant elements and successfully identify the group of non-patients, anxious patients, depressed patients, and schizophrenic patients based on only a SMC drawing without being given any further information. Also, we develop a regression model that successfully differentiates the four groups from a SMC drawing. This methodology can be used for other art therapy tools dealing with various other psychological disorders. In Chapter 14, we develop a regression model to estimate the degree of psychological disorders. As an application example, the degree of dementia is estimated from a SMC drawing. A regression model is formulated which explains the level of dementia scored by a test, from the evaluation results of elements in the SMC. In Chapter 15, an approach using a statistical method is proposed to compare the effectiveness of several art therapy tools in estimating the level of a psychological state, which includes various symptoms and disorders. We analyze the effectiveness of art therapy tools and measure it in a comparable quantitative term, so they can be compared with other tools. We apply this approach to three widely used art therapy tools, the PPAT, the FSA, and the SMC in estimating the level of dementia. In Chapter 16, final chapter of Part Two, we suggest probabilistic interpretation using the Bayesian network instead of traditional deterministic

21

22 Prologue Introduction to the Computational Art Therapy (CAT)

Figure P.7 Organization of the book.

interpretation as a solution for the problems of diversity, uncertainty, and even contradiction in the interpretations of drawings, The proposed Bayesian network methodology is expected to contribute to widening the scope of analysis and applications of art therapy tools, as well as the theoretical development of art therapy. The Epilogue discusses the contributions of CAT, and suggests its further development. Figure P.7 is the organization of the book showing the flow, relation between chapters and also the main computer methods and media used.

P.6 Discussion Since the advent of mankind’s first computer in 1946, categorized as the first generation computers based on vacuum tubes, the computer has progressed remarkably and continuously at a fantastic rate. We are currently in the era of the fifth generation of computers based on very-large-scale integration (VLSI) circuits, in which personal desktop computers can easily perform 1012 operations per second and store 108 pages of news articles.

Prologue Introduction to the Computational Art Therapy (CAT)

The advances in computer networks gave birth to the internet in the late 1980’s, connecting the computers throughout the world. In 1997, the computer chess player Deep Blue with artificial intelligence won against the human champion Gary Kasparov. In 2008, the computer was able to tell what a person was thinking about, for example, a car, an elephant, a house, a tree or a person from the images of functional Magnetic Resonance Imaging (fMRI) (Mitchell, Shinkareva, Carlson, Chang, Malave, Mason, & Just, 2008). Very recently in March, 2016, AlphaGo, the artificial intelligence program developed by Google DeepMind beat Lee Sedol, one of the greatest players of Go, a game which has the near-infinite number of board positions and, thus, was considered as the ultimate realm of human intuition. Rapid advancement of computer technology has brought about remarkable progress in many fields including manufacturing, design, education, law, and medicine. It has hence made a great contribution to human welfare. Austin (2009) stressed that professionals of art therapy should adopt computers as an innovative therapeutic medium, and be proficient in using them. Small, Moody, Siddarth, and Bookheimer (2009) predicted that future generations would need to master both technological and face-to-face human contact skills. Chilton, Gerity, LaVorgna-Smith, and MacMichael (2009) reported that the internet can be used as a virtual studio, where art therapists and artists can gather, create, and share their work. Kapitan (2009) asserted that this new trend is renewing the profession of art therapy through exposure to global ideas and practices, infusion of knowledge from the outside of the field, rise of grass root affiliations, and progress of the field beyond its traditional boundaries. In summary, computer technology is changing the way art therapists practice their profession (Peterson et al., 2005). A number of art therapists have already acknowledged that a scientific approach is essential to resolving implicit ambiguity and uncertainty in art therapy, and stressed that the academic sustainability of art therapy depends on science (Kaplan, 2000). Kaplan (1998) stated that subjective impressions need to be carefully evaluated in light of existing empirical evidence. He argued that, although science has not yet provided, and may remain not being able to provide, all the answers to problems related to human minds, rapid progresses are being made, and hence the subjective nature of human emotions can be better interpreted in the light of science based on empirical evidence. Gantt (2004) argued that we can develop scientifically sound and clinically effective art-based assessments. Various techniques in artificial intelligence, techniques of expert system, computer algorithms, and statistical models are emerging as useful methods to overcome limitations of art therapy. The computer systems for art evaluation and art interpretation presented in this book can be easily implemented as computer programs. There are a number of possible extensions. Examples include a computer system with a series of the SMC, the KFD, and the Kinetic School Drawing (KSD) - have

23

24 Prologue Introduction to the Computational Art Therapy (CAT) children fill out questionnaires on their color preference, ask them to color the SMC, and then complete the patterned KFD and the KSD. Then the system enables art therapists to obtain information on the children’s clinical status in their primary living space such as home and school. There has been no development of any computer systems for the Silver assessments, Levick assessment, and Bird’s nest drawing. It is difficult for the existing computer technology to identify forms and understand language, or evaluate elements in these assessments as well as the HTP, the KFD, and the DAP, whose elements include symbol, feature, and form. However, these assessments can become subjects of computerization as appropriate methods are developed through the progress of computer technology. Although a reliable approach to evaluating art places emphasis on the global aspects of form rather than on content, sign, or interpretation (Kaplan, 2003), the major challenge for computer technology related to art evaluation is that we are in a preliminary stage of technology development for understanding arbitrary drawings. Currently, there are few software packages that can only recognize characters or scenes in pictures. As we make progress in recognizing and understanding pictures and drawings, we will get closer to a fully automatic art evaluation. Recently the projective tools have encountered a number of serious challenges. Their reliability and validity have been questioned (Betts, 2005; Groth-Marnat, 1990; Chapman & Chapman, 1967; Dawson, 1984; Kahill, 1984; Klopfer & Taulbee, 1976; Roback, 1968; Russell-Lacy, Robinson, Benson, & Cranage, 1979; Suinn & Oskamp, 1969; Swensen, 1968; Wadeson & Carpenter, 1976), and many psychologists no longer use most projective tools. Groth-Marnat’s 2003 Handbook of psychological assessment, 4th edition, has even removed material on the projective tools. Given that questions have arisen about the lack of reliability and validity of projection techniques and art interpretation, the proposed approaches for the computer systems in Part Two, such as the expert systems approach (Kim, Ryu, et al., 2006; Kim, Kim, et al., 2006; Kim, Yoo, Kim, & Lee, 2007; Kim, Kim, & Kim, 2008), Bayesian network approach (Ghil & Kim, 2010), the statistical approach (Kim, Betts, et al., 2009), and the computer algorithms approach (Kim, Bae, & Lee, 2007; Kim, Kang, & Kim, 2008; Kim, 2008a; Kim, Kang, & Kim, 2009; Kim & Hameed, 2009), can contribute to making art therapy more structured and principled through scientific methods, as well as the analysis and applications of various art therapy assessments. There have been many debates and much effort to set the foundation of art therapy as a discipline (Kapitan, 2010; McNiff, 1998; Gantt & Tabone 2003; Kaiser, John, & Ball, 2006; Tibbetts, 1995). The computer systems can provide a way to overcome the limitations of conventional art therapy assessments such as subjective evaluation, contradictory interpretation, and dictionary type of one-to-one interpretation by building a structured and

Prologue Introduction to the Computational Art Therapy (CAT)

25

systematic knowledge of art therapy. Artificial intelligence techniques including expert systems and Bayesian networks present valuable approaches for making art therapy a scientific discipline. In order to advance the CAT, we need to collaborate with multiple disciplines The CAT as an such as psychology, psychiatry, art education, art therapy, and computer sci- interdisciplinary ence. We also need to address ethical issues such as security and privacy. approach The professional societies of art therapists would need to understand the importance of the CAT and prepare personnel for dissemination and initiatives. In conclusion, the CAT as an interdisciplinary approach applying computer science to art evaluation and interpretation can significantly expand the potential and the value of art therapy by overcoming limitations of existing art therapy assessments (Kim, 2008b). Art therapy, accompanied by scientific thought, can achieve its greatest effects in its purpose of improving and enhancing the physical, mental, and emotional well-being of individuals of all ages.

PART ONE ART EVALUATION

Evaluation of Basic Elements in the Computerized Color-Related Elements Art Therapy Evaluation System (C_CREATES) (I) Chapter 1 Abstract and summary Computerized methods are devised for the evaluation of 5 elements among the 12 basic elements in the Computer Color-Related Elements Art Therapy Evaluation System (C_CREATES) (Kim, Bae, & Lee, 2007; Kim, 2010), which consists of a total of 19 elements with addition of the 7 applied elements; the number of used colors, list of colors, area of each color, number of clusters, and length of edges. These elements are analyzed and evaluated by applying the built-in functions of a computer and the techniques inherent to digital image processing. Computer functions provide color recognition in a single pixel, color classification, and edge detection by identifying objects and colors. The digital image processing techniques include blurring and clustering. The evaluations of basic elements in this and the following chapters are so automatic, accurate, and correct that there is no dispute regarding its objectivity. The analysis of two sample drawings demonstrates how the computerized methods quickly and easily provide useful and fundamental information on these elements in accurate, detailed, and quantitative measures, in a significant departure from traditional method. The devised computerized methods can overcome difficulties such as subjectivity and inconsistency, and save labor time and effort of human raters.

1.1 Built-in functions of a computer 1.1.1 Color in art therapy Art therapy has been widely accepted as an effective tool in providing valuable information on the psychological or emotional states and disorders of a drawer based on his or her drawings. Color is one of the most important elements in art therapy. First, a brief examination of current literature on color and art therapy is necessary. It must be noted that these are only a few of the research efforts that focus on the important role of color in art therapy. According to Lowenfeld and Brittain (1982), color becomes significant 29

30

PART ONE ART EVALUATION

to children for the first time at about 5 - 7 years of age; at 9 years of age a child begins to use color to represent specific objects; and by the age of 14 a child is able to respond with specific emotions to different colors. Milne and Greenway (1999) presented a gender- and age-based hypothesis: statistically males and females differ in their use of color in drawings; older males tend to use fewer colors than younger males, whereas no similar divergence is shown in females. Colors have a physiological impact on human thought processes and behavior (Steinhardt, 1977). They also have an impact beyond people’s consciousness (Kreitler & Kreitler, 1980). Colors are perceived to be closely related to emotions (Hollins, Horrocks, & Sinason, 1998; Malchiodi, 1998). How people use and react to color can provide important diagnostic information regarding their current emotional status (Lev-Wiesel & Daphna-Tekoha, 2000). For instance, although colors do not mean the same thing to everyone, emotions attributed to red are often thought of as violence, passion, aggression, and anger; yellow, hospitality; brown, timidity (Hammer, 1953, 1969; Klepsch & Logie, 1982; Precker, 1950). Moreover, the therapeutic value of color has long been recognized (Ghaffurian, 1995). Rorschach (1951) regarded color as providing a means through which people can reveal emotion. There exist reports that child victims of severe sexual abuse (Malchiodi, 1990) and depressed patients (Gantt & Tabone, 1998; Wadeson, 1980) tend to use only one or two colors in their drawings. Victims of trauma express their psychological pain, anxiety, fear, sorrow, loneliness, and hopelessness through their selection of certain colors. Children who have experienced natural disasters such as earthquakes, hurricanes, and plane crashes tend to use a limited number of colors, not more than two or three, mostly consisting of black, white, and sometimes red (Gregorian, Azarian, DeMaria, & McDonald, 1996). 1.1.2 Color recognition The importance of color in art therapy is beyond dispute. Let us now turn to analyzing how a computer recognizes color. Figure 1.1a is a sample crayon drawing by a 4th grader. The computer divides a drawing sheet into many tiny pixels (dots), each of which is the final element for analysis. For example, if the vertical side is divided into 480 points and the horizontal side into 640, the sheet consists of a total of 480 x 640 = 307,200 pixels. Portion A with 15 x 20 = 300 pixels in Figure 1.1a is magnified for the purpose of explanation in Figure 1.1b. There are several color spaces such as RGB, CIELAB, CIEXYZ, HVC, HSV, LUV, etc., which have their own concepts (Wan & Kuo, 1998). Color monitors and most computer graphic systems adopt RGB color space representing colors by a 3-dimensional cube with red R, green G, and black B as in Figure 1.2a. Also the CIELAB and HVC color spaces are shown. Red, green, and blue are at the corners on each axis: Red is at (1, 0, 0), green at

Chapter 1 Evaluation of Basic Element in the C_CREATES (I)

31

Figure 1.1 Color recognition and edge detection before blurring and clustering.

(0, 1, 0), and blue at (0, 0, 1). Black is at the origin, and white is at the opposite end of the cube. The gray scale follows the line from black to white. The computer recognizes the color of each pixel, and expresses it in a set of three numbers (R, G, B). We often hear that a computer can express 16,777,216 colors. This means that when each value of R, G, and B ranges from 0, 1, to 255 or has 256 = 28 possible different values using 8-bit for each component

Figure 1.2 Three color spaces.

32

PART ONE ART EVALUATION

Figure 1.3 Color examples and (R, G, B).

of color in 24-bit color graphics, there are 256 x 256 x 256 = 16,777,216 colors. For example, color red (1, 0, 0) on the color cube is (255, 0, 0). The color of pixel (point) A in Figure 1.1b is (R=53, G=137, B=77), the color of B is (167, 97, 86), and the color of C is (188, 94, 82). On the contrary, when you designate a set of values, (R, G, B), the computer monitor produces a composite color. For example, for the 10 values of (R=184, G=43, B=37), (234, 114, 18), (233, 199, 0), (128, 191, 72), (0, 115, 59), (0, 86, 102), (45, 98, 163), (46, 49, 88), (103, 52, 122) and (143, 82, 32), their corresponding colors are shown in Figure 1.3. 1.1.3 Color classification One of the computer methods to measure the difference between two colors, (R1, G1, B1) and (R2, G2, B2) is Distance between two colors = {(R1-R2)2 + (G1-G2) 2 + (B1-B2)2}1/2. The shorter the distance between the two colors, the more similar they are. For example, in Figure 1.1b, the distance between the colors of A (53, 137, 77) and B (167, 97, 86) is 121.1; between A and C (188, 94, 82), it is 141.8; and between B and C, it is 21.6. Thus, the computer perceives that A is relatively quite different from B and C, whereas colors B and C are similar. However, readers are not required to understand the details of the equation or computer procedures. They only need to grasp the concept correctly. We will not discuss complex computer operations; this book focuses on how the readers can learn to use the books companion software package downloadable from the CD or http://www.ccthomas.com/cat. We use a set of fifteen standard colors adopted by the Korea Industry Standards (KIS) consisting of 5 basic colors, red, yellow, green, blue, and purple, and seven intermediate colors, orange, yellow-green, blue-green, blue-purple, and red-purple, pink, brown, and three colors without hue and chroma, white, grey and black. The set consisting of these 15 colors is denoted by: A15 (15 standard colors) = {red(R), orange(O), yellow(Y), yellow-green(YG), green(G), blue-green(BG), blue(B), blue-purple(BP), purple(P), red-purple(RP), pink(PI), brown(BR),white(W), grey(GR), black (BL)}.

Chapter 1 Evaluation of Basic Element in the C_CREATES (I)

_________________________________

33

Table 1.1 Color R G B Fifteen standard colors _________________________________ Red (R) 185 32 33 Orange (O) 235 111 14 Yellow (Y) 237 196 0 Yellow-Green (YG) 129 191 72 Green (G) 0 116 56 Blue-Green (BG) 0 85 102 Blue (B) 0 112 153 Blue-Purple (BP) 40 45 87 Purple (P) 101 49 121 Red-Purple (RP) 133 26 77 Pink (PI) 234 149 160 Brown (BR) 144 79 28 White (W) 241 241 241 Grey (GR) 121 121 121 Black (BL) 14 14 14 _________________________________

The number of colors that have specified knowledge and research results in art evaluation and art interpretation does not exceed 10, even when classified in detail. It is therefore acceptable and sufficient for art evaluation and art interpretation to classify the color into one of the above 15 colors. We note that the number of used colors in a drawing is an important element in art evaluation and art interpretation and henceforth, all elements considered in art therapy assessments (tools) will be mentioned in italics. In Table 1.1, the values of R, G, B are given for the colors in A15. The color of each pixel is classified into the one closest in A15. For example, the distance between color A (53, 137, 77) in Figure 1.1c and color red (185, 32, 33) is calculated as 174.3 by the above equation. The distance from the color of A to each color in Table 1.1 is calculated: orange 194.3; yellow 208.0; yellow-green 93.4; green 60.8; blue-green 78.3; blue 96.0; blue-purple 93.5; purple 109.5; red-purple 136.8; pink 199.5; brown 118.5; white 270.3; grey 82.6; black 143.6. Among them, the distance from A to green, 60.8, is shortest, and thus the color of A is classified into green. Likewise, the colors of B and C are classified into brown. The colors in A15 can be classified in more detail. For example, red can be further classified into dark red, red, or bright red. We adopt 47 standard colors which we denote as A47, the names and values (R, G, B)s of which are not given here. When colors in a drawing are classified, we suggest that they be classified using the A47 scheme first, and then into the A15 scheme. For example, when a color is closest to dark blue in A47, it is classified as blue in A15, and when closest to light green in A47, classified as green in A15. This two-step color classification is more reasonable because people usually classify a color into, for example, a reddish or bluish tone initially, and then distinguish the slight differences between color tones. Since slight

34

PART ONE ART EVALUATION

differences between color tones are not shown clearly enough to be recognized in the distance measure, classifying into A15 first often produces wrong results. Experiments have shown that using this two-step procedure leads to a more accurate outcome than using direct classification on the A15 scheme. There is a set of 157 standard colors, but this 3-step procure requires considerable computing time in calculating the distances. Figure 1.1c shows the results of color classification of all 1,200 pixels into A15. 1.1.4 Edge detection Now, the computer can detect edges, i.e., locations where the color change occurs. Figure 1.1d shows the edges detected. For example, one of the neighboring pixels of point D of color gray is a different color. When we count the number of pixels in edges, it gives us the length of edges. In Figure 1.1d, the length of edges is 428. Also, we count the number of clusters. A cluster refers to a group of contiguous pixels colored with the same color, not separated by different colors. There are 108 number of clusters as shown in Figure 1.1d. For example, region B is a cluster colored with pink, and region C is another cluster colored with brown. Figure 1.4 shows the results of analysis for the whole drawing in Figure 1.1a and an additional sample.

1.2 Techniques in digital image processing 1.2.1 Noise removing Two samples We illustrated the methods of color recognition, color classification, and edge detection by examining two crayon drawings made by 4th graders in Figure 1.4. Noise, a term in digital image processing (Gonzalez & Woods, 2002), means unintended ‘touch’ or ‘non-touch’ of the drawing material, which occurs when the thickness of the tips of crayons or width of brushes causes the drawer to color in unintended space or not fill in the intended space for coloring. Some areas are untouched and remain white, with others only slightly touched with very light color. Some colors are mixed, resulting in a completely unintended color. Drawing A represents weak noise and drawing B very strong noise in Figure 1.4. The classifications of these drawings into A47 and then into A15 are shown in Figure 1.4b, and Figure 1.4c, respectively. Both classifications contain a great deal of noise. The edges detected from these classifications in Figure 1.4d show highly complex results that render them unsuitable for the purpose of analysis. 1.2.2 Blurring The ‘noise’ of a drawing must be initially removed, and this is where blurring and clustering come in, which are technologies in digital image processing (Bartkowiak & Domanski, 1995; Wan & Kuo, 1998; Ye, Gao, & Zeng, 2003). Blurring is a method to remove noise and finer details, not

Chapter 1 Evaluation of Basic Element in the C_CREATES (I)

35

Figure 1.4 Classifications of color and detection of edges in two case examples.

36

PART ONE ART EVALUATION

Figure 1.5 Color classification and edge detection after removing noise.

intended by the drawer in digital images. We apply one of 3 x 3, 5 x 5, or 7 x 7 mask blurring (Gonzalez & Woods, 2002). Blurring is simply neighborhood averaging. For example, in a 3 x 3 mask, nine elements are added and divided by nine. The bigger the mask, the greater the blurring effects and the greater the time required for computation. We apply a 3 x 3 mask blurring to Figure 1.1b and the results are shown in Figure 1.5a. 1.2.3 Clustering Clustering of similar colors together can also be used to remove noise. We apply the method of Ye et al. (2003) that clusters color images using color feature and spatial connectivity. Clustering merges pixels in a circular mask of radius 3, 5, or 7. If the proportion of pixels of similar colors in the circle exceeds the predetermined value, the color of the cluster becomes that dominant color. We set the radius of circular mask and the value of proportion as 3 and 50%, respectively, suitable to crayon drawings of brush paintings. We get Figure 1.5b. Now after removing noise by blurring and clustering we apply the classification into A47, followed by classification into A15 and we get Figure 1.5c. 1.2.4 Edge detection Edge detection is exactly the same as in section 1.1.4 except that it is per-

Chapter 1 Evaluation of Basic Element in the C_CREATES (I)

37

Figure 1.6 Procedure of removing noises and classification of colors.

38

PART ONE ART EVALUATION

Figure 1.7 Edges detected.

formed after blurring and clustering. We obtain simpler edges in Figure 1.5d compared to the edges in Figure 1.1d, 428 length of edges. Figure 1.5d consists of 4 clusters, as numbered in Figure 1.5d. After all, three colors are used: green; pink; orange. The areas (number of pixels) for green, pink, and orange are 355(29.58%), 548(45.67%), and 297(24.75%), respectively. There are 4 number of clusters; cluster-1 of size 548, colored pink; cluster-2 of size 178, colored green; cluster-3 of size 297, colored orange; cluster-4 of size 177, colored green. The length of edges is 109 (pixels). Figure 1.6 illustrates the procedures of the entire two drawings. Figure 1.7 shows the edges that have been detected in the two drawings. The results in Figure 1.7 show the improvement by blurring and clustering compared with the edges in Figure 1.4d. Most noise has been removed, which gives the edges appropriate detail for the purpose of analysis. We can easily imagine the usefulness of these kinds of exact, detailed, and quantitative information.

1.3 Evaluation of basic elements in the C_CREATES 1.3.1 Procedures Now we will evaluate 5 basic elements in the Computer Color-Related Elements Art Therapy Evaluation System (C_CREATES) (Kim, Bae, & Lee, 2007; Kim, 2010), a system developed by the author to evaluate color-related elements in a drawing. Its taxonomy of 19 elements were introduced in the Prologue: the number of used colors, list of colors, area of each color, number of clusters, and length of edges. After blurring and clustering, we can quickly find the number of used colors and the list of colors by classifying the color of each pixel into a color in A15. It is easy for a human to count the number of clusters for each color, but it requires complex computer coding for the computer to do the same. This is a topic beyond the scope of this book, and thus will not be explained here. The area of each color can be obtained by counting the number of pixels that have the corresponding

Chapter 1 Evaluation of Basic Element in the C_CREATES (I)

39

Table 1.2 Measurement of color-related basic elements in two case drawings __________________________________________________________________________________________________ Drawing A Drawing B _______________________________ _______________________________ Color in A15 Area of each color Area of each color __________________________________________________________________________________________________ (number of pixels) Number of clusters (number of pixels) Number of clusters Red 2,852 ( 0.8 %) 1 ( 4.3 %) Orange 5,256 ( 1.8 %) 2 ( 8.7 %) Yellow 21,830 ( 7.1 %) 5 ( 21.7 %) Y-G 158 ( 0.0 %) 2 ( 8.7 %) Green 143,371 ( 46.0 %) 2 ( 8.7 %) G-B 345 ( 0.0 %) 1 ( 4.3 %) Blue 130,267 ( 42.4 %) 2 ( 8.7 %) Violet 924 ( 0.3 %) 2 ( 8.7 %) Purple 0 ( 0.0 %) 0 ( 0.0 %) R-P 0 ( 0.0 %) 0 ( 0.0 %) Pink 2,206 ( 0.7 %) 2 ( 8.7 %) Brown 155 ( 0.0 %) 1 ( 4.3 %) White 0 ( 0.0 %) 0 ( 0.0 %) Grey 216 ( 0.0 %) 2 ( 8.7 %) Black 620 ( 0.2 %) 1 ( 4.3 %) ________________________________________________

1,857 ( 0.6 %) 3 ( 5.4 %) 207 ( 0.0 %) 3 ( 5.4 %) 6,205 ( 2.0 %) 10 ( 17.9 %) 2,738 ( 0.9 %) 6 ( 10.7 %) 0 ( 0.0 %) 0 ( 0.0 %) 2,992 ( 1.0 %) 2 ( 3.6 %) 245,052 ( 79.8 %) 8 ( 14.3 %) 0 ( 0.0 %) 0 ( 0.0 %) 138 ( 0.0 %) 1 ( 1.8 %) 0 ( 0.0 %) 0 ( 0.0 %) 0 ( 0.0 %) 0 ( 0.0 %) 0 ( 0.0 %) 0 ( 0.0 %) 7,426 ( 2.4 %) 14 ( 26.0 %) 40,342 ( 13.1 %) 8 ( 14.3 %) 243 ( 0.0 %) 1 ( 1.8 %) _____________________________________

Total 307,200 23 ________________________________________________

307,200 56 _____________________________________

Number of colors used in A15 12 10 Number of colors used in A47 15 16 Length of edges 4,813 7,422 __________________________________________________________________________________________________

color. The length of edges can be obtained by counting the number of pixels in the boundary between two different colors. In Table 1.2, we show the results of the two case drawings in Figure 1.6d. In Table 1.2, drawing A in Figure 1.5 represents an analysis example. At first impression this drawing seems to use a smaller variety of colors than drawing B. However, drawing A actually uses 15 colors of A47, compared to 16 colors in drawing B, which is a very slight difference, as shown in Table 1.2. The total numbers of clusters in drawings A and B are 23 and 56, respectively, and the lengths of edges are 4,813 and 7,422. Thus, the impression that drawing A uses a smaller variety of colors than drawing B is not based on the number of used colors, but on the detailed description and complexity of the drawings. These measures enable easy and objective comparisons of the details of the drawings. This kind of information can be used to judge the details of a drawing and the effort, elaboration, and sophistication applied by the drawer. The application of these methods would have greatly simplified the methodology used in earlier studies by various authors, such as Milne and Gre-

40

PART ONE ART EVALUATION

enway (1999) who examined the differences in the number of used colors by different genders and ages, Richards and Ross (1967) who analyzed the number of used colors and the area colored, Furth (1988) who noted that special attention should be paid to out-of-placement or odd colors, such as a green cow or a purple person, which is a subscale of color fit in the Formal Elements Art Therapy Scale (FEATS) (Gantt & Tabone, 2003), and that the meaning of such a displacement needs to be determined, and Wadeson (1980) who studied the details of drawings by hospital patients suffering from depression. 1.3.2 Verification of the C_CREATES Case study I The C_CREATES can run on a common PC. Fifty drawings made by 3rd, 4th, and 5th grade children were collected for system verification. The reader can view and download all drawings in the Appendix of companion software package. Using the C_CREATES we can obtain the evaluation and analysis of 5 basic color-related elements. Ten drawings are presented in Figure 1.8. Among 50 drawings the system could not classify colors correctly in only two cases. In Figure 1.8A, the noise of the brown tone in the middle horizontal stripes has been misclassified into yellow. In Figure 1.8B, the edge between dark blue and bright blue was not detected. However, this kind of misclassification can be resolved by using the more detailed classification of 157 colors instead of 47 colors, although this would yield complicated edges and require more computer time for calculation. The system can accurately obtain the numbers of used colors and edge pixels in 48 drawings. Among the 48 samples, the number of used colors in A47 was 19.0 on average, with a standard deviation of 4.1.We can easily detect the difference in the number of used colors between Figures 1.8F and 1.8.G. Figure 1.8F uses 13 colors corresponding to a standard deviation of -1.49, two sided p value (Walpole & Myers, 2006) of 0.14. It thus uses a relatively small number of colors. Figure 1.8G, on the other hand, uses 28 colors corresponding to a standard deviation of 2.20, two sided p value of 0.03, a relatively large number of colors. In the same 48 samples, the area colored (the number of colored pixels) was 9,890 on average, with a standard deviation of 2,685. The difference in complexity between Figures 1.8H and 1.8I is immediately detected. Figure 1.8H has 6,113 length of edges (number of edge pixels) corresponding to a standard deviation of -1.37, two sided p value of 0.17, relatively simple edges. Figure 1.8I on the other hand has 13,549 length of edges corresponding to a standard deviation of 1.36, two sided p value of 0.18, relatively complex edges. Also, we can obtain other useful information. We speculate that there exists a general tendency that a drawing using a small number of colors has also a short length of edges, and vice versa. In fact, Figures 1.8F and 1.8H each use a small number of colors, 13 and 15, respectively. Both have a small number of edge pixels, 7,126 and 6,113, respectively. On the other

Chapter 1 Evaluation of Basic Element in the C_CREATES (I)

41

Figure 1.8 Analysis of sample results and other useful information.

42

Figure 1.8 (continued).

PART ONE ART EVALUATION

Chapter 1 Evaluation of Basic Element in the C_CREATES (I)

43

_______________________________________________________________________

Table 1.3 Analysis of sample Number Color (Number of clusters, Area) results and other useful ________________________ ______________________________ Figure information Colors Clusters Edges Area colored Largest 2nd largest _______________________________________________________________________ 1.1A 15 23 4,813 G(2, 46.7%) B(2, 42.4%) 1.1B 16 56 7,422 B(8, 79.8%) GR(8, 13.1%) _______________________________________________________________________ 1.8A 22 67 8,833 B(7, 68.3%) R(8, 14.5%) 1.8B 20 61 7,197 P(1, 83.4%) BR(15, 5.1%) 1.8C 18 29 6,361 B(1, 43.8%) YG(2, 41.9%) 1.8D 13 38 6,208 B(1, 43.8%) G(1, 10.0%) 1.8E 21 216 17,722 YG(11, 30.1%) Y(45, 21.5%) _______________________________________________________________________ 1.8F 13 44 7,126 B(6, 69.7%) GR(5, 18.4%) 1.8G 28 174 14,958 Y(9, 29.7%) B(14, 17.7%) 1.8H 15 48 6,113 YG(1, 80.2%) BL(7, 6.5%) 1.8 I 24 122 13,549 Y(5, 23,3%) B(14, 22.2%) 1.8 J 10 14 11,465 Y(11, 44.9%) B(7, 25.1%) _______________________________________________________________________

hand, Figures 1.8G and 1.8I each use a large number of colors, 28 and 24, with a long length of edges, 14,958 and 13,549, respectively. Of course, some opposite cases can exist. The C_CREATES easily identified such divergent cases among 48 sample drawings. For example, Figure 1.8J uses a small number of colors, 10, but a large number of edge pixels, 11,465. In a more detailed analysis, Figure 1.8G uses the largest number of colors, 28, Figure 1.8J the smallest, 10, Figure 1.8E the largest number of edge pixels, 17,722, and drawing A in Figure 1.1 the smallest, 4,813. Table 1.3 summarizes the results of analysis for ten drawings cited in this chapter with the exception of Figures 1.5A and 1.5B. Also, Table 1.3 provides the colors of the largest and second largest areas with percentages of areas colored and their numbers of clusters. The C_CREATES has been applied to evaluate drawing A in Figure 1.1a Case study II using the DAPA (The Descriptive Assessment of Psychiatric Artwork). The DAPA was developed by Hacking (1999) as a psychiatric assessment to objectively define and describe psychopathological criteria of paintings. It is comprised of 6 rating scales of mostly formal elements, the color, intensity, line, space, emotional tone, and form, which were derived from predicted psychiatric symptoms from clinical observations of psychiatric pictures drawn by patients and the relevant literature. For the evaluation of main color, the blue of the background sky has been removed by a computer algorithm that will be explained later in chapter 4. The procedure is illustrated in Figure 1.9 and the results are summarized in Table 1.4. We can instantaneously obtain not only the list of colors in the DAPA but the area

† background color ‡ Area of each color in the first line and number of clusters and % of pixels (in parenthesis) in the second line

(2, 1) 0 0 0 0 2,631 0 12,729 † 0 0 0 0 0 0 0 0 0 0 0 0 1 (17.1) 0 1 (82.9) 0 0 0 0 0 0 0 0 (2, 2) 0 0 0 0 6,444 0 8,916 † 0 0 0 0 0 0 0 0 0 0 0 0 1 (42.0) 0 1 (58.0) 0 0 0 0 0 0 0 0 (2, 3) 0 0 0 0 6,980 0 8,380 † 0 0 0 0 0 0 0 0 0 0 0 0 1 (45.4) 0 1 (54.6) 0 0 0 0 0 0 0 0 (2, 4) 0 0 0 0 5,073 0 10,287 † 0 0 0 0 0 0 0 0 0 0 0 0 1 (33.0) 0 1 (67.0) 0 0 0 0 0 0 0 0 (2, 5) 0 0 0 0 4,160 0 11,200 † 0 0 0 0 0 0 0 0 _______________________________________________________________________________________________________________________________________________________ 0 0 0 0 1 (27.1) 0 1 (72.9) 0 0 0 0 0 0 0 0

(1, 1) 0 0 0 0 0 0 15,360 † 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 (100.0) 0 0 0 0 0 0 0 0 (1, 2) 0 0 0 0 0 0 15,360 † 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 (100.0) 0 0 0 0 0 0 0 0 (1, 3) 0 0 0 0 0 0 15,360 † 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 (100.0) 0 0 0 0 0 0 0 0 (1, 4) 0 0 0 0 0 0 15,360 † 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 (100.0) 0 0 0 0 0 0 0 0 (1, 5) 2,816‡ 0 572 0 0 0 11,813 † 0 0 0 33 0 0 126 0 1 (18.3) 0 1 (3.7) 0 0 0 1 (76.9) 0 0 0 1 (0.2) 0 0 1 (0.8) 0 _______________________________________________________________________________________________________________________________________________________

(Row,Column) R O Y YG G BG B BP P RP PI BR W GR BL _______________________________________________________________________________________________________________________________________________________

Table 1.4 Application to the DAPA: Area of each color (number of pixels), number of clusters (% of pixels) _______________________________________________________________________________________________________________________________________________________

44 PART ONE ART EVALUATION

† background color ‡ Area of each color in the first line and number of clusters and % of pixels (in parenthesis) in the second line

(4, 1) 0 0 0 0 15,360 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 (100.0) 0 0 0 0 0 0 0 0 0 0 (4, 2) 0 0 349 0 15,011 0 0 0 0 0 0 0 0 0 0 0 0 1 (2.3) 0 1 (97.7) 0 0 0 0 0 0 0 0 0 0 (4, 3) 0 0 6,166 0 9,194 0 0 0 0 0 0 0 0 0 0 0 0 1 (40.1) 0 1 (59.9) 0 0 0 0 0 0 0 0 0 0 (4, 4) 0 173 1,786 85 12,569 0 747 0 0 0 0 0 0 0 0 0 1 (1.1) 2 (11.6) 1 (0.6) 1 (81.8) 0 1 (4.9) 0 0 0 0 0 0 0 0 (4, 5) 0 820 74 0 13,333 0 1,133 0 0 0 0 0 0 0 0 _______________________________________________________________________________________________________________________________________________________ 0 1 (5.3) 1 (0.5) 0 1 (86.8) 0 1 (7.4) 0 0 0 0 0 0 0 0

(3, 1) 0 0 0 0 13,098 0 2,262 † 0 0 0 0 0 0 0 0 0 0 0 0 1 (85.3) 0 1 (14.7) 0 0 0 0 0 0 0 0 (3, 2) 0 0 412 0 14,948 0 0 0 0 0 0 0 0 0 0 0 0 2 (2.7) 0 1 (97.3) 0 0 0 0 0 0 0 0 0 0 (3, 3) 0 0 11,269 0 4,091 0 0 0 0 0 0 0 0 0 0 0 0 1 (73.4) 0 4 (26.6) 0 0 0 0 0 0 0 0 0 0 (3, 4) 0 1,266 846 0 12,124 0 611 † 0 0 0 279 0 0 65 169 0 1 (7.8) 1 (5.5) 0 1 (78.9) 0 1 (4.0) 0 0 0 1 (1.8) 0 0 1 (0.4) 1 (1.1) (3, 5) 0 2,639 177 0 8,192 0 803 † 414 0 0 1,863 443 0 0 829 0 1 (17.2) 1 (1.2) 0 3 (53.3) 0 1 (5.2) 1 (2.7) 0 0 1 (12.1) 2 (2.9) 0 0 1 (5.4) _______________________________________________________________________________________________________________________________________________________

(Row,Column) R O Y YG G BG B BP P RP PI BR W GR BL _______________________________________________________________________________________________________________________________________________________

Table 1.4 (continued) _______________________________________________________________________________________________________________________________________________________

Chapter 1 Evaluation of Basic Element in the C_CREATES (I)

45

46

PART ONE ART EVALUATION

Figure 1.9 Application of the system to the DAPA.

of each color and the number of clusters for each color.

1.4 Discussion We have removed the noise of a drawing employing the methods of blurring and clustering in digital image processing, which is a field of artificial intelligence. The question remains as to how to characterize the noise inherent in crayon drawings. This problem requires the development of a special and efficient method, one that allows an automatic setting of the values of certain parameters such as the number of repetitions, the size of mask in blurring, and the radius of core pixel and the minimum ratio of the same color in clustering. Also, as we saw in Figures 1.8A and 1.8B, a color can be classified as a wrong color. There are various reasons for this such as the scanners or monitors being used, or the environment of surrounding lights, etc. To solve this problem, see the color calibration in the Appendix of companion software package manual. Moreover, when the drawing material is pastels as in the Diagnostic Drawing Series (DDS) (Cohen, 1986/1994) and markers as in the Person Picking an Apple from a Tree (PPAT) (Gantt, 1990), the suitable parameters should be determined. It is expected that an expert system can provide such a method. Extending the system to evaluate the remaining elements in the C_CREATES will be the subject of the following chapters in Part One. Determi-

Chapter 1 Evaluation of Basic Element in the C_CREATES (I)

nation of whether or not a color is a background color may be a relatively simple problem that can be solved by identifying the characteristics of background color and incorporating relevant knowledge into an expert system. Once background color is removed, we can evaluate the element of space usage. Likewise, we can determine the primary / secondary colors. We can also judge whether placement of a drawing on a paper is balanced or skewed through the distribution of edge pixels. In this manner we can evaluate the variety of colors, elaboration, emotional tone, etc. These topics will be discussed in Chapters 3 - 7.

47

Evaluation of Basic Elements in the C_CREATES (II)

Chapter 2 Summary In continuation from Chapter 1, the remaining 7 basic elements are evaluated: primary / secondary colors, warm / cool colors, complementary colors, number of colored grids, area of colored convex hull, completeness, and accuracy. The first three of the elements directly related to color can be easily evaluated by counting the colored pixels using the corresponding standard colors. Three mandala colorings show the evaluation procedures. The fourth and fifth space-related elements are evaluated by counting the colored grids and the colored pixels in convex hull, respectively. A sample drawing shows the concept and usefulness of the elements. The last two pattern-related elements are evaluated by computer algorithms. Two Structured Mandala Coloring (SMC) samples show how computerized methods provide fundamental information on these elements. Through this process, the evaluation of all the 12 basic elements can be completed accurately, quantitatively, and objectively with no room for disputes over its results.

2.1 Elements related with color definitions, space colored, and pattern coloring

In Chapter 1, we divided the entire length and width of the drawing sheet into tiny pixels, recognized the color of a pixel, and classified the color into one of the standard 15 colors. In this chapter, we evaluate the element of primary / secondary colors by measuring (counting) the colored area (pixels) with the corresponding colors. As we know, primary colors are red, blue, and yellow, and secondary colors are purple, green, and orange. After blurring and coloring using the techniques of digital image processing, the built-in color recognition function of the computer allows us to easily obtain information on the primary colors and secondary colors that are used, in the form of the number of colored pixels or the percentage of the colored pixels to the entire pixels. 48

Chapter 2 Evaluation of Basic Elements in the C_CREATES (II)

49

Figure 2.1 A structured mandala and its sample coloring.

In the same way, we can obtain the percentages of warm / cool colors and also complementary colors. These kinds of information on colors are known to be useful in inferring the drawer’s personality, character attributes, etc. To illustrate the above procedure, we use pattern coloring. It refers to coloring a given pattern of geometric lines or figures such as a flower, person, sun, river, or trees, instead of drawing such figures. The basic idea of coloring therapy is that, when individuals color complex geometric forms, they are provided with an opportunity to suspend their “inner dialogue” and to deeply engage in an activity that removes them from the flow of negative thoughts and emotions that can sometimes dominate their lives (Curry & Kasser, 2005). Curry and Kasser (2005) promised the following regarding this tool: Although, to our knowledge, coloring therapy has not been empirically tested or widely discussed in scholarly discourse, other research suggests that it may indeed hold promise as an effective tool for alleviating anxiety, as it combines elements of art therapy (i.e., coloring a form) and meditation (i.e., deeply concentrating on an experience that is soothing). (p. 81). A structured mandala is a kind of pattern with given geometric lines in a circle. We call this art therapy tool the Structured Mandala Coloring (SMC). In Figure 2.1, we provide a pattern of the structured mandala to color and an example of its coloring. See Chapter 11 for more explanation on the SMC and its usefulness. There are other kinds of pattern colorings as well: the Face Stimulus Assessment (FSA) uses the pattern of a human face, as discussed in Chapter 8: and the patterned Kinetic Family Drawings (p_KFD) (Kim, Han, Kim, & Oh, 2011) uses the pattern of typical family members

50

PART ONE ART EVALUATION

doing something, as discussed in Chapter 12. Next, we evaluate two elements which are related with space usage, the number of colored grids and the area of colored convex hull. Suppose we measure how much space of the sheet a drawer used in his or her drawing. One method may be to count the colored pixels. Another method can be to divide the sheet into several grids and count the colored grid. A grid is a rectangle and it is considered colored when more than a given proportion, e.g. 30 percent, of the pixels in the grid are colored. Also another method can be to count the pixels in the colored convex hull. The convex hull is defined as a convex set consisting of colored pixels. For the last two methods we devise computer algorithms. The methods are illustrated via three PPAT drawings developed by Gantt (1990). It is shown how these elements are used in the evaluation of space usage in Chapter 7. Finally, we evaluate two elements, the completeness and the accuracy in coloring a given pattern. The completeness measures how successful the drawer was in completing the coloring of the patterns, and the accuracy measures how accurately he or she colored in accordance to the given patterns. We devise two algorithms for these evaluations.

2.2 Primary / secondary, warm / cool, and complementary colors 2.2.1 Elements related with color definitions Once the list of colors and the area of each color have been analyzed and determined, the percentage of number of pixels colored with primary / secondary colors, the percentage of warm / cool colors, and the list of complementary colors can be determined from the knowledge base which is appropriately constructed according to their definitions. The principal color and the subsidiary color are defined as follows. The definitions are stored as knowledge in a knowledge base and are invoked whenever needed. Once the area of each color is evaluated, the system deduces information on the elements from an appropriately constructed knowledge base. [knowledge 2.1] The principal color is the color used in the largest area. [knowledge 2.2] The subsidiary color is the color used in the second largest area, not less than a given percentage, e.g. 20 %, of the pixels in a drawing. We note that these elements are so obvious and redundant with the elements of area of each color that they are omitted in the basic elements in the C_CREATES. For easy reading, we replicate the explanations in Chapter 1. The vertical side of a sheet is divided into 480 points and the horizontal side into 640 points. Thus, the drawing sheet consists of a total of 480 x 640 = 307,200 pixels. The process of blurring and clustering removes the noise and results

Chapter 2 Evaluation of Basic Elements in the C_CREATES (II)

51

Figure 2.2 Example of color analysis.

in adequate edge detection suitable for analysis. Then, the color of each pixel is transformed into the closest color in A15 consisting of the following standard colors, A15 = {red (R), orange (O), yellow (Y), yellow-green (YG), green (G), blue-green (BG), blue (B), blue-purple (BP), purple(P), red-purple (RP), pink (PI), brown (BR), white (W), grey (GR), black (BL)}. Figure 2.2a shows the results of the color transformation of the SMC example in Figure 2.1b after blurring and clustering. In Figure 2.2b, purple occupies the largest area with 30.1 % of the circle and red the second largest area with 22.0 %. Thus, purple is determined as the principal color and red as the subsidiary color. 2.2.2 Definitions The primary / secondary colors are defined as follows.

52

PART ONE ART EVALUATION

[knowledge 2.3] The primary colors consist of red, blue, and yellow. [knowledge 2.4] The secondary colors consist of green, purple, and orange. In Figure 2.2c, the primary colors are represented by red, and the secondary colors by green. The primary colors occupy 40.4 % of the circle, and the secondary colors 34.0 %. The warm / cool colors are defined as follows. [knowledge 2.5] The warm colors consist of red, orange, and yellow. [knowledge 2.6] The cool colors consist of blue and deep blue. In Figure 2.2d, the warm colors are represented by yellow and the cool colors by blue. The warm colors occupy 40.4 % of the circle and the cool colors 4.0 %. The complementary colors are defined as follows. [knowledge 2.7] The complementary colors are (1) red and blue-green, (2) blue and orange, (3) blue-purple and orange, (4) yellow and purple, or (5) green and purple among the principal color, subsidiary color, and the color occupying the third largest area with more than a given percentage, e.g. 12 %, of a drawing. In Figure 2.2b, yellow is the color with the third largest area with 14.4 %, and thus there are complementary colors of purple and yellow.

2.3 Number of colored grids and area of colored convex hull 2.3.1 Elements related with space colored Computer algorithm We devise three computer algorithms to evaluate the area colored, the number of colored grids, and the area of colored convex hull. A computer algorithm refers to a sequence of instructions or computational methods to solve a given problem. The drawing sheet is divided into several grids of rectangles. The colored convex hull is defined as a convex set consisting of colored pixels. All points on a line connecting any two colored points belong to a colored convex hull and are considered to be colored. Figure 2.3 shows a PPAT drawing, illustrating how the number of colored grids and the area of a colored convex hull are measured. It has 405,274 colored pixels, 13 number of colored grids, and 1,078,614 pixels area of colored convex hull. From these measures and the number of colored pixels, Table 2.1 summarizes the evaluation results of the elements considered in Chapter 1 (the number of used colors, list of colors, and area of each colors are omitted). See Chapter 7 for a more detailed explanation. Readers who are not interested in fully understanding the somewhat mathematical terminology of “convex hull” or the somewhat detailed sequence of the computer algorithm may skip the rigorous definition and the related steps. It is sufficient if one grasps the concept of convex hull from viewing Figure 2.3 and the outline of the algorithm from viewing Figure 2.4. The readers can evaluate the elements in

Chapter 2 Evaluation of Basic Elements in the C_CREATES (II)

53

Figure 2.3 A sample drawing illustrating the grid and convex hull.

their drawings using the companion software package downloadable from the CD or http://www.ccthomas.com/cat. 2.3.2 Algorithms Step-1. Divide the entire drawing sheet into pixels, each of which is the final Area colored element to be analyzed. The vertical side is divided into 1,600 points and the horizontal side into 978, so that the sheet consists of a total of 1,564,800 pixels (1,600 x 978). Step-2. Determine whether each pixel is colored, using the built-in computer functions. Step-3. Count the number of pixels colored. Step-1. Divide the entire drawing sheet into grids. The vertical and horizon- Number of colored tal sides are divided into 5 equal segments, so that the sheet consists grids of a total of 25 grids (5 x 5). Step-2. Determine each grid as colored if more than 20% of it is colored. Step-3. Count the number of grids colored. Step-1. Among 1,600 horizontal lines, find those having more than one col- Area of colored ored pixel. When only one pixel is colored, it is considered a line. convex hull If there exist no such lines, evaluate the area of colored convex hull as 0, and stop. Step-2. Find the leftmost and rightmost pixels colored in the lines obtained in Step-1. Connect the leftmost pixel with the rightmost pixel in the first line. When a line consists of one pixel, the leftmost pixel is also a rightmost pixel. Repeat until one reaches the last line. Step-3. Set the current left corner as the leftmost pixel in the first line obtained in Step-2. If there is only one line, go to Step-9.

_______________________________________________________________________

Table 2.1 Evaluation results of Number of Number of Area Length of Number of Area of colored elements in Figure 2.3 used colors clusters colored edges colored grids convex hull _______________________________________________________________________ 4 86 405,274 51,603 13 1,078,615 _______________________________________________________________________

54

PART ONE ART EVALUATION

Figure 2.4 Simplified flow of the computer algorithm for measuring the area of colored convex hull.

Step-4. Calculate the angles of left pixels in the following lines from the current left corner. Find the pixel having the largest angle. This is the next left corner. Draw a line connecting the current left corner with the next left corner. Step-5. If the next left corner is not in the last line obtained in Step-1, set the next left corner as the current left corner, and go to Step-4.

Chapter 2 Evaluation of Basic Elements in the C_CREATES (II)

55

Figure 2.5 Illustration of accuracy measurement.

Step-6. Set current right corner as the rightmost pixel in the first line obtained in Step-1. Step-7. Calculate the angles of right pixels in the following lines from the current right corner. Find the pixel having the smallest angle. This is the next right corner. Draw a line connecting the current right corner with the next right corner. Step-8. If the next right corner is not in the last line obtained in Step 1, set the next right corner as the current right corner, and go to Step-7. Step-9. Count the pixels within the convex hull consisting of lines obtained in the above steps, and stop.

2.4 Completeness and accuracy 2.4.1 Elements related with pattern coloring The completeness of coloring measures how much of the coloring a drawer finishes during a given period of time, and the accuracy measures the extent to which a drawer colors in accordance with the given pattern. The completeness can be easily measured by calculating the ratio of the number of colored pixels to the number of whole pixels in the circle. The accuracy is the accordance of the coloring clusters with the pattern clusters. Here, a coloring cluster refers to a contiguous area colored with the same color, not separated by different colors, which is simply the cluster explained in Chapter 1. This can be obtained from the computer analysis of colors and the detection of color edges. A pattern cluster refers to an area surrounded by a closed line of given patterns in circle, without any other lines within it. Figure 2.5 illustrates the method. In Figure 2.5a, the coloring cluster of Figure 2.1b is detected, and in Figure 2.5b, it is overlapped with

56

PART ONE ART EVALUATION

the pattern cluster of Figure 2.1a. The number of coloring clusters is 18, whereas the number of pattern clusters is 64. 2.4.2 Algorithm We have developed a computer algorithm for the evaluation of accuracy. Readers who are not interested in the detailed procedure can skip the following steps. Algorithm for the Step-1. Count the number of pixels in the circle, A. measurement of Step-2. Count the number of pixels colored outside of the circle, B. accuracy Step-3. Count the number of pixels for each of the 15 colors in each of the 64 pattern clusters, C(i, j), where index i (= 1, 2, ... , 15) denotes the color and index j (= 1, 2, ... , 64) denotes the cluster. Step-4. Choose the color with the largest number of pixels, kj, and set it as the color accurately colored in each of the 64 pattern clusters, C(kj, j) = max i {C(i, j)}, j = 1, 2, ... , 64. Step-5. If white is chosen, assume that there exists no color accurately colored. Step-6. Add the numbers of pixels for all clusters, D = ∑ j C(kj, j). Step-7. Subtract the number of pixels obtained in Step-1, E = D - B. Step-8. Divide the result of Step-7 by the number pixels in the circle in Step-1, F = E / A. The measure of accuracy is given as the value of F in the final step. We note that B acts as a kind of penalty, which decreases accuracy. Applying the computer algorithms to Figure 2.2a, the completeness is obtained as 82 % and the accuracy as 74.6 %. Progress detection We add two SMC samples in Figure 2.6 colored by the same person at different times. The evaluation results of Figures 2.6A and 2.6B on the 7 basic elements are given in Table 2.2. The computerized methods developed not only provide this kind of information of the C_CREATES, but also detect any changes in the elements between two drawings made by the same patient. We note that clinicians are under pressure to demonstrate the progress of their patients and that drawings can be compared to determine the course of patient treatment (Betts, 2006). In art therapy, for instance, an art therapy assessment can be administered at the onset of treatment, during the middle phase of treatment, and again upon its termination of services, allowing art therapists to find any changes in the drawings once the evaluations are stored in a data base. The methods can detect any changes in these elements when a series of drawings is given and provides the information in the form

Chapter 2 Evaluation of Basic Elements in the C_CREATES (II)

57

Figure 2.6 Changes in the elements evaluation.

58

PART ONE ART EVALUATION

Table 2.2 Summary of analysis

__________________________________________________________________________________________________ Number Number Primary Secon- Complemen- Warm Cool Complete- Accuracy Figure of used of colors dary tary colors colors ness (%) colors clusters (%) colors colors (%) __________________________________________________________________________________________________ 2.6A 4 19 33.6 65.3 orange / blue 38.2 29.3 0.45 60.1 2.6B 7 15 59.4 32.7 orange / blue 63.2 18.6 0.87 75.6 __________________________________________________________________________________________________

of quantitative data. For example, except for the number of clusters, all of the elements, namely, the number of used colors, completeness, and accuracy, in Figure 2.6B are larger or higher than those in Figure 2.6A. Also, the ratio of warm colors in Figure 2.6B is higher than that in Figure 2.6A. These facts are observable to the naked eye. However, the usefulness of the system lies in the fact that it can analyze and produce results in quantitative terms.

2.5 Discussion and conclusion We emphasize the fact again that the evaluation of 7 elements in this chapter, as well as the 5 elements in Chapter 1, is performed by mechanical and automatic computerized methods, ensuring its accuracy and objectivity. The results of evaluation are not subject to dispute. Obviously, computerized methods can save the time and effort of art therapists. Some of the elements proposed in the C_CREATES are new or quite different from the ones used in traditional art therapy assessments in respect to contents, details, measurement units, and methods of evaluation. For example, the area of a colored convex hull is a new element which can serve as an appropriate basis for the evaluation of space usage as we will see in Chapter 7. The information gathered through quantitative and detailed measures are much more useful for decision-making in comparison with the information gathered through traditional methods. For example, the data showing “30.1 percent of purple and 14.4 percent of yellow as complementary colors” gathered in the computerized evaluation provide more information than the data of “red and yellow as primary colors” derived from the traditional evaluation. Finally, we note that these evaluation results are stored in a data base and can be retrieved whenever needed, allowing art therapists, for example, to compare drawings at different times and track changes in a given time period.

A Computer System for Ranking Variety of Colors

Chapter 3 Summary We have developed a computer system that uses a simple rule to systematically evaluate variety of colors of drawings. This rule can solve the problems of subjectivity, preconception, and bias that can exist in this type of human decision-making process. The system utilizes the elements of number of used colors and length of edges. Based on the result of ranking 52 children’s crayon drawing samples, we found high inter-rater reliability in human ratings of the variety of colors and high correlation between the rankings given by the computer system and those by human raters. This finding implies that this computer system can provide a useful aid to human raters and, moreover, suggests the possibility of computer evaluation replacing human decisions, especially when human decisions lead to indecisive results. Furthermore, computer rating produces the results with high accuracy because it eliminates the issues of carelessness or fatigue that can arise when human raters process dozens or hundreds of drawings at a time.

3.1 Importance of color-related elements Art evaluation is widely accepted as a valuable clinical technique for mental health professionals (Oster & Gould, 2004). Many art therapy assessments have been developed to evaluate various elements of drawings so as to provide helpful information regarding drawers’ psychological or emotional states. Historically, the use of color in drawings has been considered one of the most important elements, along with theme, line, form, and other elements (Ghaffurian, 1995). Rorschach (1951) regarded color as a means for revealing a person’s emotion. Some art therapists have reported that child victims of severe sexual abuse (Malchiodi, 1990) and patients suffering from depression (Wadeson, 1980; Gantt & Tabone, 1998) tend to use only one or two colors in their drawings. One study showed that patients diagnosed with substance abuse use less color than those in the comparison 59

60

PART ONE ART EVALUATION

group (Francis, Kaiser, & Deaver, 2003). Children who have experienced natural disasters tend to use a limited number of colors (not more than two or three) and a palette mostly consisting of black, white, and sometimes red (Gregorian, Azarian, DeMaria, & McDonald, 1996). Survivors of trauma express their psychological pain, anxiety, fear, sorrow, loneliness, and hopelessness by preferring particular colors. A drawing contains various color-related elements that are believed to reveal the drawer’s psychological state at the time of its creation. These include the number of used colors, hues, prominence of color, emotional tones of color, and color mix. Most art therapy assessments include various color-related elements in their evaluation systems. For example, the DAPA includes hues and intensity; the DDS incorporates color type, blending, and idiosyncratic use of color; and the FEATS includes prominence of color and color fit. Subjectivity However, certain elements such as the elaboration and the emotional tone may be particularly subjective in creating an accurate evaluation. The limitations common to all art therapy assessments and research raise the problem of subjectivity in evaluation of elements. Raters are often compelled to proceed in their work on the basis of subjective and rather uncertain knowledge, relying on professional observation and judgment. Although raters usually are provided with concrete descriptors for their ratings, it is still conceivable that their rating of certain elements of drawings may be biased simply because of their disposition to certain styles of drawings (White, Wallace, & Huffman, 2004). All evaluations of elements are more or less subject to human bias and, therefore, the results may vary depending on the raters. Thus, the value and necessity of developing a computer system as an objective evaluation tool to aid human raters are thus critical. In this chapter, we propose the variety of colors of drawings as a new element and develop a computer system for its objective evaluation. We define the variety of colors as the comprehensive use of color in evaluation performed on the basis of the rater’s personal knowledge, experience, and intuition. We have evaluated the 12 basic elements described in the previous chapters. The overall variety of colors found in a given drawing can be evaluated based on these elements. Simple rule We propose a simple rule of evaluation in ranking the variety of colors of drawings: the larger number of used colors, the higher the rank. In the case of a tie, we assign a higher rank for the use of longer length of edges. We collected a sample of 52 children’s crayon drawings and two raters ranked them in order of the variety of colors. High reliability between the human raters and high correlations between the human raters and computer system have been observed.

Chapter 3

A Computer System for Ranking Variety of Colors 61

3.2 Method 3.2.1 Rule Drawings are ranked in order of the variety of colors. Our premise is that the computerized evaluation system can serve as an objective tool and an aid to human raters. First, the human raters rank the drawing samples by comparing the variety of colors comprehensively. They are blinded from each other’s ratings. The reliability of evaluation between raters is then examined by using the Spearman’s Rank Correlation Coefficient (RCC) (Walpole & Myer, 2006). Next, the computer ranks the same set of drawing samples by detecting the number of used colors. In the case of a tie, the drawing with longer length of edges is given a higher rank. To detect the number of used colors and the length of edges, we use the C_CREATES. Finally, the RCCs are used as non-parametric measures of correlations in the pairs of rankings given by the two human raters and the computer system. Here, the pairs refer to three different pairs: Rater-1 and Rater-2; Rrater-1 and the computer; Rater-2 and the computer. Denoting the difference of the ranks between two sets of rankings in the ith drawing (i = 1, 2, ... , n) as di, then the RCC is n

Rs = 1 - 6 ∑ di2 / {n (n2 - 1)}. i=1

Here, - 1 ≤ Rs ≤ 1, Rs = 1 is found when the two ranks are identical for every drawing and Rs = - 1 when the ranks are completely in the reverse order. Next, the test statistic Rs = r (n - 1)1/2 is used, which is asymptotically standard normal as the sample size increases to over 30 (Walpole & Myer, 2006), to determine whether there exist correlations in the pairs of rankings given by the two human raters and the computer system. We test the null hypothesis of no correlation versus the alternative hypothesis of positive correlation. The critical region at significance level of 0.05, for n = 52 is z ≥ 1.6448. We may conclude that the computer’s evaluation provides useful information to human raters if the null hypothesis is rejected. 3.2.2 Sample We collected 52 crayon drawing samples (n = 52) from third to fifth grade elementary school students with no known history of psychological or emotional disorders. We selected crayons because it is the most popular drawing material in Korea. Two art therapists compared the variety of colors in pairs of drawings and assigned a higher rank to the drawing with a greater variety of colors. One rater was an expressive arts psychotherapist registered with the Korean Expressive Arts Psychotherapy Association and the other was a color psychology instructor at the Heart & Color School in Korea. The raters were asked to compare the variety of colors of each pair of drawings based on their overall impression. We expected that the subjectivity of the rater would be inevitable in the rating of color because of the individual

62

PART ONE ART EVALUATION

Figure 3.1 Drawings ranked Rank-1, Rank-39, and Rank-52 by Rater-1.

rater’s particular intuition and color perception. The raters were not given any definitions of the variety of colors or specific criteria for its evaluation, but were just simply asked to “rank the variety of colors of drawings.” In the context of judging the variety of colors of a drawing, the issue of rater subjectivity arises due to the difference in raters’ understanding of the term “variety.” Had we provided them with a clearer, more specific definition of variety of colors, we might have obtained greater consistency in their ranking. We assume that the principle of the transitivity applies to the variety of colors: If drawing A is ranked higher than drawing B, and drawing B is, in turn, ranked higher than drawing C, then drawing A is ranked higher than

Chapter 3

A Computer System for Ranking Variety of Colors 63

Figure 3.2 Drawings ranked by the computer as opposed to the drawings of the same ranks by Rater-1 from Figure 3.1.

drawing C. Thus, all possible pairs need not be compared. The drawing with the greatest variety of colors is the most highly ranked, the drawing with the second greatest variety of colors is ranked second, and so on. The drawings are ranked from 1 to 52 and no tie is allowed. In our data collection, 52 drawings are denoted by A, B, ... , Z, a, b, ... , z, the two raters by as Rater-1 and Rater-2, and the 52 ranks by Rank-1, Rank-2, ... , Rank-52. Figure 3.1 shows 3 drawing samples rated by Rater-1 as Rank-1, Rank-39, and Rank-52. The computer evaluated the variety of color of the same set of drawings. In case of drawings with the same number of used colors, the one with the longest length of edges was ranked the highest. For the detection of the

64

PART ONE ART EVALUATION

number of colors and the length of edges, we used the C_CREATES: after blurring and clustering, the color of each pixel was classified as the closest to one of 47 standard colors defined by the Korea Industry Standards (KIS). Blurring and clustering are methods to remove “noise” in crayon drawings. Noise is a technical term in digital image processing that refers to unintended touches or un-touches due to the thickness of the crayon head. We expressed colors in Munsell’s color system, called HVC (representing the three elements of colors, hue H, brightness V, and chroma C), which has been accepted as being most similar to humans’ perception of colors (Wan & Kuo, 1998). A specific color is designated by the numerical values H, V, and C. For example, H = 7.5, V = 4, and C = 14 for red, H = 5.8, V = 5, and C = 12 for yellow, and H = 10, V = 4, and C = 10 for blue by the KIS. As a measure of similarity between two colors, a distance between them is defined so that the computer can determine the color closest to any given color. There is no problem in using the HVC color system instead of the RGB system described in Chapter 1. Similar result is expected. Figure 3.2 shows the three drawings ranked by the computer as opposed to the drawings of the same ranks by Rater-1 from Figure 3.1.

3.3 Results 3.3.1 Inter-rater reliability The RCCs in pairs of rankings given by the two human raters and the computer system are presented with the raw data in Table 3.1. The scatter plot of ranks between the two human raters is shown in Figure 3.3. The comparison of the computer detection of the number of used colors and the length of edges of drawings that received Rank-1, Rank-13, Rank-26, Rank-29, and Rank-52 between Rater-1, Rater-2, and the computer are summarized in Table 3.2. Rater-1 ranked drawings J, f, b, Z, and N as Rank-1, Rank-13, Rank26, Rank-39, and Rank-52, respectively. The number of used colors of each drawing was 25, 20, 23, 19, and 11. One thus observes the general tendency that the rank order assigned by Rater-1 was proportional to the number of used colors of the drawing; that is, the greater the number of used colors of a drawing, the higher the rank. The same result occurred with Rater-2. As can be seen in Figure 3.3, six drawings out of a total of 52 samples, J of Rank-1, X of Rank-3, p of Rank-6, o of Rank-10, m of Rank-14, and N of Rank-52, were given the same ranks by both raters. As an example of a relatively large discrepancy in ranks, C has Rank-38 by Rater-1 and Rank15 by Rater-2. The RCCs and the test statistic are Rs = 0.8249 as in Table 3.1, z = 5.8909 (p-value < 0.0001) respectively. The relatively large value of the RCCs indicates high reliability between the two raters. We may also conclude that correlation exists between the ranks from the

Chapter 3

A Computer System for Ranking Variety of Colors 65

__________________________________

Table 3.1 RCCs in the pairs of RCC (Rs) _______________________ rankings given by the two human raters and Raters 1 2 3 the computer system __________________________________ with raw data 1. Rater-1 − .8249 .7592 2. Rater-2 − .6621 3. Computer − __________________________________

________________________________________________________________________ Code of drawing

Number of used colors

Ranks _______________________________

Length of edges

Rater-1 Rater-2 Computer ________________________________________________________________________ A 19 9,558 32 30 32 B 17 8,942 40 35 42 C 14 14,207 38 15 47 D 18 11,443 48 42 36 E 15 8,916 44 33 46 F 19 9,804 19 8 31 G 20 11,323 34 38 27 H 16 6,386 35 36 45 I 11 5,972 46 37 52 J 25 14,181 1 1 7 K 22 12,000 12 18 17 L 17 9,630 47 47 41 M 21 12,030 11 17 20 N 11 10,607 52 52 51 O 21 13,221 18 13 19 P 22 12,680 7 28 15 Q 18 10,629 20 40 38 R 26 15,992 16 20 2 S 21 11,369 24 23 21 T 16 12,362 33 43 44 U 5 11,043 31 22 8 V 19 9,209 30 21 33 W 21 9,242 43 44 22 X 24 16,992 3 3 9 Y 25 14,850 9 5 5 Z 19 10,982 39 31 30 a 18 11,427 27 29 37 b 23 16,847 26 24 10 c 29 10,895 22 11 1 d 22 12,954 42 41 13 e 21 14,726 21 32 18 f 20 12,626 13 7 24 g 17 11,350 23 19 40 h 25 14,188 4 2 6 i 14 7,274 41 50 48 j 13 8,577 51 39 50 ________________________________________________________________________

66 Table 3.1 (continued)

PART ONE ART EVALUATION

________________________________________________________________________ Code of drawing

Number of used colors

Ranks _______________________________

Length of edges

Rater-1 Rater-2 Computer ________________________________________________________________________ k l m n o p q r s t u v w x y z

25 15,908 2 4 4 19 12,311 36 48 29 22 12,196 14 14 16 20 13,881 29 16 23 20 12,575 10 10 25 23 12,194 6 6 12 23 14,997 8 12 11 18 20,270 49 45 34 17 15,274 28 49 39 18 11,713 37 34 35 16 13,994 50 51 43 25 21,364 5 25 3 19 12,718 15 9 28 22 12,725 17 27 14 20 12,304 25 26 26 13 16,713 45 46 49

________________________________________________________________________ large value of the test statistic. It is interesting that the raters thought there would be little correlation since they were told only to compare “variety of colors” without being given any criteria and the definition of the variety of colors. The demonstration of high reliability despite the absence of a clear definition may be ascribed to human agent’s perceptual ability to make collective and comprehensive decisions. 3.3.2 System validity

Figure 3.3 Scatter plot of the ranks between Rater-1 and Rater-2.

Chapter 3

A Computer System for Ranking Variety of Colors 67

Table 3.2 Computer detection of number of used colors and length of edges of Rank-1, Rank-13, Rank-26, Rank-39, Rank-52 rated by Rater-1, Rater-2, and the computer __________________________________________________________________________________________________





Rater-1 _________________________

Rater-2 ________________________

Computer _________________________

Code of Number Length Code of Number Length Code of Number Length drawing of used of edges drawing of used of edges drawing of used of edges colors colors colors __________________________________________________________________________________________________ Rank-1 J 25 14,181 J 25 14,181 c 29 10,895 Rank-13 f 20 12,626 O 21 13,221 d 22 12,954 Rank-26 b 23 16,847 y 20 12,304 y 20 12,304 Rank-39 Z 19 10,982 j 13 8,577 s 17 15,274 Rank-52 N 11 10,607 N 11 10,607 I 11 5,972 __________________________________________________________________________________________________

Next, we analyzed the correlation between the ranks assigned by the computer and by each rater. The scatter plots in Figure 3.4 show the correlations between the ranks given by the computer and those by Rater-1 and by Rater-2, respectively. In the left scatter plot, drawing R, c, b, U, d, and W are examples ranked higher by the computer than by Rater-1, and drawing o, Q, and g are the examples ranked in the reverse order. Drawing R in the left scatter plot is an example with a large difference in ranking, ranked Rank-16 by Rater-1 and Rank-2 by the computer. In Figure 3.5, we compare drawing R with drawing m of Rank-16 by the computer, and k of Rank-2 by Rater-1 in Figure 3.5. After careful examination, we noticed that the computer’s rating is more appropriate in some cases, while the opposite is true in other cases. In summary, the RCC between the computer and Rater-1 was found to be higher than the one between the computer and Rater-2. Recall we see in Table 3.1 that for Rater-1, Rs = 0.7592, and for Rater-2, Rs = 0.6621.

Figure 3.4 Scatter plots of the ranks between the computer and Rater-1 (Left) and between the computer and Rater-2 (Right).

68

PART ONE ART EVALUATION

Figure 3.5 Comparison of drawings ranked Rank-2 and Rank-16 by Rater-1 and drawings of the same ranks ranked by the computer.

The correlation of the computer evaluation is relatively high with both raters. We may also conclude that high correlations exist between the computer’s evaluation and those by two raters as indicated by the values of test statistics, Rater-1: z = 5.4217 (p-value < 0.0001), Rater-2: z = 4.7283 (p-value < 0.0001). It further confirms the validity of the computer rating. The high reliability between the human raters on the one hand and the high correlations between the human raters and computer on the other hand validate the usability and usefulness of the computer evaluation system.

Chapter 3

A Computer System for Ranking Variety of Colors 69

3.4 Discussion Relatively large differences in the ranks assigned by the two raters can be seen in the cases of drawings C, s, Q, P, and v. We may reduce these differences in the evaluations by eliciting the reasons for such a gap and incorporating them in the knowledge base of an expert system. Defining the variety of colors in more specific terms is essential. We see relatively large differences in ratings between Rater-1 and the computer for drawings W, d, b, U, R, and c. Consideration of brightness V in the HVC color space and the list of colors with the number of clusters may reduce the differences in the ranks given by the raters and by the computer. As an alternative method to rate the variety of color, a regression analysis (Kutner, Nachtsheim, Neter, & Li, 2005) can be applied where the dependent variable represents the ranks assigned by the raters and independent variables represent the 12 basic elements in the C_CREATES. The drawing samples were collected from children with no known history of psychological and emotional disorders. The drawings were done in school art classes, not during art therapy sessions. If the samples were collected from those with some kind of psychological and emotional disorder, or if the samples had been a series of drawings by the same person, higher reliability and validity could be expected. The children used their own 12, 18, 24, or 36 color crayon sets, with the 18-color being the most common palette. In this study, we only examined the variety of colors in the sample, not the influence of the number of used colors of the drawing medium on the variety of colors. Also, we note that the samples were collected from classes of typical third, fourth, and fifth grade students. We expect that the same conclusion may be found in the drawings of children of other age groups with respect to the effectiveness of the computer evaluation system. However, this can be the subject of further research. In conclusion, although the human raters acknowledged the difficulty in the rating task, they showed relatively high reliability in their evaluation of the variety of colors. The high correlations between the human raters and the computer validate the computer evaluation system as an objective tool to aid art therapists.

Judgment of Main Color Using a Computer Algorithm

Chapter 4 Keynote Color in drawings is one of the most important elements in art evaluation. The number and list of used colors, the area of each color, and the blend of colors are commonly considered in the evaluation of drawings. The main color in a drawing is also a major concern in the evaluation and may be defined as the most important element that expresses the theme of the drawing. We have developed a computer system that can automatically judge the main color of a given drawing. The proposed algorithm focuses on distinguishing the main color from the background color which commonly covers the largest area in the drawings with a wide background.

4.1 Introduction Definition The main color may be defined as the most important one that expresses the and importance drawing’s theme. Sometimes the main color in a drawing is more important than the list of colors. How people react to and use color can provide important information regarding their current psychological or emotional state (Lev-Wiesel & Daphna-Tekoha, 2000). Colors do not, of course, hold the same meaning for everyone, and there are cultural and individual differences. However, it is reported that emotions commonly attributed to red are violence, passion, aggression, and anger; yellow hospitality; and brown timidity (Hammer, 1953, 1969; Klepsch & Logie, 1982; Precker, 1950). In any case, the main color may provide useful information about the drawer. Yet there has been no evaluation system that considers the main color as an element to be evaluated. This chapter delineates the development of a computer system that judges the main color in a drawing by applying the methods available in the field of digital image processing and by devising a computer algorithm. 70

Chapter 4 Judgment of Main Color Using a Computer Algorithm

71

Art therapists are often compelled to rely on their professional observation Difficulty and judgment in evaluating elements in drawings, which is subject to human bias. This implies that their evaluations of colors may also be inconsistent. Consequently, judging the main color may be subjective and the outcome may differ depending on the raters. As will be demonstrated in one of the examples in this chapter, some raters may choose red as the main color of a given drawing sample, while others choose green for the same drawing. Also, in some cases, it is unclear as to whether a certain color is a main color, a principal color, or a background color. Usually, background colors are not considered in interpreting drawings and, thus, sometimes need to be excluded from the list of colors. New technologies can overcome these difficulties in subjectivity, inconsistency, and uncertainty through the quantification and objectification process. In an effort to apply computer graphics to art therapy, Hartwich and Brandecker (1997) directed patients to draw using a tool such as Adobe Photoshop (Adobe Creative Team, 2004) and examined their responses. This attempt shows, although on a limited level, the innovative use of computer technology applications in art therapy. As explained in Chapter 1, Kim, Bae, and Lee (2007) applied the technologies available in the field of digital image processing to developing a computer system that automatically evaluates the number of used colors and the list of colors, as well as the area of each color, the number of clusters, the length of edges, etc. However, this system can only perform the automatic and mechanical analysis of colors in drawings, and is not capable of reaching judgments of a main color as human raters do. The computer system developed in this chapter judges main color in a A solution drawing. First, the preliminary procedure, consisting of blurring, clustering, color recognition and classification, and edge detection, is applied using the method for basic elements in the C_CREATES. Next, the main color is determined by applying a simple algorithm based on the characteristics of main color represented in the numbers of pixels and clusters of each color. Here, the cluster refers to a contiguous area colored with the same color, not separated by different colors. These methods allow us to deduce other useful information, such as the principal color and background color as well as the details of drawing, the variety of colors, and the concentration, effort and sincerity applied by the drawer. The process is illustrated through case studies and the usability and usefulness of the system are verified through the analysis of 50 crayon drawing samples. The computer process can reduce the subjectivity in human experts’ judgments and increase consistency. Furthermore, since this system is capable of analyzing hundreds of drawings in a short period of time, it is expected to alleviate the present shortage of art therapy experts. Moreover, the system can process, store, and provide a large volume of fundamental and useful information accurately and quickly.

72

Figure 4.1 Preliminary procedure in two drawing samples.

PART ONE ART EVALUATION

Chapter 4 Judgment of Main Color Using a Computer Algorithm

4.2 Procedure of main color judgment We use the same method proposed in Chapter 1. Blurring is a method of removing noise not intended by the drawer. The clustering together of similar colors can also be used to remove noise. We apply the method of Ye et al. (2003), which clusters color images using color features and spatial connectivity. As explained in Chapter 1, the number of used colors that provide useful information in art therapy assessments does not exceed 10, even when classified in detail. It is therefore acceptable and sufficient for art evaluation to classify the main color into one of the 15 colors adopted by the KIS, A15. The set of 15 colors is {red (R), orange (O), yellow (Y), yellow-green (YG), green (G), blue-green (BG), blue (B), blue-purple (BP), purple (P), red-purple (RP), pink (PI), brown (BR), white (W), grey (GR), black (BL)}. After the color of each pixel is classified into the one closest to the colors in A15, we count the numbers of clusters and pixels for each color. 4.2.1 Case examples In Figure 4.1, we illustrate the preliminary procedure by examining two crayon drawings made by two children in fourth grade, which are the same as the ones used in Chapter 1. Although the main color usually occupies a large area, it is not simply the color with the largest area. The area of each color here, is determined by the number of pixels (P) colored with it. When the drawing has a wide background, which is common in drawings, the color with the largest area is often the background color. Now, we need to depict the characteristics of the main color. We use the number of clusters (C) featured in the main or background color. We anticipate that the background color will have a large P but a small C, and that the main color will have a large P as well as a large C. To reflect this characteristic, we limit the candidates of main color to those with large values of %P times %C, where %P and %C, respectively, denote P and C percentages of the total pixel numbers in a drawing. In Table 4.1, we show the results of the two drawing samples in Figure 4.1 with a resolution of 480 x 640 = 307,200 pixels. In drawing A, green is selected as the most likely main color, with blue and yellow being the next most likely main colors in descending order of probability. Of the five raters, three selected yellow, and two green. In drawing B, blue is selected as the most likely main color, followed by grey, and then white, while all five raters selected grey. 4.2.2 Computer algorithms The criterion based on both P and C can still cause a background color to

73

74

PART ONE ART EVALUATION

Table 4.1 Measurement of color-related elements and judgment of main color in the two drawing samples

__________________________________________________________________________________________________ Drawing A Drawing B Color __________________________________________ __________________________________________ P %P C %C %P Rank a b P %P C %C %P Rank a b x%C x%C __________________________________________________________________________________________________ Red 2,852 0.8 1 4.3 4 1,857 0.6 3 5.4 3 Orange 5,256 1.8 2 8.7 14 207 0 3 5.4 0 Yellow 21,830 7.1 5 21.7 150 3 √ 3 6,205 2.0 10 17.9 30 Y-G 158 0 2 8.7 0 2,738 0.9 6 10.7 6 Green 143,371 46.0 2 8.7 404 1 2 0 0 0 0 0 G-B 345 0 1 4.3 0 2,992 1.0 2 3.6 2 Blue 130,267 42.4 2 8.7 368 2 245,052 79.8 8 14.3 1,136 1 Violet 924 0.3 2 8.7 2 0 0 0 0 0 Purple 0 0 0 0 0 138 0 1 1.8 0 R-P 0 0 0 0 0 0 0 0 0 0 Pink 2,206 0.7 2 8.7 6 0 0 0 0 0 Brown 155 0 1 4.3 0 0 0 0 0 0 White 0 0 0 0 0 7,426 2.4 14 26.0 56 3 Grey 216 0 2 8.7 0 40,342 13.1 8 14.3 184 2 √ 5 Black 620 0.2 1 4.3 0 243 0 1 1.8 0 __________________________________________________________________________________________________ Total 307,200 23 307,200 56 __________________________________________________________________________________________________ a b

system determination, checked number of human raters determination

be mistakenly selected as the main color. The criterion selected the background colors, green and blue, as the main colors in drawings A and B, respectively. To remedy this problem, we devise a simple algorithm with the following three steps. Algorithm Step 1: Limit the candidates of main color to three determined by the above criterion. Step 2: Remove colors with clusters that have a relatively long edge on the paper boundary. Step 3: Select the color with maximum %Pi x %Ci. Figure 4.2 presents three samples whose main colors were determined to be the background colors by the criterion, and shows how these errors can be corrected by the 3-step algorithm above. The proposed algorithm selected grey correctly as the main color in drawing A. Also, the algorithm selected yellow and red correctly in drawing B and C, respectively. The algorithm, although the steps are not presented here, selected yellow correctly in drawing A in Figure 4.1.

Chapter 4 Judgment of Main Color Using a Computer Algorithm

75

Figure 4.2 Three steps of algorithm in removing background color.

76

PART ONE ART EVALUATION

4.3 System verification 4.3.1 Inter-rater reliability The same set of 50 drawing samples by children in third, fourth, and fifth grade used in Chapter 3, were used for the purpose of system verification. Five raters with sound color perception identified the main color for each drawing: one art therapist, one color therapist, one elementary school teacher, and two graduate students. The system determined the main color, by applying the procedure and the algorithm. The system was coded in Microsoft Visual C++. Each drawing required 35 seconds on average to be input by a scanner, an HP Scanjet 4470c with resolution of 1,200 x 2,400 dpi, and 42 seconds on average to be analyzed by an IBM PC, Pentium 4 with CPU of 2.4 GHz, memory of DDR 512 MB and video memory of DDR 64 MB. The camera’s digital images can be processed instead of the digital images of a scanner. The specifications of the scanner and PC are old ones, and thus modern specifications will reduce the time quite a bit. The value of Fleiss’ kappa (Fleiss, 1971) for the overall inter-rater reliability of five raters was obtained as κF = 0.547, which indicates a moderate degree of agreement among the raters, but still a significant number of disagreements. This statistic supports our claim that the main color judgment is prone to human subjectivity. However, among 50 drawings, the main colors were identified in a total of 47 drawings (94 %) by the majority selection with at least 3 raters: all five raters coincided in their decision on main colors in 16 drawings, four raters in 14 drawings, and three raters in 17 drawings. Figure 4.3A is one of the three drawings in which majority decisions were not reached. For this drawing, two raters chose red as the main color, two raters chose blue, and one rater did not identify any main color. The system determined blue as main color. Figure 4.3B is another similar example. For this drawing, two raters determined red as the main color, two raters yellow, and one rater brown. The system determined brown as the main color. These cases show that determination of a main color is a delicate matter. Out of the 47 drawings, six drawings were deemed to have no main color by the majority of the raters. Thus, we excluded these six drawings and only took 41 of the above 47 drawings into consideration. In three of these drawings, the system could not correctly identify the main color. In Figure 4.3C, the noise with a brown tone in the middle horizontal stripes was misclassified as a yellow color. However, this kind of misclassification could be solved by using a more detailed classification of 157 or 47 colors adopted by KIS, instead of 15 colors. In Figure 4.3D, the edge between dark blue and bright blue was not detected. This problem can also be solved by using more detailed color classifications, although this would yield complicated edges and require more computer time for calculation.

Chapter 4 Judgment of Main Color Using a Computer Algorithm

77

Figure 4.3 Analysis of system.

78

PART ONE ART EVALUATION

Figure 4.4 Other useful information: The number of clusters and the area of each cluster.

Chapter 4 Judgment of Main Color Using a Computer Algorithm

79

Table 4.2 Other useful information: The number of clusters and the area of each cluster

__________________________________________________________________________________________________

Number Color (Number of clusters, Area) _______________________ _____________________________________________________ Drawing Color Cluster Edge Area Main color nd Largest 2 largest System Humana __________________________________________________________________________________________________ 4.1A 15 23 4,813 G(2, 46.0%) B(2, 42.4%) Y(5, 6.8%) Y3 4.1B 16 56 7,422 B(8, 79.8%) GR(8, 13.1%) GR GR5 __________________________________________________________________________________________________ 4.2B 18 49 7,353 B(1, 51.8%) BR(3, 21.3%) Y(2,17.0%) Y5 4.2C 23 94 9,339 Y(2, 75.2%) W(13, 10.2%) R(9, 4.4%) R3 __________________________________________________________________________________________________ 4.3A 22 67 8,833 B(7, 68.3%) R(8, 14.5%) B † 4.3B 20 61 7,197 P(1, 83.4%) BR(15, 5.1%) BR † 4.3C 18 29 6,361 B(1, 43.8%) YG(2, 41.9%) R(4, 2.7%) R3 4.3D 13 38 6,208 B(1, 43.8%) G(1, 10.0%) R(1, 2.0%) B5 4.3E 21 216 17,722 YG(11, 30.1%) Y(45, 21.5%) YG O3 __________________________________________________________________________________________________ 4.4A 13 44 7,126 B(6, 69.7%) GR(5, 18.4%) BR(6,10.2%) BR5 4.4B 28 174 14,958 Y(9, 29.7%) B(14, 17.7%) BL(41, 6.4%) BL3 4.4C 15 48 6,113 YG(1, 80.2%) BL(7, 6.5%) BL ‡ 4.4D 24 122 13,549 Y(5, 23.3%) B(14, 22.2%) B B4 4.4E 10 14 11,465 Y(11, 44.9%) B(7, 25.1%) B ‡ __________________________________________________________________________________________________ number of human raters no majority decision ‡ no special main color a

†

4.3.2 System validity We have then considered the remaining 38 drawings excluding the three drawings above. The detection of main color by the system coincided with the raters’ decisions in 32 drawings. In six cases in which the system and the human raters made different decisions, the system erroneously determined the background color as the main color in four drawings, even with the application of the algorithm. Figure 4.3E shows one of the six cases. It must be noted, however, that the same kind of confusion between the background color and the main color happened in 16 drawings when we did not apply the algorithm. These results show that the algorithm corrected errors in 12 drawings. In two cases, the system still made errors, despite the use of the 157 color classification. In summary, the value of Cohen’s kappa between the raters-system reliability in judging the main color of the 41 drawings above was obtained as κC = 0.729, which indicates a good degree of agreement between the majority of the human raters and the system. The system’s decisions coincided, or could be made to coincide, with those of the human raters in 38 drawings out of

80

PART ONE ART EVALUATION

the 44, excluding the 6 drawings that were deemed to have no main color from the total of 50 samples, yielding a consistency rate as high as 84.1%. Thus, we conclude that the system can be used as an effective method for automatic identification of the main color. Other useful We illustrate other useful information that can be obtained through the study information of more drawing samples. In Figure 4.4, we present the same drawings as those used Chapter 2, which will be useful in comparing other features of the analysis. The detailed statistics of all drawing samples are given in Table 4.2. The number of clusters and the area of each color allow the system to obtain information on such diverse matters as the relative importance of each color, the drawer’s taste, etc. In many cases, the number of clusters for the color with the second largest area is not smaller than the one with the largest area. The exceptions are Figure 4.4A, and Figure 4.4E. Among the 50 drawings, 37 drawings show this tendency. Also, in some cases, the main color is different from the colors with the largest and the second largest areas, as in the case of the drawings in Figure 4.1A, Figure 4.2B, Figure 4.2C, Figure 4.3C, Figure 4.3D, Figure 4.4A, and Figure 4.4B. Out of the 50 drawings, the system sees this pattern in 28 drawings, whereas the human raters see it in 26 drawings.

4.4 Discussion We have developed a computer system that can automatically identify the main color in a drawing. This system can quickly, easily, and accurately analyze hundreds of drawings and produce useful and fundamental information for art therapy. The application of this system will enable human art therapists to provide more detailed evaluation. The specific numbers provided by the system are especially helpful in indecisive cases. In this chapter, we did not touch upon the projective relationships between the main color and psychological states, but the results from this study can be utilized to develop an expert system for such psychological diagnostic and therapeutic purposes.

Determination of Placement Using Digital Image Processing

Chapter 5 Key points Placement of subject in a drawing is an important element in art evaluation. In this chapter, placement is classified into one of ten categories, one usual placement and nine unusual ones. Like other elements of drawings, such as color, theme, line, shape, and structural organization, determining placement is not immune to human subjectivity. This chapter explores the development of a computer system that determines placement category automatically and objectively by implementing digital image processing for color recognition and edge detection. This system divides the entire drawing sheet into several regions and considers the distribution of edge pixels in each region as the criteria for determining placement categories. The information is also useful for the determination of space usage and details of drawing. The system has been verified through drawing samples.

5.1 The element of placement in art therapy tools Along with color, theme, line, shape, and structural organization, placement of subject in a drawing is an important element of art evaluation. In this chapter, we classify placement into a total of ten categories with one considered as a usual placement and the rest as unusual. A usual placement refers to an image arranged in a balanced manner over a drawing sheet, and unusual placements refer to images that are predominantly allocated to certain regions. The unusual placements are classified into nine categories based on the region where the subject occupies predominantly. Each of the nine categories may reveal information about the psychological or emotional state, or social status of the drawer. The rating system of the DDS specifies the unusual placement as one of its 23 elements to be rated (Cohen, 1986/1994). In the DDS, placement is determined by regions, which are vertical and horizontal divisions of a drawing sheet. According to the DDS, three categories of unusual placements, upper (U), left (L), and right (R) 81

82

PART ONE ART EVALUATION

are defined as an image drawn predominantly above the center of the sheet (horizontal axis) or drawn mostly on the right or left of the vertical axis, especially when the remainder of the sheet is left blank. Interpretation Interpreting drawer’s psychological, emotional, or social states based on based on placement their placements of subjects should be performed in a very cautious manner since a single drawing element may not provide substantial evidence of the drawer’s characters (Cohen & Mills, 1994). The same placement may have different meanings depending on the drawer’s nationality, personal, cultural, social, and educational experience. However, a single sign can raise questions that must be addressed (Cohen & Mills, 1994). Several studies have reported that unusual placements of subjects may reveal psychological, emotional, and social states of the drawers. Lev-Wiesel and Drori (2000) have analyzed the DAP drawings by elderly widows and wives based on projective hypotheses. The results of the analysis show that widows usually have fears about their physical safety, being alone, darkness, victimization, and feelings of being unwelcome in the world of couples (Hammer & Piotrowski, 1997; Lopata, 1976, 1996). The sizes of subject figures drawn by widows tend to be smaller and placements of the heads tend to be lower than the ones in wives’ drawings. A subject placed in the center of the drawing sheet may indicate an appropriate control of the environment, while the upper-center may reflect an optimism about goals (Wanderer, 1997; Machover, 1949), haughtiness, or a sense superiority over others (Buck, 1964). The bottom-center implies a sense of insecurity, inadequacy, and depression (Machover, 1949). We note that in Swenson (1957), Machover’s hypotheses on the element, placement, was not supported by the later studies, and that it was not “generally” supported in Roback (1968). Swenson (1968) reported nine positive studies supporting the hypotheses and six negatives which do not, and concludes that such findings should be accepted with caution. As a rule, white patients’ drawings typically demonstrate normal sizes (about 18 cm) with the subject placed in the upper-center of the sheet, while black patients’ drawings tend to be smaller in size with the subject placed in the bottom-center (Adler, 1970, 1971). The bottom category reflects low self-esteem, resulting from lowly-valued social roles (Lorge, Tuckman, & Dunn, 1958). Difficulty In determining a placement category, art therapists are often expected to rely on their subjective observation, professional judgment, and individual experience, which may result in different evaluation outcomes. For instance, the placement category in one drawing could be classified as a usual one by one rater and as bottom-left by another, as shown in the samples of this study. A solution These difficulties can be overcome by utilizing the digital image process-

Chapter 5 Determination of Placement Using Digital Image Processing

83

ing technology, which improves the quantification and objectification of the evaluation. This chapter explores the development of a computer system which determines placement category. The system uses the same method that was used in Chapter 1 for color analysis and edge detection through blurring, clustering, and transforming colors into standard colors. Placement category is determined on the basis of the distribution of edges pixels after the edges have been detected through color transformations. Here, the edge consists of pixels of which the color is different from that of its neighboring pixels. The placement category is determined depending on the region where length of edges (number of edge pixels) is greater than the critical (threshold) value that has been established as a criterion. We determine placement categories based on the areas of regions covered and length of edges in each region, and establish the criteria. We will illustrate the evaluation process through 50 drawing samples. This system provides numerical information on placement categories and other elements such as space usage and details of the drawing, which will be evaluated in Chapter 7 and Chapter 6, respectively. We validate the usability and utility of the system by comparing the system’s determination of placement categories of the given drawing samples with those of human experts.

5.2 Methods for edge detection and definition of placement category 5.2.1 Methods for edge detection First, we detect edges in a drawing by using the same methods that were used for color analysis in Chapter 1. For the convenience of the reader, we will explain the steps of the methods again. We use a 15-color set A15 adopted by the KIS, and A47, which consists of 47 standard colors, which enables more detailed classifications than A15. In order to measure the closeness between two colors, we use the National Bureau of Standards (NBS) distance (Bartkowiak & Domanski, 1995). A sheet of paper is divided into pixels, each of which is the final element to be analyzed. If the vertical side is divided into 480 points and the horizontal side into 640, the paper consists of a total of 480 x 640 = 307,200 pixels. Before transforming the colors, we apply a 3 x 3 mask median filter (Gonzalez & Woods, 2002) and the clustering method of Ye et al. (2003) in order to reduce the noise, which is unintended ‘touch’ or ‘non-touch’ of drawing materials (Gonzalez & Woods, 2002). Noise occurs when the thickness of the tips of crayons causes the drawer to color in unintended space or to keep them from filling in the intended space. Some areas are untouched and remain white, while others only slightly touched with very light color. Figure 5.1 presents the results of the procedures above applied to two cray- Two samples

84

PART ONE ART EVALUATION

Figure 5.1 Color transformations and edge detections in two drawing samples.

on drawings by two fourth graders, which are the same samples used in Chapter 1. They represent typical cases of a drawing with extremely weak noise and a drawing with strong noise. The transformations of these drawing samples into A47 and A15 and the edge detections are shown in Figure

Chapter 5 Determination of Placement Using Digital Image Processing

85

Figure 5.2 Illustration of regions classified as unusual placement categories.

5.1. We observe that applying the process of burring, clustering, and color transformation to the samples results in successful edge detection, making them suitable for analysis. 5.2.2 Definitions of placement categories We divide a drawing sheet into nine regions: upper (U), bottom (B), left (L), right (R), upper-left (UL), upper-right (UR), bottom-left (BL), bottom-right (BR), and center (C). The drawing sheet is divided based on 2/3 of horizontal and vertical lines as illustrated in landscape drawings in Figure 5.2. A similar division method is applied to portrait drawings. Note that the region right in Figure 5.2 occupies 2/3 (67%) of the paper. When the number of edge pixels in the right region is over 81% (9/10 x 9/10) of the total number of pixels, we define it as an unusual placement category of left (L). Unusual placement categories of right, upper and bottom are defined in the same manner. Note that the region upper-left in Figure 5.2 occupies 4/9 (44%) of the paper. When length of edges (number of edge pixels) in the upper-left region is over 67 % (2/3) of the total length of edges of a drawing, we define it as an unusual placement category of upper-right (UR). Unusual placement categories of upper-left, bottom-left, bottom-right and center are defined in the same way. When the result does not belong to any of the above nine categories, it is classified as a usual placement. We note that dividing a drawing sheet into regions and setting critical

86

PART ONE ART EVALUATION

Table 5.1 Comparison of the results of placement category determination of the two drawing samples between A47 and A15 __________________________________________________________________________________________________

Drawing Figure 5.1A Drawing Figure 5.1B ____________________________________ _______________________________________ Unusual placement A47 A15 A47 A15 _______________ ___________________ ________________ __________________ type # pixels (%) Sa # pixels (%) Sa Hb # pixels (%) Sa # pixels (%) Sa Hb __________________________________________________________________________________________________ Upper 3,045 (58.2) 2,795 (58.1) Bottom 4,557 (87.1) √ 4,312 (89.6) √ 1 Left 2,300 (43.9) 2,183 (45.4) Right 4,705 (90.0) √ 4,295 (89.2) √ Upper-Left 2,100 (40.1) 1,245 (25.7) Upper-Right 3,725 (71.2) √ 2,496 (51.9) Bottom-Left 3,118 (59.6) 2,115 (43.9) Bottom-Right 4,506 (86.1) √ 3,843 (79.8) √ 3 Center 3,447 (65.9) 2,105 (43.7) 1 __________________________________________________

10,057 (81.8) √ 6,517 (87.8) √ 4 8,041 (65.4) 4,376 (59.0) 7,413 (60.3) 4,077 (54.9) 1 7,884 (64.1) 4,983 (67.1) 7,800 (63.4) 3,589 (48.4) 8,317 (67.7) √ 4,363 (58.8) 8,050 (65.5) 2,078 (28.0) 8,176 (66.5) 3,171 (42.7) 7,883 (64.1) 3,085 (41.6) ______________________________________

Total 5,234 ( 100) 4,813 ( 100) 12,295 ( 100) 7,422 ( 100) __________________________________________________________________________________________________ a b

Checked on unusual categories Number of human raters

values of pixel numbers may appear somewhat arbitrary. Type I and Type II errors are considered in determining the critical value. Here, Type I error refers to the probability of the system mistakenly classifying a placement into an unusual category, when it is, in fact, a usual one. Type II error refers to the probability of the system mistaking a usual placement for an unusual one. For a given region, the higher the critical value, the lower the Type I error, and the higher the Type II error, in turn. In case of a low critical value, the results are reversed. Here, the critical values of 81% (9/10 x 9/10) and 67% (2/3) were obtained by simulation with Type I error of approximately 0.05. As the system accumulates more drawing samples and the corresponding human expert judgments, more appropriate divisions and critical values can be obtained.

5.3 Determination of placement category 5.3.1 Information on placement category We explain the procedures of determining placement categories through the evaluations of the two drawing samples above. Length of edges provided by color transformations into A47 and A15 are presented in Table 5.1. For drawing Figure 5.1A, color transformation into A47 classified its placement categories as bottom (B), right (R), upper-right (UR), and bottom-right (BR), while A15 did as bottom (B), right (R), and bottom-right (BR). Among five

Chapter 5 Determination of Placement Using Digital Image Processing

human raters, three identified it as bottom-right (BR), one as center (C), and the other as bottom (B). For drawing Figure 5.1B, color transformation into A47 identified its placement categories as upper (U) and upper-right (UR), while A15 did as upper (U). Out of the five human raters, four judged it as upper (U), and one as left (L). Color transformation into A47 appeared to be more accurate in general. In other words, more computation generally yielded more accurate results. We note that if the critical values for unusual categories, UL, UR, BL, BR, and C were 75% of the total length of edges rather than 67 %, the decisions after A47 and A15 transformations yield same determination of unusual categories in both drawings. 5.3.2 Information on other elements of drawings It is obvious that space usage, an element of in the DAPA and the FEATS can be easily evaluated by the distribution of edge pixels. Also, area of each color can give useful information on space usage when the background color is identified and then removed. Wadeson (1980) has studied details of drawings of hospital patients suffering from depression. We can determine the level of details of a drawing by obtaining numbers of clusters and length of edges. For drawings Figure 5.1A and Figure 5.1B, numbers of clusters are 23 and 56, and lengths of edges in the number of pixels are 4,813 and 7,422, respectively. These measures enable easy and objective comparisons of details of the samples.

5.4 System verification We analyzed 50 drawing samples by third, fourth, and fifth grade children for system verification. The colors are transformed into A15 and the edges are detected for each drawing. Five human raters with sound color perception and a sense of balance, one art therapist, one color therapist, one elementary school teacher, and two graduate students, classified the samples into the given placement categories. The system is coded in Microsoft Visual C++. The required specifications of computer system are identical to the ones used for main color judgment in the previous chapter. 5.4.1 Sample examples Figure 5.3 illustrates the category determination process by the system. Figure 5.4 presents drawing samples to illustrate certain features of the system evaluation. In Figure 5.4A, some edges are not detected by color transformation into A15, because it cannot differentiate dark blue from bright blue. However, in Figure 5.4B, the edges are detected by color transformation into A47. As shown in Figure 5.4C, there is inconsistency among human raters with two raters determining placement as upper (U), two as center (C), and one as left (L). The system determined it as upper (U). In this case,

87

88

PART ONE ART EVALUATION

Figure 5.3 Category determination process by the system.

Chapter 5 Determination of Placement Using Digital Image Processing

89

Figure 5.4 Illustration of system features.

90 Table 5.2 Comparison of placement categories of 11 drawing samples determined by the system and five human raters

PART ONE ART EVALUATION

________________________________________________________________________ Figure

Number of edge pixels of placement category (%) ____________________________________________________________

U B L R UL UR BL BR C Usual ________________________________________________________________________

5.1A 58 90 45 89 26 52 44 80 44 Systema √ √ √ Humanb 1 3 1 5.1B 88 59 55 67 48 59 28 43 42 Systema √ Humanb 4 1 ________________________________________________________________________ 5.3A 58 7 63 74 34 50 47 53 52 Systema √ Humanb 5 5.3B 82 64 70 62 58 51 42 40 51 Systema √ Humanb 5 5.3C 50 82 73 69 35 49 41 38 45 Systema √ Humanb 5 5.3D 68 69 76 66 53 67 46 52 58 Systema √ Humanb 3 1 1 5.3E 71 46 71 74 47 54 39 37 49 Systema √ Humanb 1 4 ________________________________________________________________________ 5.4A 83 48 76 65 67 52 36 31 48 Systema √ √ Humanb 1 4 5.4C 86 65 68 67 59 57 49 4 48 Systema √ Humanb 2 1 2 5.4D 68 74 72 72 52 51 52 51 55 Systema √ Humanb 4 1 5.4E 70 59 71 60 50 43 44 34 42 Systema √ Humanb 5 ________________________________________________________________________ a b

Checks on significant unusual placement categories Number of raters

the system’s judgment proves to be more rational. It can be assumed that the human raters may have been affected by fatigue after evaluating 50 drawing samples in a row. Figure 5.4D and Figure 5.4E present a typical case of a drawing with simple edges and a drawing with complex edges, respectively, based on which details can be rated. Table 5.2 shows the comparison of the placement categories of 11 draw-

Chapter 5 Determination of Placement Using Digital Image Processing

ing samples illustrated in this chapter determined by the system and the five human raters. 5.4.2 Inter-rater reliability The value of Fleiss’ kappa (Fleiss, 1971) for the overall inter-rater reliability of five raters was κF = 0.447, which indicates a moderate degree of agreement among the raters, but still a significant number of disagreements. This statistic supports our claim that placement category judgments are prone to human subjectivity. However, 47 drawings (94 %) resulted in majority judgments of three or more raters, with 30 of them classified as the usual category, demonstrating the robustness of the majority judgments. All five human raters unanimously agreed upon placement categories for 16 drawings, four raters for 17 drawings, and three raters for 14 drawings. 5.4.3 System validity For the 47 drawings that obtained majority judgments by three or more raters, the value of Cohen’s kappa between the raters-system reliability was κC = 0.787, which indicates a good degree of agreement between the majority decision and the system. In summary, among the 44 drawings, excluding two drawings that had inappropriate transformations of colors and one drawing with no edge detection, the judgments of the system coincided with those of human raters regarding 41 drawings, achieving a consistency rate of 93 %. When we applied the A47 transformation, the accuracy increased to 98%. In conclusion, the system’s judgment was proven to be more reasonable as it provides accurate and quantitative information that is useful for determining placement categories. Thus, the proposed system can be used with high reliability in art evaluation. 5.4.4 Other useful information In the 50 samples, lengths of edges in numbers of pixels are on average of 9,890 with a standard deviation of 2,685. Figure 5.4D has 7,126 edge pixels corresponding to a standard deviation of -1.03, a two sided p-value of 0.36 with relatively simple edges, whereas Figure 5.4.E has 13,549 edge pixels corresponding to a standard deviation of 1.36, a two sided p-value of 0.11 with relatively complex edges. Thus, the level of details of Figure 5.4D is rated very high, whereas that of Figure 5.4E is very low. Figure 5.4E has the largest numbers in both number of clusters and length of edges, 216 and 17,722, respectively, whereas Figure 5.1A has the smallest, 23 and 4,813 respectively. From this quantitative information, we are able to deduce that numbers of clusters and length of edges are closely related. The analysis of all 50 drawings shows the same tendency. Number

91

92

PART ONE ART EVALUATION

of used colors, number of clusters, and length of edges provide various information that allows us to measure details, variety, or complexity of each drawing and estimate the amount of effort made by the drawer.

5.5 Discussion We defined unusual placement categories more concretely. A region refers to the larger side of 2/3 of the horizontal and vertical axes instead of 1/2 in the DDS. The minimum number of pixels therein has also been redefined from ‘the image drawn predominantly,’ ‘most of the image,’ or ‘blank’ as defined by the DDS. For the areas occupying 67 % (2/3) and 44 % (4/9) of the whole sheet, the minimum numbers of pixels in the areas were 81 % and 67 % of the total, respectively. These criteria can be modified in consideration of Type I and Type II errors.

Grading and Ranking Prominence of Color and Details of Drawing Using Regression Models

Chapter 6 Key point We have developed four computer systems using regression models that grade and rank the two elements in the FEATS, prominence of color and details of drawing. We demonstrate how each regression model is developed to evaluate in grade and rank an element (dependent variable) from the other elements (independent variables) in the C_CREATES, for which the evaluation methods have been developed in the previous chapters.

6.1 Regression analysis The prominence of color (or simply prominence) and the details of ob- Definitions jects and environment (or simply details) are two of the 14 elements in the FEATS (Gantt and Tabone, 1998). The prominence of color measures how much color a person uses in the entire drawing. The details quantifies the relative amount of detail in the drawings. In this chapter, we apply regression models to evaluate these two elements in grade and rank. Regression analysis is a statistical tool widely used to predict the values of one variable from the values of the others. We have developed two regression models for each element, one of which is for evaluating in grade and the other for evaluating in rank. We here provide a brief explanation of regression analysis. The readers who wish to know more about the regression analysis are advised to consult textbook such as Applied linear statistical models (Kutner et al., 2005). In addition, any statistical software packages including the SPSS and the SAS can be used for practical applications. Regression model can be used not only for art evaluation, but also for art interpretation. In our explanation of regression analysis below, corresponding terms used in art therapy to terms such as “predict,” “dependent variable,” and “independent variables” used in statistics are mentioned in 93

94

PART ONE ART EVALUATION

parentheses. A regression model is a statistical tool widely used to predict (evaluate, estimate) the values of one dependent variable (element) (space usage, degree of dementia) from the values of the independent others (elements in art therapy assessments) (Kutner et al., 2005). Once all of the dependent and independent variables (elements) have been measured and expressed quantitatively, the relationship can be modeled into a regression analysis. Regression model Denoting the dependent variable that is to be predicted by Y, and m independent variables that predict Y by X1, X2, ... , Xm, from n samples. The linear regression model is formulated as Yi = a0 + a1 Xi1 + a2 Xi2 + ... + am Xim + εi, i = 1, 2, ... , n, where a0, a1, ... , am are coefficients estimated by samples, and εi is assumed to be an identically distributed independent normal random variable with mean 0 and constant variance σ2. For example, we consider nine (m = 9) elements of drawings as potential independent variables to explain the grade of prominence of color as a dependent variable from 135 (n = 135) sample grades determined by the three raters for 45 PPAT drawings in the FEATS, as we will see in section 6.4. Stepwise regression Stepwise regression is a method to select some important variables that “well” explain the dependent variable. For example, three elements are selected from the nine potential variables by stepwise regression for the same case in section 6.4. Standardized To compare the relative effect (importance) of each independent variable on regression the dependent variable, we also formulate a standardized regression model Yi’ = a1’ Xi1’ + a2’ Xi2’ + ... + am’ Xim’ + εi, i = 1, 2, ... , n with standardized variables, Yi’ = {Yi - E(Y)} / S(Y) / (n -1)1/2, Xij’ = {Xij - E(Xj)} / S(Xj) / (n -1)1/2, i = 1, 2, ... , n, j = 1, 2, , ... , m, where E(∙) and S(∙) denote the sample average and sample standard deviations of a variable. The standardized regression coefficients ai’ means the effect of one standard deviation of Xi on Y in units of standard deviation of Y. However, one must be cautious to interpret the magnitude of standardized regression coefficients to reflecting the comparative importance of independent variables (Kutner et al., 2005). Logistic regression The regression model above having linear regression function is called linear regression model. Whereas, the regression model having non-linear regression function is called non-linear regression model. Logistic regression model is one of non-linear regression models, which will be used in Chapter 13.

Chapter 6 Prominence of Color and Details of Drawing

95

As a measure for describing the usefulness of a regression model, the Coef- Measure of model ficient Of Determination (COD), denoted by R2 (0 ≤ R2 ≤ 1), is widely used. appropriateness There is some rule of thumb to interpret R2: When R2 ≥ 0.9, the model is considered extremely appropriate, 0.9 > R2 ≥ 0.8, very appropriate, 0.8 > R2 ≥ 0.7, appropriate, 0.7 > R2 ≥ 0.6, marginally appropriate, and 0.6 > R2, inappropriate. However, readers must be cautious to interpret R2 in this way since there are many cases where the larger R2 does not necessarily mean better appropriateness.

6.2 Method and samples The systems are verified through 45 PPAT drawing samples included in its manual FEATS. Three human raters grade them on a 5-level grading system according to the guidelines given in the FEATS. They also rank them from 1 to 45. Then, all of the inter-rater reliabilities are examined. The computer systems also grade and rank them by the methods developed in this chapter. Then, the consistencies of the grades and ranks between the human raters and the system are examined. As for samples, we use 45 PPAT drawings in the FEATS. There is a set of three PPAT samples for each of 13 element that have been graded as Grade-1, Grade-3, and Grade 5, with the exception of element perseveration, which have two sets, hence a total of 45 PPAT samples. We code each sample in the FEATS with its element number (1 = prominence of color, 2 = color fit, 3 = implied energy, 4 = space, 5 = integration, 6 = logic, 7 = realism, 8 = problem-solving, 9 = developmental level, 10 = detail of objects & environment, 11 = line quality, 12 = person, 13 = rotation, 14 = perseveration-1, 15 = perseveration-2) and grade (1, 3, 5). For example, Drawing 2-3 is the PPAT samples in its manual FEATS with grade-3 for color fit. In addition, Drawing 1-1 is denoted by A, Drawing 1-3 by B, Drawing 1-5 by C, Drawing 2-1 by D, … , Drawing 9-3 by Z, Drawing 9-5 by a, Drawing 10-1 by b, … , Drawing 15-5 by s. Figure 6.1 illustrates the original, edges, and the colored convex hull versions of two samples, Drawing 4-3 (K) and Drawing 13-3 (l).

6.3 Evaluations by human raters and their inter-rater reliabilities 6.3.1 Grade Three human art therapists (Rater-1, Rater-2, and Rater-3) graded the two elements of prominence of color and details of drawings of the 45 PPAT samples on a scale of 1 to 5 (Grade-1, Grade-2, … , Grade-5) according to the guidelines of the FEATS. For example, the prominence and the de-

96

PART ONE ART EVALUATION

Figure 6.1 Original, edges, colored convex hull versions of two PPAT samples.

tails of Drawing 4-3(K) were graded as Grade-4 and Grade-2 by Rater-1; Grade-3 and Grade-2 by Rater-2; and Grade-4 and Grade-2 by Rater 3. Drawing 13-3 (l) was graded as Grade-5 and Grade-5 by all three raters. Table 6.1 shows the results of the grading of the 45 PPAT samples. All inter-rater reliabilities in the grading of the three elements were satisfactory, as demonstrated by Cohen’s Quadratic Weighted Kappa (QWK) values (Altman, 1991) between the pairs of raters below: Cohen’s weighted kappa κ2 Prominence Details Rater-1 and Rater-2 0.907 0.920 Rater-1 and Rater-3 0.896 0.914 Rater-2 and Rater-3 0.976 0.948 6.3.2 Rank The three raters also ranked each of the two elements in the 45 PPAT sam-

A B C

D E F

G H I

J K L

M N O

P Q R

S T U

V W X

1-1 1-3 1-5

2-1 2-3 2-5

3-1 3-3 3-5

4-1 4-3 4-5

5-1 5-3 5-5

6-1 6-3 6-5

7-1 7-3 7-5

8-1 8-3 8-5

Drawing

3 5 8

2 6 6

7 5 4

0 4 6

1 5 9

1 9 10

1 5 6

1 6 9

Number of used colors

14 185 283

10 136 71

74 121 85

6 133 252

9 178 608

2 169 391

12 73 110

33 60 215

Number of clusters

15,474 84,336 130,843

23,779 58,502 42,790

60,053 56,275 41,225

3,796 37,842 124,084

3,468 69,780 225,822

4,886 89,364 156,804

12,538 76,271 81,541

35,981 33,400 116,372

Length of edges

Basic elements

55,513 321,764 389,224

83,938 352,621 321,178

223,322 283,394 264,412

6,154 97,607 473,399

13,611 271,023 757,247

14,455 318,192 1,193,250

31,178 159,993 461,895

93,943 308,973 1,564,800

Area colored

367,519 1,369,193 1,564,676

632,171 1,088,621 691,462

1,442,634 1,481,951 653,432

189,906 495,700 1,421,505

25,651 750,402 1,560,019

117,102 1,461,294 1,564,800

154,676 1,564,800 1,564,800

636,885 626,227 1,564,800

Colored convex hull R

1 42 3 18 5 7

1 41 3 16 4 12

3 25 3 19 4 13

1 36 2 31 5 6

4 14 4 11 5 3

1 37 3 21 5 2

1 40 1 44 5 9

1 43 3 17 5 1

G

R-1 R

2 35 4 14 4 12

1 41 3 20 4 11

3 27 3 26 5 10

1 45 2 31 5 8

4 15 3 21 5 4

1 44 3 28 5 2

1 42 1 37 5 7

1 38 3 18 5 1

G

R-2 R

2 35 5 11 4 13

1 41 3 17 4 12

3 26 3 20 5 10

1 45 2 31 5 8

4 15 4 14 5 4

1 44 3 25 5 2

1 42 1 38 5 7

1 40 3 16 5 1

G

R-3

Prominence

R

2.1 31.1 3.3 22.3 4.3 14.6

2.0 32.8 3.4 19.6 3.1 20.1

3.6 19.5 3.2 22.9 2.6 25.4

1.3 38.5 2.6 28.2 3.7 17.6

1.5 36.2 3.1 23.1 5.2 6.2

1.5 36.2 4.3 13.6 5.0 -3.2

1.5 35.9 3.2 25.0 3.5 17.8

1.8 34.9 3.0 20.3 4.4 -7.2

G

System R

1 39 3 19 5 3

1 42 2 26 3 15

3 21 4 9 3 16

1 45 1 33 5 5

2 30 2 25 5 2

1 40 3 18 4 10

2 31 3 22 5 8

1 34 2 24 4 11

G

R-1 R

R

1 39 4 13 5 10

1 36 4 15 5 9

1 41 4 10 2 28

4 16 5 2 3 17

4 12 4 11 3 16 1 42 2 26 2 22

1 45 1 33 5 3

2 29 2 23 5 8

1 39 4 14 4 11

1 34 3 21 5 4

1 38 2 24 4 12

G

R-3

1 45 1 32 5 6

1 30 2 28 5 3

1 40 3 18 5 2

1 35 4 14 5 5

1 37 2 25 5 7

G

R-2

Details

1.3 3.5 4.6

1.6 3.2 2.5

3.8 3.6 2.1

0.5 1.9 3.9

0.5 2.6 5.4

0.6 4.3 5.1

0.7 3.6 3.9

1.4 2.3 4.6

G

34.1 16.7 8.0

32.7 18.9 24.4

14.0 16.3 28.1

41.3 29.2 13.4

41.5 23.9 1.8

40.5 9.1 3.1

40.0 16.0 13.8

34.1 25.3 7.2

R

System

Table 6.1 Quantitative measures of the basic five elements by the computer system and the results of grading and ranking of the two elements by the three human raters and the system

Chapter 6 Prominence of Color and Details of Drawing

97

k l m

n o p

q r s

13-1 13-3 13-5

14-1 14-3 14-5

15-1 15-3 15-5

6 1 6

7 4 1

3 8 4

1 1 5

*

123

120

68 24 126

58 124 29

77 145 106

20 26 58

27 88 333

29 79 340

39 95 279

Number of clusters

System values are in parenthesis G : Grade, R: Rank S.D. : Standard Deviation R-1: Rater-1, R-2: Rater-2, R-3: Rater-3

2.89

h i j

12-1 12-3 12-5

3 3 8

S.D.

e f g

11-1 11-3 11-5

4 4 11

4.80

b c d

10-1 10-3 10-5

1 8 8

Number of used colors

Mean

Y Z a

9-1 9-3 9-5

Drawings

Table 6.1 (continued)

46,872

59,448

68,782 17,567 43,380

23,705 66,709 42,241

68,706 61,161 48,285

16,598 28,586 32,913

17,327 30,679 152,455

14,582 41,113 110,735

24,643 68,158 111,587

Length of edges

Basic elements

319,648

304,831

229,276 44,733 122,474

148,349 253,986 148,725

384,348 437,219 270,673

38,543 53,115 239,392

53,022 105,573 898,625

52,003 105,237 768,357

258,909 741,231 302,520

Area colored

515,144

931,059

982,791 230,182 862,409

331,709 560,923 1,297,313

1,359,108 1,134,908 815,274

470,719 1,126,588 1,016,268

286,494 474,354 1,563,740

298,665 1,108,920 1,496,917

420,226 1,555,008 1,514,918

Colored convex hull R

2 30 1 38 3 27

3 15 3 23 2 34

3 20 5 10 3 24

1 39 3 29 3 22

2 32 2 33 5 4

2 35 3 28 5 5

1 45 5 8 3 26

G

R-1 R

R

3 27 2 37 3 24

3 30 3 19 2 36

3 22 5 9 3 23

1 43 3 29 3 18

2 33 2 32 5 6

2 34 3 28 5 5

1 39 5 3 3 21

G

R-3 R

3.2 21.6 1.6 35.7 3.3 23.4

3.2 20.8 2.6 25.6 2.0 33.9

2.6 25.6 3.9 13.8 2.7 25.3

1.7 35.8 1.9 35.5 2.9 23.7

2.1 31.1 2.3 30.3 4.4 6.1

2.4 29.0 2.8 28.1 5.2 1.7

1.7 32.1 4.0 8.7 4.3 16.0

G

System

G: 1.40 (1.06)* R: 13.04 (10.89)*

G: 3.01 (2.98)* R: 23.00 (23.00)*

3 24 1 39 3 23

3 16 3 17 2 36

3 22 5 9 3 25

1 43 3 30 3 19

2 34 2 32 5 3

2 33 3 29 5 5

1 40 5 6 4 13

G

R-2

Prominence R

R

3 21 1 34 2 23

1 43 2 24 1 36

2 27 5 9 3 20

1 41 2 29 3 17

1 33 1 31 3 15

1 38 3 19 5 1

1 44 5 4 5 8

G

R-2

R

3 22 1 35 2 25

1 43 2 27 1 40

2 26 5 7 3 20

1 42 2 30 3 18

1 32 1 31 4 13

1 37 3 19 5 1

1 44 5 5 5 6

G

R-3

2.9 0.8 2.8

2.1 2.0 2.4

3.0 3.6 2.4

1.2 2.2 2.8

1.2 1.6 4.7

1.4 2.8 5.1

1.1 4.2 4.5

G

21.1 38.9 21.7

27.1 28.6 26.5

21.7 14.9 25.9

36.2 28.5 22.5

34.8 31.8 7.3

33.0 22.9 2.9

36.5 10.7 8.6

R

System

G: 1.53 (1.38)* R: 13.04 (11.41)*

G: 2.67 (2.72)* R: 23.00 (23.00)*

2 23 1 35 2 28

1 43 4 13 1 38

2 27 5 6 3 20

1 41 1 32 3 17

1 36 2 29 4 12

1 37 3 14 5 1

1 44 5 7 5 4

G

R-1

Details

98 PART ONE ART EVALUATION

Chapter 6 Prominence of Color and Details of Drawing

99

Figure 6.2 Plots of the rank correlation for prominence of color between the pairs of raters.

ples from 1 to 45 (Rank-1, Rank 2, … , Rank-45). The highest ranking in prominence of color is denoted by Rank-1, the second highest by Rank-2, and so on. We assume that the transitivity principle applies to the ranking: if drawing A is ranked higher than drawing B, and drawing B higher than drawing C, then drawing A is ranked higher than drawing C. Thus, all possible pairs need not be compared. This was equally applied to ranking of details. For example, Drawing K was ranked as Rank-11 and Rank-25 for the evaluation of prominence and details, respectively, by Rater-1, as Rank-21 and Rank-28 by Rater-2, and as Rank-14 and Rank-23 by Rater-3, while Drawing l was ranked as Rank-10 and Rank-6 by Rater-1, as Rank-9 and Rank-9 by Rater-2, and as Rank-9 and Rank-7 by Rater-3. The ranks for the 45 PPAT samples are shown in Table 6.1. It is interesting to note that, although the raters confessed having some difficulty in applying the criteria to their ranking, all of the inter-rater reliabilities in the ranking were satisfactory, as shown in the following the Rank Correlation Coefficients (RCCs) between two raters (Walpole & Myers, 2006): Prominence Details RCC Rs Rater-1 and Rater-2 0.943 0.957 Rater-1 and Rater-3 0.950 0.938 Rater-2 and Rater-3 0.972 0.954 The plots of the rank correlation for prominence of color between the pairs of raters are shown in Figure 6.2. The plots for details of drawing are omitted. In Figure 6.2, we can find two outliers that represent large discrepancies between the pairs of raters in their evaluations, which are shown in Figure 6.3. Drawing A shows a tree trunk with no color filled in, but very detailed outlining and many, albeit small, objects colored in. Drawing n is far from the PPAT drawings and, thus, logic and realism might affect the evaluation of its prominence of color. The raters agreed on the potential inconsistency in applying the criteria and the possible correlations among

100

PART ONE ART EVALUATION

Figure 6.3 Two outliers accorded by the raters in the ranks of the prominence of color.

the elements in their evaluations.

6.4 Evaluations by regression models For each sample, the computer system analyzed the basic elements included in the quantitative measures of the methods by Kim, Bae, and Lee (2007), and Kim, Han, and Oh (2012), such as the number of used colors, number of clusters, length of the edges, area colored, area of colored convex hull, and color occupying the largest area (principal color). The results are shown in Table 6.1. We used the regression model for the computer system to grade each of the two elements in the 45 PPAT samples. The model can be formulated as Yi = a0 + a1 Xi1 + a2 Xi2 + ... + a9 Xi9 + εi, i = 1, 2, ... , 135, where εi is assumed to be an identically distributed independent normal random variable with mean 0 and constant variance σ2. As each of the two dependent variables, Yi, we used the 135 grades of the 45 samples evaluated by the three raters. 6.4.1 Grade The following independent variables were considered as potential variables to explain the grades evaluated by the three raters: number of colors (X1), principal color with the largest colored area (X2), percentage of primary colors (X3), percentage of warm colors (X4), existence of complementary colors (X5), number of clusters (X6), length of edges (X7), colored area (X8), and area of colored convex hull (X9). Principal color is one of the 15 colors defined by the KIS. To represent each of the 15 colors, the variable X2 was replaced with 14 indicator variables: X21 = 1 in the case of red, X22 = 1 for orange, X23 = 1 for yellow, X24 = 1 for light green, X25 = 1 for green, X26 = 1 for blue-green, X27 = 1 for blue, X28 = 1 for blue-purple, X29 = 1 for purple, X2,10 = 1 for red-purple, X2,11 = 1 for pink, X2,12 = 1 for brown, X2,13 = 1 for white, and X2,14 = 1 for grey. Otherwise, these variables had values of zero. In the case of the color black, X2,1 = ... = X2,14 = 0. The existence of complementary colors was represented by X5 = 1, and X5 = 0 otherwise.

Chapter 6 Prominence of Color and Details of Drawing

By stepwise regression, three statistically significant elements, whose p-values were less than 0.10 and whose interpretation could be drawn, were chosen as important variables that “well” explain the human raters’ grading of prominence of color, whose average and standard deviation were 3.01 and 1.40, respectively: number of used colors (X1), number of clusters (X6), and area of colored convex hull (X9), whose averages were 4.80, 120, and 931,059, respectively, and standard deviations were 2.89, 123, and 515,144, respectively. It should be noted that some principal colors were selected by the stepwise regression, but were subsequently omitted, since they could not be interpreted reasonably. The regression equation was obtained: Grade of prominence = 1.201 + 0.247 (0.504) x X1: Number of used colors [0.000] + 0.002 (0.203) x X6: Number of clusters [0.020] + 0.376 10-6 (0.137) x X9: Area of colored convex hull [0.093], where the values in the parenthesis indicate the standardized coefficients and the values in the square brackets indicate the p-values of the elements. For Drawing K in Figure 6.1 whose values for the number of used colors, the number of clusters, and area of colored convex hull are 5,178, and 750,402, respectively, we obtained: Grade of prominence = 1.201 + 0.247 x 5 + 0.002 x 178 + 0.376 10-6 x 750,402 = 3.1, whereas the grades determined by the three human raters were Grade-4, Grade-3, and Grade-4. For Drawing l in Figure 6.1 whose values for the number of colors used, the number of clusters, and the area of colored convex hull are 8, 145, and 1,134,908, respectively, we obtained: Grade of prominence = 1.201 + 0.247 x 8 + 0.002 x 145 + 0.376 10-6 x 1,134,908 = 3.9, whereas the grades determined by the three human raters were all Grade-5. Here is the interpretation of the equation above: As the number of used colors increases by one, the grade for prominence of color increases by 0.247 on average. As the number of clusters increases by one, the grade increases by 0.002 on average. As the area colored in number of pixels increases by 106, the grade increases by 0.376 on average. According to the standardized coefficients in parenthesis, as the number of used colors increases by its one-standard deviation (2.89), the grade increases by 0.71 on average, which is 0.504 of its one-standard deviation (1.40) when the other two elements are fixed. The other standardized coefficients can be interpreted in the same manner. Thus, when we disregard the correlation among the three elements, the number of used colors has the largest effect on the prominence of color (0.504), the number of clusters the second largest (0.203), and the number of pixels in area of colored convex hull (0.137) the smallest among the chosen independent variables.

101

102

PART ONE ART EVALUATION

As for the grades of details of drawing, whose average and standard deviation were 2.67 and 1.53, respectively, we obtained: Grade of details = 0.233 + 0.176 (0.330) x X1: Number of used color [0.000] + 0.002 (0.126) x X6: Number of clusters [0.062] + 0.150 10-5 (0.502) x X9: Area of colored convex hull [0.000]. For Drawing K, grade 2.6 is determined by the system, whereas Grade-2 was determined by all three human raters. For Drawing l, grade 3.6 is determined by the system, whereas Grade-5 was determined by all three human raters. The independent variables chosen, their coefficients, and their relative importance can be interpreted in the same manner as they were in the case of prominence of color. CODs Overall, the CODs, which are the measures representing the usefulness of the model, were obtained as R2 = 0.601 and 0.756 for prominence of color and details of objects and environment, respectively, which implies the application of the model is marginally satisfactory in the case of prominence of color. However, the value of Cohen’s weighted kappa showed the ability of the model to explain the grade for prominence of color, as we will see in the section of System validity below. 6.4.2 Rank The same regression model was used except that, instead of using grade as the value of the dependent variable, we used the 135 ranks determined by the three raters for the 45 PPAT samples whose average and standard deviation were 23.00 and 13.04, respectively. We note that these values are the same for the details since the 135 ranks consist of three sets whose elements are from 1 to 45. By stepwise regression, two elements were chosen to explain the ranking of prominence of color: number of used colors (X1) and colored area (X8). It is reasonable to expect that the set of independent variables chosen for grading an element is chosen again for ranking the same element. For instance, the number of used colors, which was used to explain the grading, was also used to explaining the rank. However, the number of clusters was omitted and the area of colored convex hull was replaced by the area colored. The different sets of explaining variables between the grade and rank imply that the independent variables are correlated with each other. The regression equation was: Rank of prominence = 38.612 - 2.195 (- 0.483) x X1: Number of used colors [0.000] - 0.167 10-4 (- 0.405) x X8: Area colored [0.000]

Chapter 6 Prominence of Color and Details of Drawing

103

For Drawing K in Figure 6.1, the values of the number of used colors and the area colored (the number of colored pixels) are 5 and 271,023, respectively, Rank of prominence: = 38.612 - 2.195 x 5 - 0.167 10-4 x 271,023 = 23.1, whereas the ranks evaluated by the three human raters were Rank-11, Rank21, and Rank-14. For Drawing l, the values of the number of used colors and the area colored are 8, and 437,219, respectively, Rank of prominence: = 38.612 - 2.195 x 8 - 0.167 10-4 x 437,219 = 13.8, whereas the ranks evaluated by the three human raters were Rank-10, Rank9, and Rank-9. The selected independent variables, their coefficients, and their relative importance in the ranking can be interpreted in the same manner as they were in the case of the grading. As the number of used colors increases by one, the rank of prominence of color decreases (i.e. the ranking becomes higher) by 2.195 on average. As the area colored increases by 104 pixels, the rank of prominence decreases by 0.167 on average. According to the standardized coefficients in parenthesis, assuming no correlation among the elements, the number of used colors has a greater effect on the ranks of prominence than the area colored. For the rank of the details of drawing we obtained: Rank of details = 33.609 - 1.693 (- 0.372) x X1: Number of used colors [0.000] - 0.014 (- 0.128) x X6: Number of clusters [0.061] - 0.116 10-4 (- 0.457) x X9: Area of colored convex hull [0.000]. We note that the same variables that were used to explain the grade were chosen again to explain the rank, which is reasonable. For Drawings K and l, the ranks of details obtained were 24.9, and 14.9, respectively. CODs Overall, the CODs, were obtained as R2 = 0.685 and 0.753, for prominence of color, and details of drawing, respectively, which demonstrate the marginal applicability of the model in the case of ranking prominence of color. However, the usefulness of the model in explaining the ranks of the two elements is demonstrated by the RCCs, as seen below.

6.5 System validities The applicability of the regression models to grading and ranking of each

104

PART ONE ART EVALUATION

____________________________________________________________

Table 6.2 Classifications of grades between the human raters and the system

Prominence Details

______________________

_____________________

G-1 G-2 G-3 G-4 G-5 G-1 G-2 G-3 G-4 G-5 Grade-1 1 7 1 - - 9 6 - - Grade-2 - 5 1 - - 1 4 4 - Raters Grade-3 - 1 12 2 - - 2 3 2 Grade-4 - 1 3 1 - - - 1 3 2 Grade-5 - - 1 3 6 - - - 6 2

____________________________________________________________ 1-1: M, 1-2: A D G S Y h r, 1-3: E 2-2: V b e f p 2-3: N 3-2: i, 3-3: B P Q T c j k m n o q s, 3-4: H a 4-2: J, 4-3: K U W, 4-4: X 5-3: R, 5-4: F O l, 5-5: C I L Z d g

1-1: A D G M V Y e h r, 1-2: N S b f n p 2-1: J, 2-2: B K i o, 2-3: T U k s 3-2: R m, 3-3: c j q, 3-4: E H 4-3: W, 4-4: P Q g, 4-5: C I 5-4: F O X Z a l, 5-5: L d

____________________________________________________________ G: Grade i-j: Grade-i and Grade-j by the raters and the system, respectively

element was determined by Quadratic Weighted Kappa (QWK) between the grades and the Rank Correlation Coefficient (RCC) between the ranks, respectively, determined by the system and the human raters. 6.5.1 Grade QWKs We rounded off the grade determined by the system to the nearest integer and modified grades below 1 to Grade-1 and grades above 5 to Grade-5. For the QWK values, the median grade of the three raters was used. For example, when the three grades were Grade-2, Grade-2, and Grade-3, we used Grade-2, when they were Grade-3, Grade-4, and Grade-5, we used Grade-4, and when they were Grade-2, Grade-3, and Grade-5, we used Grade-3. The QWK values were as follows, which show the applicability of the grading system to all of the elements: Prominence Details QWK κ2 Human and System 0.781 0.867 Table 6.2 shows the classifications of the grades between the human raters and the system. For prominence of color, the grades in 25 drawings determined by the system are identical to the median grades determined by the human raters, the grades in 17 drawings determined by the system are one grade higher or one grade lower than the median grade determined by the human raters, the grades in two drawings determined by the system are two grades lower than the median grade determined by the human raters, and the grade in one drawing determined by the system is two grades higher than the median grade determined by the human raters. We find that there is no grade difference greater than one between the human raters and the system for details of drawing.

Chapter 6 Prominence of Color and Details of Drawing

105

Figure 6.4 Plots of the rank correlation between the raters and the system for prominence of color and details.

6.5.2 Rank We examined consistencies between the ranks determined by the system RCCs and the median averages of the three human raters. The regression models seemed appropriate for both elements. The RCC values are shown below, which show the applicability of the ranking system to both elements: Prominence Details RCC Rs Human and System 0.833 0.869 The correlation plots between the median ranks of the three human raters and the ranks of the system are shown in Figure 6.4. We find two outliers for prominence of color acknowledged by both the human raters and the system, as shown in Figure 6.5: Drawing E shows only the outlining of objects with no color filled in, but a relatively large number of used colors in a large portion of the space. Drawing J has all of the objects colored in, but extremely small space usage. These drawings may represent cases in which there is a possible inconsistency between the human raters in applying the criteria to ranking elements or a possible limitation in the application of the system to special or abnormal drawings. After all, the two systems, the re-

Figure 6.5 Two outliers in the ranks of prominence of color acknowledged both by the raters and the system.

106

PART ONE ART EVALUATION

gression of grades and regression of ranks, showed almost the same degree of appropriateness for both elements.

6.6 Discussion and conclusion In this chapter, we developed four statistical regression models. Two for evaluation of prominence of color, among which one is in grade and the other is in rank. Other two are for details of drawing. Readers can use statistical packages such as the SPSS and the SAS in model building. However, readers must understand the basic concept of various statistical terms such as standardized regression, stepwise regression, coefficients of determination, etc. Inter-rater We obtained satisfactory inter-rater reliabilities in evaluations between the reliabilities human raters. However, there exist some outliers in the drawing samples that caused large discrepancies in the evaluation outcomes of the three human raters. This involves the problem of human subjectivity in art evaluation, a limitation inherent in all art evaluations. Raters, despite their subjective and possibly uncertain knowledge, are often compelled to execute their work by relying on professional observation and judgment. It has been pointed out that, even when raters are provided with concrete descriptors for all levels of elements evaluations, they are likely to evaluate specific aspects of drawings differently, simply because they like certain drawings better than others (White et al., 2004). All evaluations of elements in the DAPA, the DDS, and the FEATS are more or less subjective and their evaluation results may differ depending on the human raters. Also, there exists a halo effect between elements, which means that when some elements of a drawing are evaluated either high or low, the other elements of the drawing are also likely to be evaluated either high or low, proportionally. Nonetheless, we obtained satisfactory inter-rater reliabilities in both grading and ranking for both elements. System reliability We also obtained high consistencies between the human raters and the system in grading and ranking of the elements. Thus, the usefulness of the system in terms of automatically providing art therapists with quantitative, objective, and detailed information and of reducing the time and effort involved in evaluating various elements in large quantities of drawings is proven. However, there are some outliers showing large discrepancies between the human raters and the system. The system detected a general tendency between the variables in the normal drawings, but not in the abnormal or special drawings. This implies that the computer system only offers assistance to human raters and that the final decision should be made by human experts. The system could be rendered more useful by using an

Chapter 6 Prominence of Color and Details of Drawing

expert system approach covered in Chapter 9 (Kim, Ryu, et al., 2006) and Chapter 10 (Kim, Kim, et al., 2006). For most elements in art therapy assessments, we recommend using ranks rather than grades, since the former provides more detailed information. Experts in art therapy could compose a standard sample of one hundred drawings, which appropriately represents the whole range of grades for each of the elements and establish a system based on this sample. Then, some authoritative institution, such as the American Art Therapy Association (AATA), could officially acknowledge the system as a standard method and criteria for the evaluation of that element. We admit that the system obtained relatively low values of the QWK and RCC for prominence of color. The development of some new independent variables to explain prominence of color by applying one or more computer algorithms deserves further research.

107

Evaluation of Space Usage in a Drawing and Degree of Concentration in a Pattern Coloring

Chapter 7 Key points We delineate the development of computer systems that objectively and quantitatively evaluate space usage in drawings and degree of concentration in a pattern coloring using statistical regression models and computer analysis. Space usage (or simply space) in drawings is an important element that provides useful information about the drawers’ level of energy and psychological state. Degree of concentration (or simply concentration) involved in the activity of coloring a given pattern is known to be useful for establishing emotional stability and for healing of psychological disorders. Dependent variables that are to be explained by the regression models are grade and rank in the case of evaluation of space usage and only rank when it comes to evaluation of concentration. Human raters evaluate these variables in the samples and inter-rater reliabilities are examined in terms of kappa values. Possible independent variables that are to explain the dependent one are all the elements in the C_CREATES. These elements are evaluated by the computerized methods developed in the previous chapters. Some important independent variables are selected by the stepwise regression, and their relative effects on the dependent variable are determined by the standardized regression. We emphasize the contribution of area of colored convex hull in explaining space and accuracy in explaining concentration. We measure the validities of the systems in terms of the kappa values, rank correlation coefficients, and coefficients of determination.

7.1 Importance of space usage and degree of concentration Definition of space Space usage (or simply space) in art evaluation refers to the extent to which a drawer uses a drawing sheet as an outlet for expression. Along with other elements, such as the type of drawing, firmness of line, specificity or vagueness of contours, and types of configurations chosen, space usage is known to reveal important information about the experience submerged in the drawers’ inner mind (Garai, 1976). Particularly, space was assumed to be 108

Chapter 7 Space Usage and Degree of Concentration

correlated with the drawers’ energy (Gantt & Tabone, 1998). Lehmann and Risquez (1953) considered space as an element to determine the degree of energy output in a finger painting. Space usage in schizophrenics’ drawings was very low, because people who could not use their energy on focusing on external objects were not very interested in new experiences and tended to cling to their personal problems (Russell-Lacy, Robinson, Benson, & Cranage, 1979). This was also true for melancholiacs (Gulbro-Levit & Schimmel, 1991; Wadeson, 1980; Dawson, 1984). In contrast, the space of patients suffering from brain disease was high (Reitman, 1947). In HFD drawings, the size of a person represents space. Compared with white patients’ figures, which were typically of normal size (about 18 cm), black patients’ figures tended to be smaller (Adler, 1970, 1971). The figures drawn by widows, who usually have fears about their physical safety, being alone, darkness, victimization, and feeling unwelcome in the world of couples, tended to be smaller than the ones by wives (Hammer & Piotrowski, 1997; Lopata, 1976, 1996). It was reported that the particular emotional terms used to describe a human figure (“nice” vs. “happy”, “nasty” vs. “sad”), as well as the valence of the affective characterization (i.e. positive vs. negative), can influence the size of children’s HFDs and, thus, children drawing “nice” and “happy” men systematically drew larger figures than those drawing “nasty” and “sad” ones (Brukitt, Barrett, & Davis, 2009). Thus, space was considered to be a very important element in various art therapy assessments for art evaluation and for art interpretation. The definitions of space usage in various art therapy assessments are essentially the same. Most art therapy tools use an interval scale, such as the Likert type. The only difference, if any, is their specific criteria. The Descriptive Assessment of Psychiatric Artwork (DAPA) (Hacking. 1999) uses five grades corresponding to space usages of 10%, 10 - 25%, 25 - 55%, 55 - 80%, and 80 - 100%. The FEATS of the PPAT, uses six grades corresponding to the space usages, which are Grade-0 for 0%, Grade-1 for 0 25%, Grade-2 for 25%, Grade-3 for 50%, Grade-4 for 75%, and Grade-5 for 100%. The system also allows for grades between the numbers. The DDS uses four grades, which are 0 - 32%, 33 - 66%, 67 - 99%, and 100%. The DDS provides particularly specific criteria that divides the whole paper into 3 x 3 = 9 grids and counts the grids that are colored over a length of more than 2 inches. Also, the KFD provides a specific criterion that uses rulers or grids made of tracing paper marked off in millimeters to measure the sizes of family members in drawings and the distances between them. While space usage is widely used in art therapy assessments, some inconsistencies in the evaluation outcomes are hardly avoidable, since the evaluation requires human raters’ intuition, judgment, and subjective determination. For instance, coloring the background of a picture is counted as space usage by some raters, but not by others. In spite of the use of rulers and grids, as in the KFD, which facilitates objective and quantitative eval-

109

110

PART ONE ART EVALUATION

uation, and of the effort and time consumed in the process of evaluation, complete accuracy may not be guaranteed. Thus, we develop computer systems for automatic, objective, and rapid evaluation of space, using regression models and computer analysis. Definition Degree of concentration (or simply concentration) involved in the activity of concentration of coloring a given pattern is an element first proposed in the C_CREATES, but not yet considered in other art therapy assessments. Coloring a reasonably complex geometric pattern may induce a meditative state that benefits individuals suffering from anxiety (Curry & Kasser, 2005). Belchamber (in Curry & Kasser, 2005) recommended the use of structured mandalas with symmetrical figures that have long been used as meditative objects in spiritual traditions. However, Curry and Kasser reported that the anxiety level declined to approximately the same extent for the mandala- and plaid-coloring groups. Based on this observation, it is believed that the key element is not the circle, but concentration involved in the activity of coloring a given pattern. Thus, we propose concentration as an important element. We use the structured mandala introduced in Chapter 2 as a pattern. The basic function of mandala coloring is to provide order and balance for the person who concentrates on and completes coloring it. Bell and Robbins (2007) examined the long-held claim that art making reduces stress and elevates mood. The results of their study demonstrated that the simple act of creating a work of art can produce dramatic reductions in negative mood and that such reductions can be attributed specifically to the producing of art rather than viewing art. Riedel (1999) reported that patients participating in mandala art therapy produced clearer drawings and exhibited increased concentration. Cornell (1994) stated that in coloring a mandala, a patient forgets his or her everyday work, worry, and anxiety, and becomes comfortable, calm, and engaged in coloring. Thus, degree of concentration involved in coloring determines the effectiveness of the art therapy tool. Such an element has not been considered in previous studies, not because it is not important, but because there has been no appropriate method of measuring it objectively and quantitatively in numbers. In Chapter 14, degree of concentration will be described as an important variable that explains the degree of dementia (Kim, Kim, & Kang, 2008). Now we devise a system for measuring degree of concentration. Here, we will describe an approach we adopted to develop a computer system using regression models and computer analysis for the evaluation of space and concentration. Human raters evaluate the dependent variable that is to be explained in grade and rank for space in 145 PPAT drawings, and rank for concentration in 58 SMC colorings. The inter-rater reliability of grades is measured by the Quadratic Weighted Kappa (QWK) value and that of ranks is measured by the Rank Correlation Coefficient (RCC). All elements in the C_CREATES are considered as possible independent

Chapter 7 Space Usage and Degree of Concentration

111

variables that explain the dependent variable. The most important independent variable is expected to be area of colored convex hull in evaluation of space, and accuracy in the evaluation of concentration. Important exploratory (independent) elements are selected by stepwise regression and their relative effects are determined by standardized regression. The validity of the computer systems for grades is measured by the QWK value between the median grade of the raters and the round-off grade of the system. The validity for ranks is measured by the RCC between the mean rank of the raters and the rank of the system. Also, the usefulness of the regression models is measured in terms of Coefficient Of Determination (COD) and RCC. We emphasize the necessity of this kind of computer system approach, which supplements the existing art therapy methods.

7.2 Regression models for the evaluation of space usage in grade and rank 7.2.1 Possible independent variables Two regression models have been developed for the evaluation of space Two models usage. See Chapter 6 for detailed explanation on these regression models. The dependent variable of the first model is the grade for space, and that of the second model is the rank. The former model is referred to as Model-G (Grade) and the latter as Model-R (Rank). The area colored and its dispersion can be considered as factors affecting space. The methods of evaluating space can be classified into two classes, depend- Three methods ing on these factors. The first one includes the DAPA and the PPAT, and consists of methods that use the area colored. From the viewpoint of a computer system, the area colored corresponds to the number of colored pixels. A piece of paper is divided into pixels, each of which is the final element to be analyzed. For example, if the vertical side is divided into 1,600 points and the horizontal side into 978, the paper consists of a total of 1,564,800 pixels (1,600 x 978). The second class consists of those methods in which the number of colored grids is counted when more than a given percentage (i.e., 20%) is colored. For example, if the vertical and horizontal sides are divided into 5 equal segments, the paper consists of a total of 25 grids (5 x 5). The DDS adopts this method, and hence belongs to the second class. The former class is referred to as Method-P (Pixels) and the latter as Method-G (Grids). However, both methods have problems in appropriately evaluating space usage. Method-P cannot discriminate between two different drawings with a similar number of colored pixels of different dispersions. Conversely, Method-G cannot discriminate between two drawings with a similar dispersion of different numbers of colored pixels. However, these two methods are expected to complement each other when used together.

112

PART ONE ART EVALUATION

Figure 7.1 Three drawing samples illustrating grid and convex hull.

Also, neither Method-P nor Method-G can count as colored the empty space between two subjects that are a certain distance apart from each other. In general, it is reasonable to consider this empty space to have been used by the drawer as an expression. To overcome these weaknesses, a method using area of colored convex hull (or simply convex hull) is proposed in this paper. A convex hull includes straight lines connecting any two points contained within it. When two points are considered to have been colored in a drawing, every point between them is also considered to have been colored. This method is called Method-C (Convex hull). Three examples Figure 7.1 illustrates the three methods applied to three PPAT drawings from the 145 samples used for the model development. Art therapists naturally rank drawing A as the one with the most space usage, drawing B as the second most, and drawing C as the least: A > B > C. Drawings A and B have similar areas colored and drawings B and C have

Chapter 7 Space Usage and Degree of Concentration

113

Table 7.1 Values of independent variables evaluated by the C_CREATES and the results of grading and ranking by the raters and the system

__________________________________________________________________________________________________ Number of Number of Area Length of Number of Area of Figure __________________________________________________________________________________________________ colors clusters colored edges grids convex hull 7.1A 8 169 302,133 94,809 25 1,379,088 7.1B 4 86 405,274 51,603 13 1,078,615 7.1C 5 75 104,506 33,679 15 657,136 __________________________________________________________________________________________________

_______________________________________________________________ Model-G

___________________________

Rater

___________________

Model-R

____________________________

System Rater

_____

____________________

System

_____

1 2 3 Median 1 2 3 Mean _______________________________________________________________ 7.1A 5 5 5 7.1B 4 4 4 7.1C 3 3 3

5 4.90 10 8 11 9.67 0.68 4 3.44 28 35 26 29.67 52.30 3 2.87 54 70 65 63.00 69.54

_______________________________________________________________ similar numbers of colored grids. Method-P evaluates drawing B as having slightly more space usage than drawing A, and drawing C as the least: B≥A>C based on the order of number of colored pixels, 405,274, 302,133, and 104,506, respectively. Thus, Method-P committed an error in evaluating between drawings A and B, which actually have similar areas colored, but with different dispersions. On the other hand, Method-G evaluates them as A>C≥B based on the order of number of colored grids, 25, 15, and 13, respectively. Thus, Method-G committed an error in evaluating drawings B and C, which have actually similar numbers of colored grids but with different areas colored. Now, Method-C correctly evaluates the three drawings as A>B>C based on the order of numbers of pixels in the colored convex hull, 1,379,088, 1,078,615, and 657,136, respectively. Thus, the area of colored regions, number of colored grids (or simply grids) and area of colored convex hull (or simply convex hull) are all considered as independent variables. Along with these three independent variables, all elements in the C_CREATES such as number of used colors, number of clusters, and length of edges are also considered as independent variables, and thus evaluated by the computerized methods developed in the previous chapters. Art therapists graded (Model-G) and ranked (Model-R) 145 PPAT drawing samples consisting of 45 drawings in the manual FEATS and 100 drawings by patients with suspected depression. The detailed information about

114

PART ONE ART EVALUATION

the three drawings by the C_CREATES and the results of grading and ranking by the raters and by the system are summarized in Table 7.1. We note that grades and ranks by the system which are late to be addressed are also presented in the table. 7.2.2 Inter-rater reliabilities in the evaluation of dependent variables The dependent variable in Model-G is the grade for space usage, 6-point Likert scale, and the dependent variable in Model-R is the rank, 1 to 145 (n: sample size). The same sample of drawings mentioned in the previous section was graded from 0 to 5 (from Grade-0 to Grade-5), according to the FEATS guidelines. However, grades between numbers were not allowed. The same sample was also ranked for space usage applying the transitivity principle: If drawing A is ranked higher than drawing B, and drawing B higher than drawing C, then drawing A is ranked higher than drawing C. Rank-1 is assigned to the drawing with the most space, Rank-2 to the second most, and so on, up to Rank-145, to the least space. We examined all of the inter-rater reliabilities. The QWK values are used in Model-G, and the RCCs in Model-R. QWKs and RCCs The measures of inter-rater reliabilities, as indicated in the QWK values in Model-G and the RCCs in Model-R are as follows: Rater-1 and Rater-2 Rater-1 and Rater-3 Rater-2 and Rater-3 Model-G κ2 0.980 0.951 0.939 0.973 0.984 0.970 Model-R Rs The very high QWK values at 0.939 - 0.980, and also the very high RCC values at 0.970 - 0.984 prove that very reliable data were obtained for developing the regression models. The inter-rater reliabilities between Rater-2 and Rater-3 were the lowest in both Model-G and Model-R. The classification of grades between them is shown Figure 7.2a and the correlation plot of ranks between them in Figure 7.2b. The 145 samples are designated as A, B, ... , Z, a, b, ... , z, A’, B’, ... , Z’, a’, b’, ... , z’, A’’, B’’, ... , Z’’, a’’, b’’, ... , n’’, o’’. In Figure 7.2a, among 145 drawings, 117 drawings show the coincidence and 28 drawings show onegrade differences, which prove the high inter-rater reliability in Model-G. In Figure 7.2b, all drawings except a few show very high consistencies in evaluating the ranks, which prove high inter-rater reliability for Model-R. 7.2.3 Regression models In Model-G, using stepwise regression, number of colored grids whose average is 13.2 and standard deviation is 5.52; area of colored convex hull whose average is 692,716 and standard deviation is 448,125; and number of clusters whose average is 82.6 and standard deviation is 88.48 were selected as independent variables to “well” explain the grade whose average is 2.77 and standard deviation is 1.255, and the regression function was obtained as

Chapter 7 Space Usage and Degree of Concentration

115

Figure 7.2 Classification of grades and the correlation plot of ranks between Rater-2 and Rater-3.



Grade of space = 0.599 + 0.073 (0.319) x Grids [0.000] + 1.670 × 10-6 (0.595) x Convex hull [0.000] + 0.001 (0.052) x Clusters [0.020], where the numbers in parenthesis are standardized coefficients and those in square brackets are significance levels. In Model-R, number of colored grids and area of colored convex hull were selected and the regression function obtained was: Rank of space = 149.501 - 2.930 (- 0.385) x Grids [0.000] - 5.48 x 10-5 (- 0.584) x Convex Hull [0.000]. The CODs of the two models were CODs 2 R = 0.876 and 0.912, which suggest the high aptness of the two models. For example, the grade and rank of drawing C in Figure 7.1, whose number of colored grids, area of colored convex hull, and number of clusters are 15, 657,136, and 75, respectively, were: Grade of space = 0.599 + 0.073 x (15) + 1.670 x 10-6 x (657,136) + 0.001 x (75) = 2.87, Rank of space = 149.501 - 2.930 x (15) - 5.48 x 10-5 x (657,136) = 69.54, and these values were close to the average grade and rank obtained by human raters at 3.00 and 63.00, respectively. Now, we interpret the above regression function for the grade. As number of colored grids increases by 10, the grade increases by 0.73 on average. As

116

PART ONE ART EVALUATION

Figure 7.3

Classification of grade and rank between the human raters and the system.

the number of pixels in convex hull increases by 105, the grade increases by 0.167 on average. And as number of clusters increases by 100, the grade increases by 0.1 on average. According to the standardized coefficients in parenthesis, as grids increases by its one-standard deviation of 5.52, the grade increases by 0.400 on average, which is 0.319 of its one-standard deviation of 1.255. The other standardized coefficients can be interpreted in the same manner. Thus, assuming no inter-correlation among these elements, convex hull has the largest effect on the grade of space usage (0.595), grids the second largest (0.319), and number of clusters (0.052) the smallest among the chosen independent variables. The interpretation of the regression function for the rank can be done in the same manner. 7.2.4 System validity QWKs and RCCs As the measures of system validities, the QWK values between the raters and the system in Model-G and the RCC values in Model-R are as follows: Rater-1 Rater-2 Rater-3 Median/Mean 0.927 0.928 0.898 Model-G κ2 0.932 0.977 0.972 0.967 0.972 Model-R Rs The kappa values were very high, κ2 = 0.898 - 0.932, and also the rank correlation coefficients were very high, Rs = 0.967 - 0.977, which means that the system can be used to provide art therapists with very useful information about the space usage in both of grade and rank. The classifications of median grades of the three raters and the system, the correlation plot of ranks between the average ranks of the three raters and the systems are shown in Figure 7.3a and Figure 7.3b, respectively. In Figure 7.3a, among 145 drawings, 115 drawings show the coincidence and

Chapter 7 Space Usage and Degree of Concentration

117

Figure 7.4 Three SMC samples.

30 drawings show one-grade differences, which prove the high validity of Model-G. In Figure 7.3b, all drawings except a few show very high consistencies in evaluating the ranks, which prove the high validity of Model-R. It is noted that Figure 7.3b shows the possibility of nonlinear regression models. However, since very slight improvement was obtained by applying nonlinear regressions, we decided to use linear models obtained hitherto for the simplicity of use.

7.3 Regression model for the evaluation of concentration in rank In the previous section, we developed a regression model for the evaluation of space usage in grade which uses a Likert type scale choosing one among four, five, or six grades. The regression model showed its applicability to this type of grading that is used in most art therapy assessments. Assigning a grade to each drawing only results in a 5 grade difference, at most. In fact, assigning an order rank to each drawing within the sample size of, for example, fifty, gives much more detailed and useful information. Thus, for the evaluation of concentration we develop only a model for ranking. 7.3.1 Sample pattern colorings Structured mandala was used as a pattern, and 58 SMC samples were collected from sixty to ninety year old persons with suspected dementia in a sanitarium. Figure 7.4 shows again the three SMC samples in Chapter 2. In Table 7.2, their quantitative information regarding the elements in the C_CREATES is given. It also shows the ranks determined by the raters and by the system.

Primary Second Complemen color (%) -tary colors

Warm Cool color (%)

4.0 29.3 18.6

40.4 38.2 63.2

11.95

14.90

0.590

0.295

35.67

30.36

42.60

12.09

System ______

________________________________________

7.4A 4 8 5 5.67 5.47 7.4B 29 32 23 28.00 26.14 7.4C 23 15 17 18.33 19.12

________________________________________

Rater Figure __________________________ 1 2 3 Mean

Rank __________________________________

________________________________________

5.79 12.11 0.226 0.139 27.83 26.95 29.77 22.12 _________________________________________________________________________________________________________________________________________

Average Standard Deviation

Total _________________________________________________________________________________________________________________________________________

_________________________________________________________________________________________________________________________________________

7.4A 10 18 0.82 74.6 purple red 40.4 34.0 purple/ yellow 7.4B 4 19 0.45 60.1 green orange 33.6 65.3 orange/ blue 7.4C 7 15 0.87 75.6 red orange 59.4 32.7 orange/ blue

_________________________________________________________________________________________________________________________________________

Figure Number Number Complete Accuracy Main Subsidiary of colors of clusters -ness (%) color

_________________________________________________________________________________________________________________________________________

Table 7.2 Summary of analysis

118 PART ONE ART EVALUATION

Chapter 7 Space Usage and Degree of Concentration

119

Figure 7.5 Correlation between Rater-A and Rater-B.

7.3.2 Inter-rater reliability Three experts, denoted by Rater-1, Rater-2, and Rater-3, evaluated the degree of concentration in the 58 SMC samples. Rater-1 is an art therapist, Rater-2 a color therapist, and Rater-3 an elementary school teacher. We denote the 58 SMC samples by A, B, ... , Z, a, b, ... , z, 1, 2, ... , 6. The SMC sample evaluated with the highest concentration is ranked first, the SMC sample evaluated with the second highest concentration is ranked second, and so on. No explanation or definition of degree of concentration was given. When a rater insisted on knowing the meaning of “degree of concentration,” the only reply given was based on the definition of “concentration” in Webster’s New World Dictionary (Webster’s New World, 1988): “to concentrate on coloring the structured mandala is to direct one’s thoughts or efforts, or to fix one’s attention to it.” No ties were allowed. Thus, the SMC samples were ranked from 1 to 58. In Figure 7.5, the correlation of the ranks between Rater-1 and Rater-2 is RCCs plotted. The RCCs between each pair of raters are obtained as follows. Between Rater-1 and Rater-2: Rs = 0.890, Between Rater-2 and Rater-3: Rs = 0.862, Between Rater-2 and Rater-3: Rs = 0.847. After all, we concluded by the high reliabilities between all pairs of raters that there were high consistencies in the ratings given by the three raters. 7.3.3 A regression model and its validity By applying stepwise regression, number of used colors, number of clusters, and accuracy have been selected as important independent variables to explain degree of concentration, the dependent variable. Completeness,

120

PART ONE ART EVALUATION

Figure 7.6 Correlation between raters and system.

one of the independent variables, has been omitted from the model, not because it is less important, but because it is highly correlated with the other independent variables included in the model. The final regression function was obtained as Rank of concentration = 65.171 - 2.973 (0.461) x Number of colors [0.000] - 0.666 (0.239) x Number of clusters [0.000] - 0.241 (0.337) x Accuracy [0.015]. COD The COD which is a measure of the model’s appropriateness was obtained as R2 = 0.740, which is a satisfactory value. The standard deviation of concentration was estimated to be 8.44. The residuals, which are not presented here, showed the normality, the independency, and the constant variance of the error term, εi. In conclusion, the proposed model is highly appropriate and can be used extremely effectively. The negative signs of coefficients coincide with our insight. The larger numbers of colors and clusters and the greater accuracy, the lower the rank of the dependent variable, i.e. the higher concentration. For example, when the values of the other independent variables held constants, as number of colors increases by one, the estimated rank decreases by 2.973 on average. According to the standardized coefficients in parenthesis, for example, as accuracy increases by one of its sample standard deviation (22.56 %), the rank decreases (concentration increases) by 0.037 times its sample standard deviation, i.e. 0.037 x 16.12 = 5.43, on average. From these standardized coefficients, assuming no inter-correlation among these elements, we

Chapter 7 Space Usage and Degree of Concentration

121

concluded that the independent variables that have the greatest effects on concentration are number of colors, accuracy, and number of clusters in descending order. Figure 7.6 shows the correlation between the average ranks of the three hu- RCC man raters and the ranks of the system. The value of the RCC was obtained as Rs = 0.860. This high value of RCC verifies the applicability of the proposed system for measuring concentration.

7.4 Discussion and conclusion We formulated regression models which determine grades and ranks of space usage and of degree of concentration in a drawing. The high values of the Coefficient Of Determination (COD) of the models and of the Rank Correlation Coefficients (RCC) between the human raters and the system verified the usability of the system. The system can be used together with other computer systems for evaluation of elements (independent variables) mentioned in the previous chapters. Until now, human raters have evaluated space usage using subjective judgments based on the human rater’s visualization of the drawing with the naked eye, which may lack consistency. It was mentioned that a standardized method of evaluating space usage is very difficult to achieve, since many factors, such as forms and their movements, affect space (Richter, 1987). However, there was a very high reliability between the raters in the evaluation, showing high QWC values from 0.939 to 0.980 for grade, and high RCC values from 0.970 to 0.984 for rank. Therefore, it is believed that consistency can be obtained even between subjective evaluations. By stepwise regression, we selected grids and convex hull as independent variables which satisfactorily predict both grades and ranks of space. In Model-G, the QWK value was κ2 = 0.898 between the median grade of raters and the grade of the system, and in Model-R, the RCC value was Rs = 0.972 between the mean rank of raters and the rank of the system, which proves high validity of the system. We evaluated space not only by grading, but also by ranking. Ranks provide more information than grades. Experts in art therapy could compose a standard set of samples which appropriately represent space, establish a system based on this set, and standardize it by transforming the ranks into a percentage value. Then, some authoritative institution, such as the American Art Therapy Association, could officially acknowledge the system as a standard method for evaluation of space. Gantt (2000) emphasized the need for standardized instruments in creative art therapies that meet the scientific

122

PART ONE ART EVALUATION

requirements of reliability and validity. The computer system developed in this study can be used for such purposes. As for the evaluation of concentration in rank, we also obtained high inter-rater reliability and validity of the system. However, a clearer definition of degree of concentration along with more sophisticated algorithms for completeness and accuracy might have allowed the system to obtain even better results. Degree of concentration involved in the activity of coloring a given pattern is first suggested as an element by the C_CREATES. Here the structured mandala is selected as an art therapy medium with a pattern. The artistic and technological tools of this age are creating a growing demand for art therapists who can interact not only with new media but also with the values and pathologies of an emerging techno-culture (Kapitan, 2007).

A Bridge from Part One to Part Two: Computerization of Art Evaluation and Its Application to Art Interpretation

Chapter 8 Abstract Hitherto, we have completed developing computer methods or systems for evaluation of all the 19 elements in the C_CREATES. Now, we show how their computer methodologies and approaches can be applied to the computerization of an existing art therapy assessment for evaluation of elements. As a case study, we evaluate several elements of the standardized pattern of the human face in the FSA using computer methods and algorithms. Then, to show how the results of art evaluation in Part One can be used as information for art interpretation, we decide whether or not an element (e.g., space usage) is statistically related to a certain group of psychological state (e.g., dementia) and measure how much it is related to the degree and the severity of the psychological state. Also, we show how the elements in the FSA differentiate and identify the severity groups of dementia.

8.1 An approach to developing a computerized evaluation system and its connection to art interpretation As in the previous chapters, we develop a computer system for objective and Computerization quantitative evaluation of elements in the FSA developed by Betts (2003) using computer algorithms and regression models. We denote this computerized FSA as the c_FSA (Computerized Face Stimulus Assessment) (Kim, Kim, & Hong, 2013). The methodologies and approaches are the same as in the previous chapters. The algorithm divides an FSA drawing into several areas, determines the main color of each area, compares the main color with one of its surrounding areas, and then evaluates the elements in grades. FSA samples are drawn by elderly persons with suspected dementia and undergraduate students with a normal state of health. Human raters grade the five elements in the FSA, based on its evaluation guidelines. Also, the computer system grades the same five elements. For each element, the in123

124

PART ONE ART EVALUATION

ter-rater reliabilities and the reliabilities between the human raters and the computer system are measured by the Quadratic Weighted Kappa (QWK) for the grading of each of the five elements. The c_FSA has been verified to be useful for art evaluation. Now we understand how the methodologies and approaches can be used in computerizing an existing art therapy assessment having elements that are different from ones in the C_CREATES. Art evaluation An element in a drawing gives information about drawer’s some kind of - Art interpretation psychological state or disorder. Of course, several elements together also give that information. We show the relationship between a single element, space usage in the PPAT evaluated in grades and ranks by the system developed in Chapter 7, and the severity groups of dementia. Here, the severity is classified into two or three groups. We also show the relationship between five elements in the c_FSA and the degree of dementia. Here, the degree of dementia is scored quantitatively by a test and the severity of dementia is classified qualitatively into two or three groups. For the former case of space usage in the PPAT, based on the samples in Chapter 7, the ANalysis Of VAriance (ANOVA) is used to conclude that the evaluation results are quite different depending on the severity groups of dementia and the correlation analysis is used to measure the strength of the relationships between the evaluation results of space usage and the degree of dementia scored by the Mini-Mental State Examination-Korean (MMSE-K) (Kwon & Park, 1989). The MMSE-K is a modified version of the Mini-Mental State Examination (MMSE) developed by Folstein, Folstein, and McHugh (1975), which is a simplified form of the cognitive mental status examination. For the latter case of c_FSA, correlation analysis (Rodgers & Nicewander, 1988) selects the elements that are correlated with the degree of dementia; factor analysis (Walpole & Myers, 2006) finds the elements which differentiate between the normal group and the dementia group; and regression analysis of which the independent variables are the grades of elements identifies these groups by estimating the probability of dementia. The system developed is verified through the analysis of sample results. The system is expected to promote the usage of the FSA and assist research of FSA validity. Now we come to understand how the art evaluation in Part One is applied to the art interpretation in Part Two.

8.2 Computerization of the Face Stimulus Assessment (FSA) 8.2.1 The FSA Betts (2003) found that clients with autism, communication difficulties, and particularly a lack of motivation benefited from stimulus drawing such as a picture of a human face. She therefore developed her own projective drawing test, the FSA. A human face to color in is provided in the FSA, Betts

Chapter 8 A Bridge from Part One to Part Two

125

states that the FSA is appropriate when: ... evaluating my nonverbal clients who had cognitive impairments, as these individuals were unable to follow directions and were not motivated to draw without a visual stimulus. The clients who had severe mental retardation or developmental delay, for example, were usually unresponsive to a basic directive, such as “Drawing a person.” (p. 77) The FSA is composed of three types of pictures. The first picture, shown Five elements in Figure 8.1a, consists of the standardized facial contours of the face, eyes, pupils, ears, nose, lips, and throat, while the second picture consists of only the face outline and neck line; and the third picture is drawn freely on blank paper. Hamilton (2008) analyzed the first and second pictures with the FEATS. In addition, Mattson (2012) evaluated the revised FSA by incorporating computerized techniques to analyze the color and free space. Since we find difficulty with using current computer technology to analyze the second and third pictures of the FSA, we develop a computer system to analyze the first picture. Actually, the first picture alone can be used as an art therapy assessment. Betts (2003) proposed the following five elements for evaluation in the first picture of FSA: (1) Whether a drawer’s motor skill permits coloring inside the lines (motor skill). (2) Whether a drawer is able to use natural colors - realism (natural color) and whether the picture can be evidence of recognized as a human face (face recognition). (3) Whether other components such as hair and jewelry are added (components addition). (4) Whether a drawer adjusts the face to look like the client and what this might indicate regarding self-perception - race and gender should also be considered here (self-perception). (5) Whether a drawer fills in the background and whether there is color differentiation between the face and background (space usage). Since element (4), with the exception of race, cannot be automatically evaluated with current computer technology, we omit this element and consider race in the element of natural color. We then divide element (2) into two elements of natural color and face recognition. Thus, we consider the following five elements: (1) motor skill, (2) natural color, (3) face recognition, (4) components addition, and (5) space usage. We call the system considering these elements and evaluating them by computer as in the next section the Computerized Face Stimulus Assessment (c_FSA). 8.2.2 Algorithms and criteria for each element Among the elements to be measured in the FSA, one is the drawer’s motor skill of coloring inside the lines, which corresponds to the accuracy in the

126

PART ONE ART EVALUATION

Figure 8.1 First picture and critical areas of the FSA.

SMC introduced in Chapter 2. Another element in the FSA is the use of picture space, which corresponds to the completeness in the SMC. However, we develop another computer algorithm to analyze these elements. The algorithm divides the first picture of FSA into 16 areas as shown in Figure 8.1b. The second, third, and fourth elements compare the main color of the designated area with its neighbor’s main color. The first and fifth elements additionally analyze the ratio of color usage. The elements are graded by the number of components which are satisfied with each element condition. Figure 8.1a is colored with 12 color markers: red, orange, yellow, light-green, green, sky-blue, blue, purple, violet, light-brown, brown, and black. We divide the sheet of a FSA into the following 16 areas, as shown in Figure 8.1b: (1) pupils, (2) around the pupils, (3) lips, (4) around the lips, (5) nose, (6) cheek, (7) around the cheek, (8) neck, (9) neck line, (10) around the neck, (11) forehead, (12) hair, (13) background, (14) earrings, (15) eyebrows, and (16) around the eyebrows. The C_CREATES (Kim, Bae, & Lee, 2007; Kim, 2010) recognizes the most frequently used color in each area, the ratio of colored area to the total area, and the difference in the color between the areas. In this evaluation, the most frequently used color in each area is defined as the principal color of that area. Now we describe how the five elements are evaluated in grade. 8.2.3 Elements in the Computerized Face Stimulus Assessment (c_FSA) Motor skill Motor skill shows whether or not the drawer accurately colors within the given line. We evaluate this element in terms of whether the drawer accurately colored the following four components: pupils (area 1), lips (area 3), cheek (area 6), and neck (area 8), which are adequate for measuring motor

Chapter 8 A Bridge from Part One to Part Two

127

_____________________________

Table 8.1 Number of components Criteria for grading Grade motor skill, natural _____________________________ colored accurately color, face recognition, and components addition 0 None 1 One 2 Two 3 Three 4 Four _____________________________

skill. For example, pupils are considered to be colored accurately when the principal color of the pupils (area 1) differs from that of around the pupils (area 2). The computer algorithm for the evaluation of motor skill is as follows: Step 1. Count the number of pixels colored in the pupils (area 1) using the C_CREATES. Step 2. If the percentage of colored area to the total pupils (area 1) is less than a certain threshold (in this paper, 70%), we consider the pupils to be not colored accurately and go to step 5. Step 3. Find the principal colors of pupils (area 1) and around the pupils (area 2). Step 4. If the principal colors of the pupils (area 1) and the surrounding area (area 2) are the same, then we consider that the pupils are not colored accurately. Otherwise, we consider the pupils to be colored accurately. Step 5. The lips (area 3), cheek (area 6), and neck (area 8) are evaluated in the same way. Step 6. Decide a grade according to the number of components colored accurately, as shown in Table 8.1. The criteria for grading motor skill are as follows: the grade is 0, when no component is colored accurately, 1, when one component is colored accurately, 2, when two components are colored accurately, and so forth, as shown in Table 8.1. Natural color indicates whether the components comprising the face are Natural color colored with the appropriate color. We evaluate this element on the following four components: pupils (area 1), lips (area 3), nose (area 5), and cheek (area 6). The appropriate colors for each component are assumed to be already known and the color is compared with the principal color of each area. For example, the lips are considered to be appropriately colored when the main colors of the lips (area 3) are black, brown, or red. As the computer algorithm for the evaluation of the natural color is similar to that of motor skill, it is not given here. The criteria for grading natural color are shown

128

PART ONE ART EVALUATION

Figure 8.2 Samples of the FSA.

in Table 8.1. Face recognition Face recognition indicates whether the given outline is recognized as a human face. We evaluate this element based on whether the drawer recognized the following four components which are the basic components of the human face: pupils (area 1), lips (area 3), cheek (area 6), and neck (area 8). The computer algorithm for the evaluation of face recognition is not given here as it is similar to that of the evaluation of motor skills. The criteria for grading face recognition are shown in Table 8.1. Components Components addition indicates whether other components have been added addition to the components of pupils, lips, nose, etc. We evaluate this element on whether the drawer added the following four components: hair (area 11 or 12), clothes (area 8), earrings (area 14), and eyebrows (area 15). The computer algorithm for the evaluation of the components addition is not given here for the same reason. The criteria for grading components addition are shown in Table 8.1. Space usage Space usage shows how much space the drawer uses. This element is evaluated through the space that has been used in the background (area 13). The computer algorithm for the evaluation of space usage is as follows: Step 1. Count the number of pixels colored in the background (area 13). Step 2. Calculate the percentage of colored areas to the total background (area 13). Step 3. Decide a grade according to the percentage of space used. The criteria for the grading of space usage are as follows: the grade is 0 when the background used is 0 - 5 %, 1 when it is 6 - 25 %, 2 when it is 26 - 50 %, 3 when is 51 - 75 %, and 4 when is 75 - 100 %.

Chapter 8 A Bridge from Part One to Part Two

129

__________________________________________

Table 8.2 2 Rater-1 Rater-1 Rater-2 QWK values ( κ ) Element : Rater-2 : System : System

__________________________________________ 1 2 3 4 5

0.957 0.854 0.980 0.951 0.941 0.891 0.951 0.819 0.977 0.960

0.858 0.970 0.907 0.822 0.958

__________________________________________ 8.2.4 Reliability and validity of the c_FSA The FSA samples were collected from 54 persons aged 60 - 90 with mild to Samples serious dementia and from 20 undergraduate students. Two human raters, one art therapist and one graduate student graded the five elements in the FSA. The computer system also graded them. Figure 8.2 shows the four pictures selected from the samples showing the comparisons of grades between the two human raters, and between the human raters and the computer system. Figure 8.2A and Figure 8.2B show examples of identical grades between both the raters and the computer system. All elements shown in Figure 8.2A were evaluated as 0 by both the human raters and the computer system. Components addition of Figure 8.2B was graded by both the human raters and the computer system as 3, and the remaining elements as 4. Figure 8.2C shows an example of different grades between the 2 human raters. While Rater-1 decided hair was added, Rater-2 decided hair was not added, showing that people can evaluate the same picture differently. The consistent outcome by the computer system is expected to be a great help to overcome this problem. Figure 8.2D is an example of the different grades between both the human raters and the computer system. While both the human raters evaluated the face as being colored accurately, the system evaluated the face as not being colored accurately. This was due to the fact that humans evaluated the picture considering light and shade, but the system evaluated the picture based only on the ratio of the colored area to the total area. The consistency of the evaluations between the two human raters was measured by Quadratic Weighted Kappa (QWK) values. The first column of Table 8.2 shows the values between Rater-1 and Rater-2. The high values ranging from κ2 = 0.941 to 0.980 show strong consistencies between the two raters. The QWK values between human Rater-1 and the system, and between human Rater-2 and the system are shown in the second and third columns of Table 8.2, respectively. The values ranging from κ2 = 0.854 to 0.958

130

PART ONE ART EVALUATION

Figure 8.3 Six sample PPAT drawings.

show strong consistencies between the raters and the computer system and thus demonstrate the applicability of the system.

8.3 Application of the evaluation results in Part One to the interpretation in Part Two 8.3.1 Relationships between the space usage in the PPAT and severity, and degree of dementia Now we exhibit how the system which evaluates the element of space usage developed in Chapter 7 can be applied to the psychological state of a real case. Usually we may have a hypothesis that space usage is related to energy, some psychological states or disorder. In this section, we investigate the relationships between the space usage and the severity group and between the space usage and the degree of dementia. For the application, 58 PPAT samples were collected from elderly persons with suspected dementia in a sanitarium. Space usage was evaluated in grade and rank by the system as in Chapter 7 and the degree of dementia was measured by the MMSE-K (Kwon & Park, 1989). The averages of the number of colored girds, area of colored convex hull, and number of clusters were 170,300 with standard deviation of 179,609, 614,573 with standard deviation of 279,757, and 74.3 with standard deviation of 54.51, respectively. The dementia severity of

Chapter 8 A Bridge from Part One to Part Two

131

________________________________________________________________________

Table 8.3 Area Number Area Grade Grade Information on the six Figure MMSE-K sample PPAT drawings colored of grids of convex hull by human by system

________________________________________________________________________ 8.3A 8.3B 8.3C 8.3D 8.3E 8.3F

2 26,741 7 26,263 12 99,552 15 1,593,554 21 225,625 27 235,050

0 58,346 0 143,086 4 544,085 10 642,367 7 1,117,670 12 768,732

1 1 2 2 3 3

0.67 0.81 1.78 2.40 3.12 2.82

________________________________________________________________________ a person who scores 24 - 30 is classified as “definitely normal,” one that scores 0 - 10 as “definitely having dementia,” and one that scores 10 - 23 as “having potential dementia.” Figure 8.3 shows six PPAT drawings among the samples and Table 8.3 provides various information. To investigate the relationship between the space usage and the severity and the degree of dementia, we used two statistical methods. (Severity of dementia classifies patients into three groups depending on the degree of dementia. There are 30 degrees of dementia.) One statistical method is the ANOVA of one-way classification which attempts to test whether several populations classified on a criterion have different averages or not (Walpole & Myers, 2006). We classified the 58 samples into three populations of “definitely normal,” “having potential dementia,” and “definitely having dementia” based on the MMSE-K scores corresponding to 24 - 30, 11 - 23, and 0 - 10, respectively. The respective sizes of the three populations were 12, 43, and 3. We tested the hypothesis that the three populations have the same average evaluations of grade and rank. The values of the test statistic having F distribution with 2 and 55 degrees of freedom were obtained as f = 6.169 and 5.981, whose corresponding p-values were 0.004 and 0.005 in grade and rank, respectively. Thus, we conclude that the space usage evaluations in grade and rank are quite different depending on the severity groups of dementia and are as a result, useful in differentiating them. The other method is a correlation analysis which attempts to measure the PCC and RCC strength of relationships between two variables by means of a single number called a correlation coefficient. We found that the Pearson’s Correlation Coefficient (PCC) between the space usage evaluation in grade by the system and the degree of dementia was Rp = 0.477, which corresponds to the p-value of less than 0.0001. Also, we found the Spearman’s Rank Correlation Coefficient (RCC) between the space usage evaluation in rank by the system and the degree of dementia was Rs = 0.502,

132

PART ONE ART EVALUATION

Table 8.4 ________________________________________________________________________ Correlation analysis Element 1 Element 2 Element 3 Element 4 Element 5

________________________________________________________________________

MMSE

PCC (Rp) .328 .432 .387 .381 .114 p-value .015 .001 .004 .005 .414

________________________________________________________________________ which corresponds to the p-value of less than 0.0001. Both correlation coefficients show statistically significant relationships between them.

8.3.2 Relationships of the five elements in the c_FSA Correlation between We analyzed the correlation between the grades of the five elements evalelements and degree uated by the computer system and the degree of dementia scored by the of dementia MMSE-K. The Pearson Correlation Coefficients (PCCs) are shown in Table 8.4. With the exception of space usage, all elements showed statistically significant positive correlation at the significance level of below 0.05. As the grade of element is increased, the degree of dementia is decreased. Differentiation Samples were divided into two groups depending on the severity of demenbetween normal and tia. A person who scores 0 - 19 was classified as one group and all other dementia groups people were classified as another group. By applying factor analysis, we examined whether or not each element differentiates the two groups. The ANOVA table for the factor analysis is shown in Table 8.5. Among the five elements, face recognition and space usage differentiated the two groups at the significance level of below 0.05. Regression analysis By applying stepwise regression, the probability that a person belongs to Table 8.5 ANOVA table

________________________________________________________________________ Element

Sum of Degree of Mean Significance Squares Freedom Square F Level

________________________________________________________________________ 1 2 3 4 5

Between Groups 1.021 Within Groups 109.627 Total 110.649 Between Groups 2.106 Within Groups 138.935 Total 141.041 Between Groups 9.074 Within Groups 85.520 Total 94.595 Between Groups 1.436 Within Groups 63.064 Total 64.500 Between Groups 6.060 Within Groups 77.953 Total 84.014

1 1.021 .671 72 1.523 73 1 2.106 1.091 72 1.930 73 1 9.074 7.640 72 1.188 73 1 1.436 1.640 72 .876 73 1 6.060 5.597 72 1.083 73

.415 .300 .007 .204 .021

________________________________________________________________________

Chapter 8 A Bridge from Part One to Part Two

133 Table 8.6 Regression results

_____________________________________________________________________________________________ Non-standardized Standized Variables ________________________ coefficient Coefficient Standard (beta) error

Value of t

Significance level

Sample average (standard deviation)

_____________________________________________________________________________________________ Constant .907 .098 9.233 .000 Element 3 -.135 .046 -.315 -2.912 .005 1.73(1.14) Element 5 -.125 .049 -.274 -2.538 .013 0.42(1.07)

_____________________________________________________________________________________________ a dementia group was estimated. The dependent variable was defined as for dementia follows: a group having a person who scores 0-19 was classified as 0, and all others were classified as 1. Independent variables were defined as grades of five elements evaluated by the computer system. The results are shown in Table 8.6. Among the five elements, face recognition and space usage were selected as variables to “well” explain the dementia. The regression function was obtained as Probability of dementia = 0.907 - 0.135 (- 0.315) x Face recognition [0.005] - 0.125 (- 0.274) x Space usage [0.013]. The numbers in parentheses are standardized coefficients and the numbers in brackets are significance probabilities. For example, as the grade of face recognition is increased by 1, the probability of dementia is decreased by 0.135. Also, as it is increased by 1 standardized deviation (1.14), the probability of dementia is decreased by 0.315 standardized deviation (0.315 x 0.49). The relative effect of face recognition (0.315), which affects the probability of dementia, is greater than space usage (0.274). A negative regression coefficient means that as the grade increases, the probability of dementia decreases. The Coefficient Of Determination (COD) was obtained as R2 = 0.174. 2 R is low since the dependent variable is an indicator variable. We can use the regression function as an estimate of the probability of dementia.

8.4 Conclusion The purpose of this chapter is to relate the information of art evaluation obtained in Part One with its use in the decision-making of art interpretation in Part Two. As a case study of considering only one element, the element evaluation of space usage in a PPAT drawing in Chapter 7 is used to analyze its relationships to the severity groups and the degree of dementia of the person who drew it. From 58 PPAT samples from elderly persons with

134

PART ONE ART EVALUATION

suspected dementia we concluded that the space usage evaluations in grade and rank are quite different depending on the severity groups of dementia at the significance level of below 0.004, and thus are useful in differentiating them. Correlation coefficients show strong relationships between the grade and the degree of dementia, and also between the ranks and the degree of dementia scored by the MMSE-K at the significance level of 0.0001. As a case study of several elements, we have developed a computerized system, the c_FSA, for the evaluation of the five elements in the first picture of the FSA. We used 54 drawings by people with suspected dementia and 20 drawings by undergraduate students. Two raters graded the five elements. The inter-rater reliabilities measured by the QWK values were κ2 = 0.941 - 0.980, which show high consistency in their gradings. Also, the system graded the five elements. The QWK values between the raters and the system were κ2 = 0.854 - 0.958, which confirm the applicability of the system. In the correlation analysis between the elements and the MMSE-K, four elements except space usage showed statistically significant positive correlation at the significance level of below 0.05. In the factor analysis, face recognition and space usage differentiated the normal group and the dementia group. In regression analysis for estimating the probability of dementia, face recognition and space usage were selected as independent variables. In this case of considering the five elements to estimate the probability of dementia, R2 was 0.174. As we will see in Chapter 15, when we consider the elements of FEATS as possible independent variables, line quality is selected as the variable to estimate the MMSE-K, and R2 is 0.390 (Kim, Kang, Chung, & Hong, 2012). When we consider all the elements in the C_CREATES, it is expected to produce a much better result of R2. This system can be applied not only to dementia but also to other psychological disorders. Also, the approach to computerizing existing art therapy assessments can be applied not only to the FSA but also to other art therapy tools. The validity of the FSA in understanding creative capabilities, cognitive functioning, and development levels is yet to be determined (Brooke, 2004). The objective and quantitative evaluation of the c_FSA will promote the usage of the original FSA and the study of the validity of the FSA. The c_FSA together with several other computer methods or systems developed in Part One can contribute to the objectification and quantification of subjective element evaluation in previous methods and thus provide information for art interpretation in an accurate, precise, and detailed way.

PART TWO ART INTERPRETATION

An Expert System Approach to Art Interpretation

Chapter 9 Abstract The expert system developed in this chapter is a prototype model to demonstrate the possibility of applying an artificial intelligence technology to art therapists’ decision-making on art interpretation in Part Two. The knowledge on art evaluation in Part One is the basis for art interpretation. However, element evaluation of a drawing is only a factor in art interpretation, and knowledge on drawer’s other various factors that may affect the decision-making need to be considered. For a computer to handle knowledge, abstract knowledge needs to be converted into a written form, during the process of which knowledge and standards become more clearly defined and objectively managed. The objective knowledge representation process of the expert system is one of the most significant and obvious advantages of using computer systems in art therapy. In the proposed expert system, similarities and contradictions of a mass of knowledge can be thoroughly reviewed, evaluated, classified, and organized to improve the quality of the knowledge base.

9.1 Various factors considered in art interpretation The purpose of Part Two is to interpret drawings and thus to gather infor- Difficulty mation about the psychological states of the drawers. This is based on the theory that the drawer unconsciously divulges his or her inner state of mind or feelings in drawings. The main difficulty in art interpretation is the fact that drawers’ personal and cultural background may affect their choice of shapes, colors, and styles in their drawings. Each person expresses an identical theme in a different manner. This implies that even seemingly identical drawings may require different interpretations and conclusions. For example, the observation that a significant number of fashion-conscious Parisians tend to wear black and grey tones can be viewed as a result of social milieu, as they are the colors that match the surroundings of an old city. On the other hand, most Koreans share a social perception that black attires are reserved 137

138

PART TWO ART INTERPRETATION

for funerals. Thus, a French child residing in Paris using black in drawings should be differently interpreted from a Korean child living in Seoul using black. Overlooking or neglecting these various personal or cultural factors and failing to reconcile seemingly contradictory elements can increase the risk of reaching inaccurate or misleading conclusions about the drawers’ psychological states. Since individuals’ personal, educational, social, and cultural experiences vary from one another, a psychological interpretation of art needs to employ not only a common denominator of a general category, but also an exclusive individual numerator. Any projective features of a drawing may not only be the results of certain psychological states of the drawers, but also be the results of a number of extrinsic elements. The latter includes cultural environments, such as values, languages, traditions, and behavioral patterns of the society in which the individual has grown up or currently resides. All aspects of an individual’s experiences and psychology are related to, influenced by, and are formed by culture (Coseo, 1997). Sentiments of human beings may be similar to one another, but that their expressions of the sentiments may vary depending on their cultural backgrounds. Thus, much caution is needed in applying predetermined methods of psychological interpretation and treatment to patients from different cultures (Wegmann & Vusenbrink, 2000). Numerous studies have warned against using only projective methods to identify abused, neglected, or indifferently-treated children (Veltman & Browne, 2000a, 2000b). Consideration of all possible supplementary materials is required in art interpretations in order to avoid exclusively relying on a singular element in a singular drawing (Veltman & Browne, 2003). Also, many studies have pointed out the problem of conflicting standards in art therapy (Hoshino, Silbert, Knapp, & Weaver, 1998). Veltman and Browne (2003) tested the consistency of art interpretation by comparing interpretation results of two professional psychotherapists using 72 samples of Kinetic Family Drawing (KFD) (Burns & Kaufman, 1972) drawings. The outcome, converted into the Cohen’s kappas of qualitative analysis, was between κC = 0.25 and 0.47, falling within the category of “standard (0.21 - 0.60)” but far less than that of “very high (over 0.81)” or “high (0.61 - 0.80).” This is not a satisfactory accuracy range, although it cannot be totally discarded (Cicchetti & Sparrow, 1982). A soloution An expert system for art therapy is proposed as a solution, at least in part, to the problems discussed above. Expert systems attempt to allow computers to substitute for human experts by endowing computers not only with experts’ knowledge and experience, but also with the ability to think and make decisions like a human being. Expert systems are one of the most productive and promising fields of artificial intelligence (Giarratano & Riley, 2005). In an expert system, the process leading to decision-making is reviewed and verified, and the decisions reached are assessed in order to make certain that

Chapter 9 An Expert System Approach to Art Interpretation

it has produced reliable conclusions (Raggad, 1996). Knowledge in art therapy is largely empirical, heuristic, and subjective. Art therapists rely on their professional expertise and experience, which are not always amenable to organized algorithms. In this sense, art therapy is analogous to such areas as economic demand forecasting or judicial sentencing in criminal cases, which expert systems designate as “ill-structured paradigm.” The use of artificial intelligence in these areas has already been in progress as in computer-assisted uniform administration of sentencing structure (Kim, Kim, Lee, Kim, & Baik, 1992). An expert system for art therapy has also been recognized as a useful method to systemize, organize, and classify various kinds of knowledge (Giarratano & Riley, 2005). It can thus provide a powerful tool to overcome the problems of complexity and subjectivity in art therapy. Some practicing art therapists are skeptical about the usefulness of computer-oriented technical tools (Fryrear & Corbit, 1992; Hartwich & Brandecker, 1997). They questioned the introduction of computer technology in the field, believing that computers may obliterate human roles in art therapy with machines replacing human sentiment and intuition. However, these concerns are largely ill-founded and unjustified. As already emphasized in previous chapters, the computer systems are not proposed to replace the existing art therapy practice, but it is rather intended to supplement and reinforce art therapy by introducing objective standards in a systematic manner. The frame of the expert system discussed below demonstrates how computer-aided technology can enhance the significance and efficiency of art therapy.

9.2 An expert system for art interpretation 9.2.1 System facilities Figure 9.1 shows the frame of an expert system for art interpretation developed in this chapter. Through its knowledge acquisition facility, knowledge is acquired, converted into a form amenable to computers, and accumulated into the knowledge base. The inference facility utilizes this knowledge to answer questions, diagnose (interpret) symptoms, and consult patients. The explanation facility provides explanations on how a certain answer, diagnosis or consultation result has been derived. The user interface of the system allows even non-experts to use and communicate with the system with relative ease. Finally, the self-learning facility makes it possible for the system to grow more intelligent with each use, as more knowledge is accumulated and refined. 9.2.2 Knowledge The system integrates expert knowledge in art interpretation gathered from

139

140

PART TWO ART INTERPRETATION

Figure 9.1 The frame of the expert system for art therapy.

existing research results in publications and bibliographies. Since human knowledge is immense and researchers use various terminologies and expressions for the same fact, it is necessary to classify and organize knowledge systematically. Note that the knowledge base is composed not only of elements of a drawing but various factors affecting the interpretation. The proposed system’s computer database functions are capable of efficient knowledge management. The knowledge structure of the system consists of four main factors: elements of a drawing, personal environments, psychological symptoms, and psychological disabilities. Each of these factors is subdivided into a number of elements as shown in Table 9.1, which is incomplete, and thus can be modified whenever desirable. The categorizations presented in the system are standards to be used as a means to classify, organize, and systemize knowledge. They are not mutually exclusive, but, in fact, overlap in many cases. Knowledge acquired through these four main factors is converted into a rule-based IF - THEN form. This type of representation is most common for expert systems (Rowe, 1998). Since it is difficult to draw definite correlations between various kinds of knowledge in art therapy, the IF - THEN

Chapter 9 An Expert System Approach to Art Interpretation

________________________________________________________________________ Main-factor

Sub-factor Classification / example ________________________________________________________________________ Elements Theme of a drawing

Kinetic or static Subject-given or subject-free / person, automobile, train, ship plane, house, flower, doll, tree, rain, selfportrait, animal, school, etc. Line Direction / vertical, horizontal, diagonal, or circles Strength of brush strokes / strong or weak Color Primary / secondary / red, orange, yellow, green, blue, purple, black, brown, etc. Feeling / warm/ cold colors Number / various or limited Composition Placement / upper left, lower left, upper right, or lower right Space usage / empty or full ________________________________________________________________________ Personal Age, gender environments Personal relations Periodical

Years / boy or girl Parents, siblings, relatives, friends, teachers, etc., Overprotection, negligence or abuse Accidents, trauma, abuse or events / death of a parent, birth of a younger sibling, etc. Spatial Home, school, and work place Method of education Status Physical Disabilities Pathology Psychological Disorders, attitude, stress, pride, emotional stimuli, enforcement, induction of inferiority, lack of understanding ability, etc. Social Nationality, cultural background, religion, etc. ________________________________________________________________________ Psychological Intellectual symptoms Emotional

Underdevelopment or degradation Dejection, jealousy, shyness, excitement, desire, nervousness, anxiety, ostentation, expansive delusion, etc. Physical / Disabilities, attitude, stress, pride, emotional stimuli, physiological enforcement, induction of inferiority, lack of understanding ability, etc. Social Evasion of personal relationships, mother complex, Oedipus complex, independent character, dependent character, domination, evasiveness, narrow-mindedness, selfishness, etc. ________________________________________________________________________ Psychological Mental retardation Mile, moderate, severe, or profound disorders Learning Reading, mathematics, written expression, or learning Motor skills Developmental coordination Communications Expressive, receptive-expressive, phonological, stuttering, or communication Developmental Autistic, Rett’s, childhood disintegrative, Asperger’s, pervasive ________________________________________________________________________

141 Table 9.1 Main factors of the knowledge structure

142

PART TWO ART INTERPRETATION

Table 9.1 ________________________________________________________________________ (continued) Main-factor Sub-factor Classification / example

________________________________________________________________________

Attention-deficit Hyperactivity d i s r u p t i v e be- Conductor havior Oppositional defiant Disruptive behavior Feeding and eat- Pica ing Rumination Tic Tourette’s Chronic motor or vocal Transient Elimination Encopresis Enuresis Others Separation anxiety, selective mutism, reactive attach ment, stereotypic movements, etc.

________________________________________________________________________ form is regarded as more suitable than forms that use a tree, network, or frame. A sample set of knowledge is listed in Table 9.2. Each knowledge is assigned a classification code: S (Standard diagnosis), I (Individual diagnosis), P (Psychological diagnosis), and F (Feedback). It is linked to the class and subclass numbers of factors, which together form a serial number. For instance, P.2-1-3 refers to the third knowledge in Psychological diagnosis related to the first subclass of the second class of the factors. 9.2.3 Reasoning process Numerous studies have attempted to connect a particular element of a drawing with a certain psychological symptom. For example, Kim (1988) claimed that when a child chooses a ship as a theme of his drawing and uses purple as a major color, the general clinical conclusion is that such elements imply the desire to express outwardly of a physical illness. We understand that this specific assertion may not be widely accepted in the field. Whether or not to include a certain kind of knowledge into a system’s knowledge base is up to the experts who develop the system. Expert systems will differ depending on who builds the system. Consideration of the various degrees of knowledge uncertainty (Gordon & Shortliffe, 1984) is beyond the scope of this chapter. Many studies have assumed that the cause and the effect correspond on a one-to-one basis, but many art therapists have acknowledged the limitations of this assumption, as illustrated in the study by Alschuler and Hattwick (1947). A drawing is a product of complex effects of numerous factors. If the child mentioned above is a well-educated sixth grader, the fact that he still likes to draw ships at this age may imply the possibility of a maternal complex, and the fact that he mostly uses the color purple may indicate the extraordinary artistic sense of this child. In this manner, interpretations and conclusions, even for an identical theme or color, can differ

Chapter 9 An Expert System Approach to Art Interpretation

143

Table 9.2 A part of the knowledge in the system

_____________________________________________________________________________________________ Code

IF-THEN rule _____________________________________________________________________________________________ [S.1-1-1]

IF large or full size, THEN hyperactivity, activation, impulsivity, defensiveness, expansion of selfego, expansive delusion. [I.1-1-1] IF large or full size and age of over 7, THEN the causes of [S.1-1-1] are intellectual underdevelopment. [I.1-1-2] IF large or full size, and juvenile, THEN the causes of [S.1-1-1] are inferiority, inaptness, frustration, compensation for disappointment. [I.1-1-3] IF large or full size and adult, THEN the cause of [S.1-1-1] is manic. [S.1-1-2] IF small size, THEN shyness, inferiority, inaptness, lack of self-confidence, contraction, melancholy, stress, loneliness. [S.1-2-1] IF placement skewed to the left, THEN impulsivity, desire to change, extrovert. [S.1-2-2] IF placement skewed to the right, THEN stableness, introvert. [S.1-2-3] IF placement skewed to the upper-left, THEN instability, degradation, contraction, unrest. [S.1-2-4] IF placement skewed to the upper-right, THEN unpleasant experience, optimism, illusion. [S.1-2-5] IF placement skewed to the lower-left, THEN melancholy. [S.1-2-6] IF placement skewed to the lower-right, THEN despair. [S.1-3-1] IF strong brush-stroke, THEN tension, unrest, paranoia, antisocial, epilepsy, retardation, contraction, stiffness, dogmatic, aggressiveness, impulsivity. [I.1-1-4] IF strong brush-stroke and age under 6, THEN [S-3-1] not-applicable. [S.1-3-2] IF weak brush-stroke, THEN timidity, indecisiveness, fear, instability, unrest, melancholy, helplessness, terror, neurosis, schizophrenia. [I.1-1-5] IF weak brush-stroke and age under 6, THEN [S-3-2] not-applicable. [S.1-3-3] IF both strong and weak brush-stroke, THEN adaptability, applicability. [I.1-1-6] IF both strong and weak brush-stroke and age under 6, THEN [S-3-3] not-applicable. [S.1-3-4] IF zigzag stroke, THEN hostility. [S.1-3-5] IF straight stroke, THEN dogmatic, promptitude, decisiveness lack of adaptability, impulsivity. [S.1-3-6] IF short stroke, THEN impulsivity, excitement. [S.1-4-1] IF no head, THEN neurotics, disturbance, delusion. [S.1-4-2] IF large head, THEN compensation for anxiety. [I.1-1-7] IF large head and age under 6, THEN [S.1-4-2] not-applicable. [S.1-4-3] IF small head, THEN passiveness, inhibition, contraction. [S.1-4-4] IF triangle or rectangle head, THEN intellectual disability, disturbance, neurotic. [S.1-4-4] IF head emphasis, THEN abnormal brain wave, epilepsy. [S.1-4-5] IF hair emphasis, THEN dogmatic, defensiveness, hysteric. [S.1-4-6] IF large eyes, THEN sensitivity. [S.1-4-7] IF small eyes, THEN contraction. [S.1-4-8] IF eyes emphasis, THEN unrest, tension, suspicion, defensive, paranoia. [S.1-4-9] IF no ears, THEN dissociation. [S.1-4-10] IF large ears, THEN sensitivity, nervousness. [S.1-4-11] IF small ears, THEN unrest, nervousness, sensitivity, distrust, defensive. [S.1-4-12] IF no mouth, THEN frustration, inability, contraction, lack/difficulties of personal relationship, confliction. [S.1-4-13] IF large mouth, THEN fear of meeting people, fear. [S.1-4-14] IF small mouth, THEN shunning people/ society. [S.1-4-15] IF tongue emphasis, THEN enuresis, sexual-problem. [S.1-4-16] IF black nose, THEN caught-cold. [S.1-4-17] IF black arm/hands, THEN guilty. _____________________________________________________________________________________________

144

PART TWO ART INTERPRETATION

Table 9.2 (continued) _____________________________________________________________________________________________ Code

IF-THEN rule _____________________________________________________________________________________________ [S.1-4-18] [S.1-4-19] [S.1-5-1] [S.1-5-2] [S.1-5-3] [I.1-1-1] [I.2-1-1] [I.2-2-1] [I.2-6-1] [P.2-1-1] [P.2-1-2] [P.2-2-1] [P.2-2-2] [P.2-2-3] [F.1-2-1] [F.2-1-1] [F.3-1-1]

IF long neck, THEN caught-cold. IF clothes button emphasis, THEN dependent to mother. IF light primary color, THEN fear, timid. IF purple color, THEN illness, unhappiness. IF red secondary color, THEN liberalness, sociability, cooperativeness, happiness, hostility. IF drawing level 2 and age 6-8, THEN normal. IF no parents, THEN high probability of ~. IF bad relationship with people, THEN pervasive development disorder not otherwise specified (PDD NOS). IF 2 or more lower drawing- level, THEN pervasive development disorder not otherwise specified (PDD NOS). IF fear, tension, terror, hysteric lack of self-confidence, difficulties of personal relationship, defensiveness, degradation, uneasy, worry, obstinacy, THEN anxiety disorder. IF impulsivity, careless, diffuseness, THEN attention deficit. IF defensive, antisocial, anger, violence, self-ego, dependent, frustration, defensive, introvert, indecisive, THEN conduct disorder. IF bad relationship with people, THEN Attention Deficit Hyperactivity Disorder (ADHD). IF attention deficit, hyperactivity, THEN ADHD. IF high anxiety index, THEN anxiety disorder. IF cerebral embolism, THEN cause of learning disability. IF communication problems with parents, THEN cause of psychological disorder.

_____________________________________________________________________________________________

according to age and individual experience. A human expert is uniquely qualified to take into consideration complicated relationships between causes and effects. The system is designed to emulate this human reasoning process of interpretation. Figure 9.2 describes the four types of diagnosis (interpretations) in the system implemented after the human expert’s reasoning process model. Standard diagnosis Standard diagnosis links causes to effects, i.e., the elements of a drawing to psychological symptoms on a 1:1 basis. It is the process of extracting the common denominator. For example, an unusually large drawing corresponds to symptoms of hyperactivity, activation, impulsivity, defensiveness, expansion of self-ego, or expansive delusion classified as knowledge [S.1-1-1] in Table 9.2 (Shin et al., 2002). A “black sun,” which is a typical metaphor for darkness, gloom, fear, terror, death, revenge, or oblivion in poems or in myths, correlates to the psychological symptoms of gloom, destruction, immorality, obsession with crime, sin, and punishment. If a picture is skewed to the upper or lower, inner or outer, left or right, or side of the paper, it implies perversion of existing senses (Gregorian et al., 1996). A drawing that depicts sexual organs or a human figure without hands or fingers, or only the head implies symptoms of insecurity, inner complications,

Chapter 9 An Expert System Approach to Art Interpretation

145

Figure 9.2 A model of art therapy diagnosis process.

desire for compensation, and tendency of body mutilation (Trowbridge, 1995). Individual diagnosis is established after considering various factors of draw- Individual diagnosis ers’ personal environments. This is the process of extracting the individual numerators where the previous schematic 1:1 match is adjusted to reflect the drawer’s personal environments. Suppose that an individual has produced a drawing with large subjects. Here, consideration of the age factor can lead to several different conclusions. When the drawing was done by a child at age seven or older, a possible cause of his or her symptoms may be traced to intellectual undergrowth [I.1-1-1]; if drawn by a teenager, inferiority, feelings of unworthiness, discouragement, or disappointment [I.1-1-2]; and finally if drawn by an adult, manic [I.1-1-3] (Shin et al., 2002). Under the assumption that Draw A Person (DAP) (Goodenough, 1926) drawings reflect social status, Lev-Wiesel and Drori (2000) revealed that the drawings by widows and wives living with spouses differed in length and arrangement of subjects, but not in area. Widows with symptoms of fear of physical safety, anxiety of spinsterhood, fear of darkness, anxiety of harm, and apprehension of getting ill-treated by others tended to draw in one cor-

146

PART TWO ART INTERPRETATION

ner of the papers (Hammer & Piotrowski, 1997; Lopata, 1976, 1996). Those who had established marital relations tended to draw large drawings (Abraham, 1989). Those who allegedly had self-confidence tended to draw in the center (Wanderer, 1997). Psychological Psychological diagnosis relates psychological disabilities to the psychodiagnosis logical symptoms identified above. For example, symptoms such as fear, tension, terror, hysteric lack of self-confidence, relationship difficulties, defensiveness, degradation, uneasiness, worrying, or obstinacy are diagnosed as anxiety disorder [P.2-1-1]. Attention Deficit Hyperactivity Disorder (ADHD) is translated into the psychological symptoms of attention deficit and hyperactivity [P.2-2-3]. Further symptoms of impulsivity, carelessness, or diffuseness are diagnosed as attention deficit [P.2-1-2]. Hitherto, conclusions are drawn on a tentative level, claiming to be part of the comprehensive diagnosis. Feedback process Finally, the feedback process reviews the collective data of the various factors obtained. The three kinds of diagnoses above are compared, examined and evaluated in this stage, where necessary adjustments are made and the justifications for conclusions examined. For example, cerebral embolism in one’s backgrounds may be singled out as the cause of his learning disability [F.2-1-2]. The black sun mentioned above points to a child who may have been mentally injured from traumas such as war, hurricanes, plane crash, earthquakes, etc. (Gregorian et al., 1996). Sexual abuse can be detected through behavior review scores (Chantler, Pelco, & Mertin, 1993). The indexes of emotional disability, anxiety, brain damage, impulsivity, etc., are calculated and compared with correlating indexes of psychological aspects gleaned from various psychology assessments [F.1-2-1]. 9.2.4 Advantages of the system In the proposed system, knowledge is represented in a written form. Abstract knowledge is turned into specific and concrete knowledge to constitute a knowledge base. This representation process of the system has thus the advantage of providing a basis for transforming a subjective diagnosis into an objective one. Furthermore, new methods may be developed in the representation process to offer important grounds for theoretical developments in art psychotherapy, facilitating more scientific diagnoses. Apart from being capable of synthesizing expert opinions from numerous sources, the proposed system can also effectively reconcile conflicting pieces of information and organize them in a systematic manner. Where the color white signifies the birth of a new child, for instance, it can also mean death, depending on circumstances. According to Gregorian et al. (1996), white or colorlessness implies suppressed emotions or the end of a life; according to the Russian artist Kandinsky who was the pioneer of abstract art,

Chapter 9 An Expert System Approach to Art Interpretation

147

Figure 9.3 Difference in the strength of brush strokes.

in contrast, the color white stood for birth (Baraev, 1991). The system also has a number of practical advantages. It is operated on a PC and its functions can be easily utilized at low cost. Contemporary art therapy experts can benefit from the system’s various functions by using an assortment of simulations provided by the system. For instance, it is often the case that the standards are not clear when we input the elements of a drawing. Determining the degree of strength of a brush-stroke, for instance, may be extremely subjective, but this problem can be rather easily dealt with by using computer graphic functions. In Figure 9.3, the picture with strong brush-strokes was produced by adjusting the original picture with light brush-strokes (Shin et al., 2002) using the graphic functions of Photoshop 7.0 (Adobe Creative Team, 2002). Other computer functions such as data-base management, expression of visual colors, spreadsheet, etc., can prove useful. Since the system is fairly easy to use, non-professionals such as parents or teachers can readily use the system to their children’s advantage. The proposed expert system can thus both enhance the expertise of professionals and foster the training of novices, contributing to spreading the benefits of art therapy. 9.2.5 System features This system that was developed in 2006 (Kim, Ryu, Hwang, & Kim, 2006), Development with an architecture presented in Figure 9.4, consists of the following: environment (1) Language: Kernel Prolog (Tarau, 1999), ASP (Active Server Pages),

148

PART TWO ART INTERPRETATION

Figure 9.4 System development environment.

VBScript (Visual Basic Script); (2) Data-base: SQL (Structured Query Language) Server 2000; and (3) Web Server: Windows 2000, IIS (Internet Information Services) 5.0. The components can be replaced with updated versions, although the system works well to date. The use of IIS 5.0 offered by Windows 2000 prevents any excessive expenses. The Kernel Prolog was used to avoid the overloading of operating file resources on the web server and to ensure smooth exchange of data. It opens all Java sources related to the inference facility and provides Prolog Applet so that the client can easily operate it on the internet. In addition, the Kernel Prolog allows swift delivery of rules and facts in the knowledge base in the form of text files. ASP and VBScript were used to provide a more convenient interface when the user inputs related information or sees diagnosis results on the screen. The SQL Server 2000, a relational data-base management system, was used to manage personal data and other information, and was connected through ODBC (Open Data Base Connectivity) (Sanders, 1998). Software menu The software has the following menu: (1) Introduction: research personnel, related research results, diagnosis process, and examples of diagnosis; (2) Data-base: add, edit, delete, and search information; (3) Input: elements of a drawing, personal environments, supplementary and other reference materials, etc.; (4) Diagnosis: relationship of causes and effects, standard diagnosis, individual diagnosis, psychological diagnosis, and feedbacks results; (5) Gallery: samples of various drawings with their diagnostic results;

Chapter 9 An Expert System Approach to Art Interpretation

149

Figure 9.5 A case study of a fiveyear-old boy with a psychological disorder.

(6) References: papers, journals, books, related web sites, and graduate schools; (7) BBS: participation and coordination of experts; (8) Knowledge base: save, search, edit, add, and update knowledge; and (9) Help: questions and answers, explanation of terms, and terminology dictionary

9.3 Case study Figure 9.5 is a drawing sample by a child who has been diagnosed with a psychological disorder (Shin et al., 2002). The child is six years old and has communication problems with his parents. Information is fed into a computer through the user interface. In Figure 9.5, a cellophane sheet overlaps with the left side of the drawing paper. The drawing takes up the entire space. We can gather the following elements from this drawing: (1) Full picture; (2) Skewed to the left side; (3) Big head; (4) Emphasis on hair; (5) Emphasis on eyes; (6) Big ears; and (7) Omission of mouth.

150

PART TWO ART INTERPRETATION

These 7 elements of the drawing lead to the corresponding knowledge in the system, and we can obtain a standard diagnosis. (1) Full picture: [S.1-1-1] hyperactivity, activation, impulsivity, defensiveness, expansion of self-ego, expansive delusion, excessive activity, activation tendency, impulsivity, expansion of self-ego, expansive delusion; (2) Skewed to the left: [S.1-2-1] impulsivity, desire to change, extrovert; (3) Big head: [S.1-4-2] compensation for anxiety; (4) Emphasis on hair: [S.1-4-5] dogmatic tendency, defensiveness, hysteria; (5) Emphasis on eyes: [S.1-4-8] unrest, tension, suspicion, defensiveness, paranoia; (6) Big ears: [S.1-4-10] sensitivity, nervousness; and (7) Omission of mouth: [S.1-4-12] frustration, inability, contraction, Lack/ difficulties of personal relationship or conflict. Specific personal environments are as follows: (1) Age: six; and (2) Communication problems with his parents. The standard diagnosis is adjusted according to these personal environments, invoking the corresponding knowledge in the following manner. (1) [I.1-1-7]: since the patient is a child, the psychological symptom in (3) in the standard diagnosis is removed; thus (2) The individual diagnosis results in the identification of psychological symptoms of (1), (2), (4), (5), (6), and (7) listed above. A tentative conclusion that can be drawn from the diagnosis of psychological diagnosis of symptoms invoking knowledge [P.2-1-2], [P.2-2-2], and [P.2-2-3] is that this child may have ADHD. In the feedback process, the background knowledge [F.3-1-1] that the child had communication problems with his parents further supports the likely diagnosis of ADHD. No further feedback exists in this simple case. Hence the interpretation. The entire process of establishing the diagnosis of ADHD is clearly and systematically documented by the system. The user interface shows the information with its codes invoked and the corresponding IF - THEN rules on the output screen. A professional therapist can review the output for its accuracy and the information provides important materials for determining the course of treatment. After all, the case study confirms the usefulness of the system in the early detection of a psychological problem. The system’s diagnosis should prompt the parent/teacher to seek a final professional diagnosis for the child in a timely manner.

9.4 Discussion

Chapter 9 An Expert System Approach to Art Interpretation

A careful and positive attention to the expert system can promote significant progress in art therapy. The intended use of the proposed system is, at present, limited to supplementing existing art therapy, enabling parents and teachers without professional training to detect their children’s psychological problems at an early stage so they can seek professional treatment in a timely manner. Further development of the system is, however, expected to provide more extensive and reliable results to meet the comprehensive needs of art therapy. This study is proposed as the first step in this development. Since art therapy has been regarded as particularly effective in treating younger patients, the current prototype expert system is limited to the children’s DAP and KFD drawings. Children’s allegedly small universe is more likely to be metamorphosed into certain subjects, colors, lines, and styles of their drawings, which can provide valuable information for understanding their psychology. Children draw not what they see but what they know (DiLeo, 1970). When children paint, they choose colors based on their feelings, not on the reality (Thomas & Silk, 1990). Children’s drawings thus tend to reflect rather precisely their mental and emotional maturity, current state of emotions, desires, fantasies, and pathological symptoms, as well as their surrounding environment such as family or personal relationships. Therefore, children’s drawings provide important clues for psychological diagnosis, and as a result, art therapy has developed into an important clinical discipline. Of course, the expert system can be extended to drawings by people at any age. The proposed expert system provides information on psychological state based on a single drawing. Comparison of several drawings produced by a drawer over a long period of time, which is essential for tracking psychological progress, is an aspect that needs yet to be integrated into the system. The interpretation process of the model in Figure 9.2 can be modified in any way thought to be more practical, reasonable, or appropriate depending on the expert’s view point. A different process will be presented in the next chapter.

151

Reasoning Process of an Expert System for Art Therapy

Chapter 10 Abstract Decision-making by human beings involves analyzing all factors concerning a given situation in a comprehensive and intuitive manner based on the decision maker’s experience and knowledge. Due to the inherent nature of human decision-making, the reasoning process from confronting the problem to finding a solution is too complicated to be explicitly represented. In this chapter, we implement an expert system that emulates the art interpretation process of art therapists. We model the complicated mechanism by dividing the process into several procedural stages and feedbacks. The system presents a suitable method of maintaining consistency among when making numerous decisions. The system has a learning facility that improves its intelligence. A case study demonstrates the usefulness and suitability of the proposed system.

10.1 Modeling human decision process Difficulty Under the projective hypothesis, numerous studies have been conducted on interpretation of drawings regarding drawers’ psychological states based on the evaluation of the elements in the drawing. However, the relationship between elements of a drawing and psychological states, symptoms, or disorders is extremely complicated. Many factors such as the drawer’s age, psychological or emotional maturity, economic situation or status, social or cultural background, and the environment where he or she was raised are hidden in the expressions of a drawing (Hanes, 1997). Such factors affect one another and interoperate in a complex manner. A certain factor in one individual does not always affect others in the same way, and can even operate in the opposite way depending on other factors. This is called interactive relations. In art therapy, therefore, the drawings cannot be interpreted on its own, but instead the implications of particular other factors associated with the individual must also be considered synthetically. 152

Chapter 10 Reasoning Process of an Expert System for Art Therapy

153

Naturally, numerous studies and methods in art therapy have produced results that are diverse, lacking in consistency and sometimes even contradictory to each other. This is because there exist few objective, conclusive, and systematic methods of interpreting drawings. Interpretation outcomes are often inherently subjective, largely based on the experience or knowhow of individual art therapists. Thus, art therapy can be classified as an Ill-Structured Paradigm (ISP) in the sense that its practice is hard to be systematically organized (Giarratano & Riley, 2005). To address this issue, we propose an art therapy expert system, which can A solution support art therapists with making art interpretation, as well as helping parents and teachers easily ascertain signs of psychological problems in their children or students. An expert system is a well-established field of artificial intelligence that incorporates human experts’ knowledge, experience, know-how, and decision process into a computer, thus allowing even non-experts to make decisions based on the captured expertise. This type of system is especially effective for ISPs such as art therapy. Expert systems have yielded the most promising results among many areas of artificial intelligence (Giarratano & Riley, 2005). They have proved beneficial in a wide range of applications such as the diagnosis of automobile defects, medical diagnosis and treatment (Clercq, Hasman, Blom, & Korsten, 2001), and sentencing guidelines in criminal cases (Kim et al., 1992). Hartwich and Brandecker (1997) used computer graphics as drawing tools for art therapy. Moving beyond the use of computer technology as mere audiovisual methods, we explore the application of artificial intelligence in art therapy, focusing on the prospect of having the machine think like a human. Several studies have been undertaken on the use of expert systems in art therapy, such as introducing the expert system approach to the field (Kim, Ryu, et al., 2006) in the previous chapter, formulating the structure of knowledge for the SMC (Kim, Kang, & Kim, 2009) in Chapter 11 and for the KFD (Kim, Han, Kim, & Oh, 2011) in Chapter 12, and suggesting a framework of knowledge base for art therapy (Kim, Yoo, Kim, & Lee, 2007). We hope that these studies will contribute to the development of art therapy. The most important task in developing an expert system for art therapy is to model human experts’ decision-making process so that it can be performed by a computer. Human beings can make instant decisions by synthetically and simultaneously considering multiple factors involved in a given situation. Complex relationships such as interactive relations can also be considered in an instant manner. It is difficult to articulate such a process in a logical sequence of events. Nevertheless, the substantiation of the reasoning process is an essential task that every expert system must achieve. In the proposed system, the decision process or reasoning process of an expert is divided into specific stages and the order of the stages is determined. A

154

PART TWO ART INTERPRETATION

tentative decision is made at each stage and there are feedback processes in which each decision is reviewed from the beginning stage. Compared with the model of Kim, Ryu, et al. (2006) in the previous chapter, this model is more systematic in deriving conclusions and thus presents a clearer method for art therapy experts to participate in refining the system. While the inherent characteristics of the ISP render it unavoidable to reach different decisions about identical situations, it is desirable that all decisions made are consistent. Human experts know the degree of reliability of decisions they make. Likewise, humans have the ability to accumulate knowledge as they gain new information or experience. We have developed an expert system that has the ability to evaluate the reliability of decisions and to draw consistent conclusions (interpretation). Our system also has the ability to increase its intelligence as it is utilized. However, it should be assumed that any conclusion reached by the system is only a supplementary aid to parents or art therapists. Final decisions should be made by human experts who have keen insights and sufficient experience of working in the field.

10.2 Process of diagnosis consisting of nine sub-processes 10.2.1 Requirements of art interpretation process in expert system Suppose a school counselor has to decide whether a student has psychological problems. The counselor tries to reach a conclusion by synthetically considering the student’s language, attitude, attire, and information concerning the student’s environment, such as friends or family background. It is highly likely for the counselor to be subjective as he or she might rely on his or her own experience. Thus, it is necessary to review the reasons and process leading to the conclusion that the student has psychological problems. Also, the counselor is expected to explain in a logical manner the reasons for and demonstrate the method employed in reaching such a conclusion. In art psychotherapy, the similar process provides the basis for theoretical and scientific development. The use of an expert system is proposed as a method of integrating this process. The detection of correlation between the elements of a drawing and the exhibited psychological states or symptoms is not simple. For example, let us say that we are to determine the presence of psychological problems based on whether or not a certain color scheme is excessively used and whether the placement on the paper is balanced or skewed. If we were to reach a conclusion based upon an excessive use of color alone, the presence of this element in a drawing would imply abnormality and the absence, normality. If we were to consider the placement alone, a balanced placement would imply normality while a skewed one would imply otherwise. However, if these two factors are considered simultaneously, a drawing that is

Chapter 10 Reasoning Process of an Expert System for Art Therapy

both excessive and unbalanced, or not excessive and balanced would imply normality, while a drawing that possesses one positive element and one negative would imply abnormality. Such interactive relations between two factors become exponentially more complex when additional factors are considered. For example, if the subject in the example above has received a significant level of education in art, or is gifted in art, all of the matters mentioned above are considered to be normal. Hence, such complex interactive relations between causes and effects are nearly infinite in art therapy. In this chapter, we use the expressions ‘the cause-and-effect relationship’ to describe their correlation. A drawing is a product of numerous factors operating simultaneously in an individual. The role of the knowledge engineer, an expert in the field of expert systems, is to model the complicated cause-and-effect relationships between the elements of a drawing and psychological states or symptoms, and the diagnosis process of a human expert into a computer. The diagnosis process does not merely involve choosing a method, but it is, in fact, a complex reasoning process that synthetically considers the correlations between a large number of factors and symptoms, consults the experience and knowhow of the past, and reaches a conclusion in a particular situation. To have a machine perform such a process, the reasoning processes of experts must be elicited through discussions with and among experts. The result is then formulated as much as possible in a written form. Ultimately, the reasoning process of an expert system must function like that of a human expert. The art interpretation mechanism of art therapy in the proposed system is composed of subdivided phases and feedback processes which operate in a certain order like a slow motion video. We propose another model of human reasoning process slightly different from the one presented in the previous chapter (Kim, Ryu, et al., 2006), which suggests the need for continuing refinement in the art interpretation model. Experts in art therapy are encouraged to participate actively to review, determine, and verify which of the existing models yields better results. The overall process consists of three stages of diagnosis and six levels of feedback. The diagnosis stages relate the effect with the cause on a 1:1 basis, and the feedback stages review the intermediate results from the diagnosis by relating the interactive relations of two or more causes. 10.2.2 Model of reasoning process The first phase of the mechanism involves extracting the relationships between elements of the drawing (size, composition, color, lines, etc.) and psychological states or symptoms. This information ultimately connects drawing elements and psychological symptoms on a 1:1 basis. This is called Standard diagnosis. For example, a drawing large enough to fill up the entire drawing sheet may imply the presence of symptoms such as hyperactivity, activation tendency, impulsiveness, offensiveness, inflation of self-ego,

155

156

PART TWO ART INTERPRETATION

and expansive delusion. A drawing far short of usual standard artistic skills may represent developmental disabilities. The influence of an environmental factor on drawing elements varies and can even have an opposite effect depending on each circumstance. Therefore, in the next phase, the factors which result in a deviation from the simple 1:1 relationships are identified. We define this phase as Feedback-1. The previous example shows such relationships, namely the interaction between color scheme and placement. The environment of an individual is taken into account to adjust the results from Feedback-1. This is subdivided into two phases, and the factors belonging to a wider category including sex, age, nationality, and cultural background, are considered first. This phase is called General diagnosis. For example, the tentative conclusion on artistic skills development is adjusted according to the age and symptoms of each subject. The complex interactive relationships are identified in Feedback-2 and Feedback-3. For example, partitions between family member figures in a KFD drawing would generally suggest lack of communication among family members in some cultures, but they can also imply a systematic or structural way of thinking in the case of a Taiwanese child (Wegmann & Vusenbrink, 2000). Next, specific environmental factors of an individual are considered. Such factors include family, friends, education, medical history, etc. This phase is called Individual diagnosis. If a patient is suspected of having difficulties in personal relationships, his or her relationship with family members is reviewed and the reliability of the conclusion is adjusted accordingly. Again, the interactive relations are identified and the factors of the previous two phases, General diagnosis and Standard diagnosis, are reviewed in Feedback-4, Feedback-5, and Feedback-6. Order of steps The diagnosis process model is summarized in Figure 10.1. The order of the steps in the art interpretation mechanism is as follows. (1) Standard diagnosis (2) Feedback-1. If adjustment is made, go to step (1), else next step. (3) General diagnosis (4) Feedback-2. If adjustment is made, go to step (3), else next step. (5) Feedback-3. If adjustment is made, go to step (1), else next step. (6) Individual diagnosis. (7) Feedback-4. If adjustment is made, go to step (6), else next step. (8) Feedback-5. If adjustment is made, go to step (3), else next step. (9) Feedback-6. If adjustment is made, go to step (1), else next step. (10) End. Example Let us examine the process of determining whether the level of artistic skills exhibited in a child’s drawing is appropriate for his or her age. According to the studies by Piaget (1959), the development of logical thinking is reflected

Chapter 10 Reasoning Process of an Expert System for Art Therapy

157

Figure 10.1 Another model of diagnostic human reasoning process.

in a child’s drawings. The ability to express realism represents the last stage of development (Golomb, 1999). In our expert system, we divide the artistic skills into eight levels, compared to the six levels provided in the studies of Lowenfield and Brittain (1982) and Malchiodi (1998). We present below the appropriate age, relevant explanations, and the upper and lower limits of age in parentheses and a standard drawing for each level in Figure 10.2. If the child’s age is below the lower limit, then the level of the artistic skills is considered to be abnormally/exceptionally high, and vice versa. Level 1: Age 3 and below, meaningless scribbles (-, age 3) Level 2: Age 4, identifiable shapes (age 3, age 5) Level 3: Age 5, in a human figure, one circle (head and body) and two lines (arms, legs) (age 3, age 6) Level 4: Age 6, in a human figure, classification of head, body, arms, legs, and hair (age 5, age 7) Level 5: Age 7, in a human figure, classification of eyes, nose, ears, mouth, and etc. in the facial area (age 6, age 10) Level 6: Age 8 - 9, three-dimensional, realistic or see-through (intestines, behind a desk, etc.) drawings (age 6, age 11) Level 7: Age 10 - 12, perspective drawings or elaborate colors (age 7, age 14) Level 8: Age 13 and above, precise details or abstract images (age 9, - ).

158

PART TWO ART INTERPRETATION

Figure 10.2 Standard drawings for each artistic development level.

Let us consider a drawing by an 8 year old child placed in the top right corner in Figure 10.2. An expert may immediately conclude that the drawing shows artistic skills far below the average level of a normal 8 year old, and note the presence of a problem. Our expert system performs the following steps to reach such a conclusion. 1) The drawing is first compared with the standard drawings for each of the 8 artistic skills levels. Let us assume that the above drawing is assigned to level 4. 2) The age factor is considered, in order to decide whether or not the child’s development is normal. If the child is younger than the lower limit for her or his age level, this suggests that the child possesses higher skills than the standard, whereas if he or she is older than the upper limit, this implies the opposite. In the present example, the child at age 8 has the artistic skills corresponding to level 4, the upper limit of which is age 7. Thus, it is concluded that the child’s artistic skills level is below the standard.

10.3 Reliability, consistency, and learning abilities Reliability The results from the expert system are in many cases uncertain. It is therefore essential that they are equipped with some sort of reliability measures. For example, if multiple elements that imply an identical symptom are expressed in a single drawing, the diagnosis of the symptom becomes more

Chapter 10 Reasoning Process of an Expert System for Art Therapy

159

Figure 10.3 Six cases illustrating the consistency of the proposed mechanism.

certain. If some information on the subject’s personal background of an individual indicates the same symptom along with the elements of the drawing, the result can be even more reliable. In our system, one of the following seven levels of reliability is assigned to each result. Reliability 1: No evidence Reliability 2: Weak evidence Reliability 3: Possibility Reliability 4: Indication Reliability 5: Moderate evidence Reliability 6: Strong evidence Reliability 7: Absolute evidence. Another important aspect of our system is maintaining consistency as illus- Consistency trated in Figure 10.3. Among the six drawings, drawings B and E are from Shin et al. (2002) and the rest are from Nakanishi (2002). Suppose only one element of the drawing, e.g. emphasis on eyes, is considered to evaluate the

160

PART TWO ART INTERPRETATION

reliability of the diagnosis of the symptom of sensitivity. The emphasis on eyes in a drawing is known to be related to the symptom of sensitivity (Shin et al., 2002). The emphasis on eyes is classified into 10 levels from none or negligible (level 1) to extraordinarily significant (level 10). Suppose these levels in the six drawings are 5, 9, 6, 9, 1, and 2, respectively. Then the reliability measures for the diagnosis of sensitivity follow this order in levels: The reliability measure of the diagnosis for drawing B and D, which have highest levels, is not less than the measures for other drawings; In the same manner, drawing C not less than drawings A, E and F; Drawing A not less than drawings E and F; Finally, drawing F not less than drawing E. We note that drawings B and D need not have the same reliability measures. Many combinations of reliability measures of sensitivity are possible to maintain such consistency. One example is 3, 6, 4, 5, 1, and 1, respectively, for the six cases. Now suppose two elements of the drawing, e.g. the size of the drawing and the emphasis on hands, are considered simultaneously to evaluate the diagnosis of offensiveness. The size of the drawing and the emphasis on hands in a drawing are believed to be related to offensiveness (Shin et al., 2002). The size of the drawing is classified into 10 levels from extraordinarily small (level 1) to extraordinarily large (level 10) and the emphasis on hands is also classified into 10 levels likewise. Suppose the levels of the sizes of the 6 drawings are 5, 9, 5, 9, 5, and 9, respectively, and the levels of the emphasis on hands 1, 3, 6, 6, 9, and 2, respectively. Then the reliability measure for drawing D is not less than drawings A, B, C, and F, since neither of its levels are less than those of the others. Likewise, several consistencies to be maintained are derived: Drawing B is not less than drawings A and F; Drawing F not less than drawing A; Drawing E not less than drawings A and C; Drawing C is less than drawing A. We note that drawings D and E need not have such ‘less than’ relationship in their reliability measures because drawing D is in a higher level in terms of the size but in a lower level in terms of the emphasis on hands than drawing E. The same is true between drawings E and F and drawings between C and F. One combination of the reliability measures for offensiveness satisfying these consistencies is 1, 4, 5, 7, 7, and 2, respectively, for the six cases. The results are summarized in Table 10.1. Learning Implementing an elaborate knowledge base requires much time and effort. In this study, we devise an easier method, which involves starting the reliability measure at 3 and increasing it by one whenever a corresponding cause occurs. We can use these six cases as a set of criteria when we diagnose a new drawing. Of course, we can delete or add cases to form a new set depending on its validity, appropriateness, or reasonableness decided by multiple therapists. As the number of cases in the criteria grows, reliability measures become subdivided. Thus, the mechanism is refined and becomes

Chapter 10 Reasoning Process of an Expert System for Art Therapy

161

________________________________________________________________________

Table 10.1 Levels of drawing elements Reliability measures of symptoms Levels of drawing (1-10) (1-7) elements and reliability _______ _______________________________ ___________________________ measures of symptoms Emphasis Size Emphasis Figure Sensitivity Offensiveness on eyes of drawing on hands

________________________________________________________________________ 10.3A 10.3B 10.3C 10.3D 10.3E 10.3F

5 5 9 9 6 5 9 9 1 5 2 9

1 3 6 6 9 2

3 6 4 5 1 1

1 4 5 7 7 2

________________________________________________________________________ more accurate as it accumulates more knowledge.

10.4 Knowledge base for each stage A number of studies have been conducted to investigate how pain, insecurity, nervousness, and other symptoms can be detected in the drawings of mental patients to establish relationships between emotional disabilities or psychological symptoms and the elements of a drawing. We cite a few of such relationships from the studies of Shin et al. (2002) and Kim (1988). However, as noted previously, many studies warn against making an interpretation only on the basis of the elements of a drawing without considering clients’ personal and environmental backgrounds. Unfortunately, only few studies have specifically explored this, and, as a result, knowledge is extremely scarce. Therefore, further research by related experts is necessary. We classify the knowledge into 10 categories. To form an easy grasp on concept, idea, or feature of the knowledge base, we present here one example of knowledge for each of the five categories selected. The entire knowledge base applicable for the sample case is given in Table 10.2. Standard diagnosis: Excessively large drawings indicate hyperactivity, a Examples of tendency toward activation, impulsiveness, offensive- knowledge ness, aggressiveness, inflation of self-ego, and expansive delusion. Individual diagnosis: Poor relationship among family members represents difficulties in personal relationships, uneasiness, worry or pervasive developmental disorder. Feedback-1: In the case of a slightly unbalanced drawing and the slight overuse of a certain color scheme, artistic talent needs to be taken into consideration. Feedback-3: A large face is normal for children under 9 years of age.

162

PART TWO ART INTERPRETATION

Table 10.2 ________________________________________________________________________ Knowledge base Common knowledge: [C.1] The terms ‘problems,’ ‘difficulties,’ ‘bad,’ and ‘symptoms’ have the same meaning. [C.2] The terms ‘development disability,’ ‘disability of development,’ and ‘underdevelopment’ have the same meaning. [C.3] The terms ‘mental’ and ‘emotional’ have the same meaning. [C.4] The terms ‘underdevelopment’ and ‘immaturity’ have the same meaning. [C.5] Whenever a condition for some psychological symptom appears or disappears, its reliability measure increases or decreases by one. ________________________________________________________________________ Standard diagnosis: [S.1] An X-ray style expression of the structure of a home is expressed in a KFD drawing implies the extinction of communication. [S.2] Frequent use of the color purple implies illness or unhappiness. [S.3] The color white implies purity or birth. [S.4] Drawings below the standard skills level indicate disabilities in intellectual, mental or physical development. [S.5] Excessively large drawings indicate hyperactivity, a tendency toward activation, impulsiveness, offensiveness, aggressiveness, inflation of self-ego, and expansive delusion. [S.6] A large face implies insecurity, childishness, and regression. [S.7] The omission of both arms implies difficulties in personal relationships. [S.8] The omission of both hands implies regression and problems in personal relationships. [S.9] Small feet indicate lack of self-control and fear. [S.10] The omission of a nose implies difficulties of personal relationships and fear. [S.11] A sense of social superiority in society appears as an imbalance clustering in the center and upper part of a drawing. ________________________________________________________________________ General diagnosis: [G.1] The contents of the DAP questionnaire and the Rorschach Index of mental disruption need to be reviewed (Kent, 1999). [G.2] In the case of an elderly person, some amount of forgetfulness is normal. [G.3] In a mixed community of white and black people, white people tend to have a sense of superiority. ________________________________________________________________________ Individual diagnosis: [I.1] Strong dependence on one’s mother means dependence on others, anxiety, lack of confidence, difficulties in personal relationships, childishness, and intellectual immaturity. [I.2] A child who refuses to go to school has difficulties in personal relationships, intellectual or mental development disability or learning disability. [I.3] Poor relationship among family members represents difficulties in personal relationships, uneasiness, worry or pervasive developmental disorder. ________________________________________________________________________ Feedback-1: [F.1.1] In the case of a slightly unbalanced placement and the slight overuse of a certain color scheme, artistic talent needs to be taken into consideration. [F.1.2] The importance attached to the order of a drawing varies depending on its theme.

________________________________________________________________________

Chapter 10 Reasoning Process of an Expert System for Art Therapy

________________________________________________________________________ Feedback-2: [F.2.1] In the case of a female subject, the presence of excessively passive characteristics is normal in cultural regions where the activities of women outside home is considered a taboo. [F.2.2] Elderly widows tend to show fear regarding physical safety and security, fear of being alone, fear of the dark, fear of victimization, and feelings of being unwelcome in a world of married couples (Hammer & Piotrowski, 1997; Lopata, 1996; Lev-Wiesel & Drori, 2000). ________________________________________________________________________ Feedback-3: [F.3.1] An X-ray drawing style of the structure of his/her home by a Taiwanese child may indicate an organized or systematic structure of mind. [F.3.2] In the case of a level 4 drawing, if the child is of age 8, this indicates mental, intellectual or physical development disabilities. [F.3.3] A large face is normal for children under 9 years of age. [F.3.4] A large drawing indicates intellectual immaturity in a younger child. [F.3.5] Low index from the Standardization Test indicates developmental disabilities. [F.3.6] For a level 4 drawing, the lower limit is age 5, and the upper limit is age 7. ________________________________________________________________________ Feedback-4: [F.4.1] When Forschah Schizophrenia Index shows severe difference with DAP questionnaire, modify the corresponding inputs. [F.4.2] Satisfaction in family relationship and social skills need to be taken into consideration together. ________________________________________________________________________ Feedback-5: [F.5.1] Lack of social skills is related to low satisfaction in family relationship. [F.5.2] Various Psychological Indexes and the applicable Picture Indexes need to be compared. ________________________________________________________________________ Feedback-6: [F.6.1] The level of drawing skills is adjusted according to the extent of art education received. [F.6.2] A slight indication of a symptom is normal in the case of a child with artistic talent. [F.6.3] In the case of spastic children, symptoms are deleted from the diagnosis even if the level of drawing skills is low. [F.6.4] Attention deficit, hyperactivity, aggressiveness, and bad relationships with people means ADHD. [F.6.5] When developmental disability or learning disability appears, intelligence testing is recommended.

________________________________________________________________________ Feedback-6: When developmental disability or learning disability appears, intelligence testing is recommended.

10.5 Case study Figure 10.4 is a drawing by an 8 year old girl (Nakanishi, 2002). She is

163 Table 10.2 (continued)

164

PART TWO ART INTERPRETATION

Figure 10.4 A case study.

highly dependent on her mother, and refuses to go to school. Note that cellophane paper is attached to the drawing in order to compare the size of the object, its placement, and other similar factors. This example attests the advantages of using a PC in analysis. The drawing has the following elements: (1) Large drawing (2) Large face (3) The artistic development level: 4 (4) Omission of the nose (5) Omission of the hands (6) Small feet. Her general environmental factors are follows: (1) Age 8 (2) Girl. Her specific environmental factors are follows: (1) Strong dependence on her mother (2) Refusal to go to school (3) Inadequate interactions within her family. The diagnosis results for each phase are as follows. The values indicated in parenthesis are the reliability measures and those indicated in square

Chapter 10 Reasoning Process of an Expert System for Art Therapy

brackets are the knowledge code evoked. 1) Standard diagnosis: (1) Large drawing - hyperactivity (3), activation tendency (3), impulsiveness (3), offensiveness (3), aggressiveness (3), inflation of selfego (3), expansive delusion (3) [S.5]. (2) Large face - insecurity (3), childishness (3), regression (3) [S.6] (3) Omission of nose - difficulties of personal relationships (3), fear (3) [S.10] (4) Omission of hands - regression (4), problems in personal relationships (4) [S.8], [C.1], [C.5]. (5) Small feet - lack of self-control (3), fear (4) [S.9], [C.5]. 2) Feedback-1: no application 3) General diagnosis: no application 4) Feedback-2: no application 5) Feedback-3: (6) Age 8, level of drawing 4 - above upper limit of age [F.3.6]. (7) Large face, age 8 - insecurity deleted, childishness deleted, regression (3) [F.3.3], [C.5]. 6) Standard diagnosis: (8) Age 8, level of drawing 4, above upper limit of age - intellectual immaturity (3), mental immaturity (3), physical immaturity (3), [S.4], [C.2], [C.4]. 7) Feedback-1: no application 8) General diagnosis: no application 9) Feedback-2: no application 10) Feedback-3: no application 11) Individual Evaluation: (9) Dependence on mother - dependence on others (3), anxiety (3), lack of confidence (3), difficulties in personal relationships (5), childishness (3) intellectual immaturity (4) [I.1], [C.1], [C.4], [C.5]. (10) Refusal to go to school - difficulties in personal relationships (6), intellectual immaturity (5), mental immaturity (4), learning disability (3) [I.2], [C.2], [C.4], [C.5]. 12) Feedback-4: no application 13) Feedback-5: no application 14) Feedback-6: (11) Hyperactivity, aggressiveness, problems in personal relationships ADHD (3) [F.6.4]. [C.1]. (12) Intellectual immaturity - Intelligence testing is necessary (3) [F.6.5], [C.4]. (13) Development disability - Intelligence testing is necessary (4) [F.6.5], [C.4], [C.5]. (14) Learning disability - Intelligence testing is necessary (5) [F.6.5], [C.5].

165

166

PART TWO ART INTERPRETATION

Figure 10.5 Diagnosis results.

(15) Physical disability - Intelligence testing is necessary (6) [F.6.5], [C.4], [C.5]. (16) Mental immaturity - Intelligence testing is necessary (7) [F.6.5], [C.4], [C.5]. 15) Standard diagnosis: no application 16) Feedback-1: no application 17) General diagnosis: no application 18) Feedback-2: no application 19) Feedback-3: no application 20) Individual diagnosis: no application 21) Feedback-4: no application 22) Feedback-5: no application 23) Feedback-6: no application 24) End. The final result is as follows. The user interface of the system indicates the result of the diagnosis process as shown in Figure 10.5. (1) Hyperactivity (3), tendency towards activation (3), impulsiveness (3), offensiveness (3), aggressiveness (3), inflation of self-ego (3), expansive delusion (3) (2) Immaturity / underdevelopment - intellectual (4) / mental (4) / physical (3) (3) Problems with personal relationships (6), dependence on others (3)

Chapter 10 Reasoning Process of an Expert System for Art Therapy

(4) Lack of self-control (3), fear (4), anxiety (3), childishness (3), regression (3), lack of confidence (3). (5) Learning disability (3) (6) Intelligence testing recommended (7) (7) ADHD (3).

10.6 Discussion We proposed an expert system model in which a machine emulates the synthetic thinking and judgment abilities of human art therapists in arriving at decision. The model consists of three diagnosis processes and six feedback processes. The diagnosis embodies the general 1:1 relationship between cause-and-effect, and the feedback interactive relationships between two or more causes and effects. The system can accumulate the knowledge of highly authoritative experts or scholars in all of the related fields. We have developed methods to provide objective standards and, thus, to insure consistency in the decision. Furthermore, the system is equipped with learning abilities by means of which the knowledge of the system expands with each use, just as a human being accumulates knowledge through learning or experience. Also, the system can provide logical reasons for the results it yields. The diagnosis mechanism of this model can resemble the reasoning process of a human expert. It would, if anything, be ideal for human experts to diagnose their clients through the rigorous application of the proposed mechanism. In practice, it is understandably difficult for human experts to emulate the machine’s mechanism thoroughly and flawlessly. Machines, i.e. computers, execute this mechanism quickly and thoroughly, without fatigue that human experts may experience. The proposed expert system is at present merely a prototype that handles an extremely small portion of the scenario-based information related to DAP drawings and KFD drawings. Also, only part of the functions of the computer are utilized. The amount of knowledge involved, as well as the capability of the computer, is in fact massive. Active participation of professionals in this field is essential for the development of this system. Much time, effort, manpower, and cost will be required before its practical use can be envisaged.

167

An Expert System for Interpreting the Structured Mandala Coloring (SMC) Drawings

Chapter 11 Abstract and summary We develop an expert system as a systematic, scientific, computerized, and interdisciplinary approach to art therapy. In particular, the expert system is designed to interpret the Structured Mandala Coloring (SMC) drawings which is selected among all of the art therapy tools. This system can assist human experts by providing the analysis and interpretation of SMC drawings. Through a questionnaire, the system obtains drawers’ personal preferences of color and automatically analyzes and evaluates the elements in the SMC by the C_CREATES. A knowledge base has been constructed on the meanings of the colors used in the SMC, the relationship between the colors and the personal characteristics in the questionnaire, and changes of elements detected in a series of SMC drawings obtained over a period of time. The knowledge is expressed in a simple IF - THEN format and accumulated in the knowledge base. The system evokes relevant knowledge corresponding to the analysis and evaluation of the elements of a SMC drawing and the answers to the questionnaire. The system operation and usability are illustrated through sample cases.

11.1 The Structured Mandala Coloring (SMC) as a subject of expert system Gantt (1998) foresaw future research for the role science plays in art therapy and for systematic application of science to art therapy. Kaplan (1998) also recognized the need for research on the integration of findings of other disciplines into art therapy. Recently, Kapitan (2007) emphasized the necessity for interdisciplinary study of incorporating digital technologies into art therapy to expand the definition of art materials and contexts across a wider spectrum. In line with these research recommendations, we suggest the use of an expert system, a promising field of artificial intelligence in computer science. The purpose of the expert system is to aid human experts in deci168

Chapter 11 An Expert System for Interpreting the SMC Drawings

169

sion-making process. From the viewpoint of artificial intelligence, the field of art therapy is a ISP so-called the Ill-Structured Paradigm (ISP), which is referred to a vague and unclear problem domain. There hardly exist organized algorithms or objective means of finding a solution, and, moreover, in certain cases the solutions found are inevitably contradictory. People and art are too complex to be evaluated based on the interpretation of a single element of art as substantial evidence of their characters (Cohen & Mills, 1994). The same element of drawings may have different meanings for different drawers depending on the drawers’ national, personal, cultural, social, and educational backgrounds. This nature of art therapy makes the decision-making process complex and difficult for art therapists. An expert system with an appropriate knowledge base is an approach to systemize such diverse knowledge. Being viewed as an ISP never indicates that art therapy is an unimportant field. If anything, it makes art therapy all the more worthwhile to be investigated. In this sense, art therapy is analogous to disciplines such as economic demand forecasting, weather forecasting, diagnosis of diseases, or judicial sentencing in criminal cases. The use of artificial intelligence in these areas is already in progress, and has yielded significant results (Giarratano & Riley, 2005). We suggest interdisciplinary research on incorporating the expert system into art therapy as a useful method of systemizing, organizing, and classifying various kinds of art therapy knowledge, and as an appropriate means of finding solutions to the problems encountered in art therapy practice. Careful and positive attention to expert systems could bring about significant progress in art therapy. Experts from various fields of art therapy, psychology, psychiatry, education, art, computer science, statistics, etc., are expected to participate in this interdisciplinary study. This chapter develops an expert system for the interpretation of drawings. Expert system The system emulates human experts’ decision-making using an art therapy tool. The system analyzes various elements in drawings by using computer technology and establishes a knowledge base through systematic examination and organization of existing theories, practices, and knowledge of experts in art therapy. The system provides information and saves time and effort of art therapists. Also, parents and teachers, who are not art therapists, can use the system in ascertaining signs of their children’s psychological problems at an early stage. Timely detection can allow them to promptly seek professional treatments. Unfortunately, we can only find a handful of expert systems, all of which are fairly elementary, such as the ones proposed in the previous two chapters and the one you will see in the next chapter (Kim, Ryu, et al., 2006; Kim, Kim, Lee, Lee, & Yoo, 2006; Kim, Yoo, Kim, & Lee, 2007, Kim, Han, Kim, & Oh, 2011). Moreover, they are prototypes, which are not quite

170

PART TWO ART INTERPRETATION

suitable for practical applications, merely suggesting directions for future development. Also, we can find only a few computer systems that are designed to automatically analyze and evaluate the elements in the C_CREATES as we saw in Part One (Kim, Bae, & Lee, 2007; Kim, Kang, & Kim, 2008; Kim, 2010). These systems identify and evaluate only the elements in the C_CREATES, which are far from general elements of drawings, such as forms, lines, symbols, etc. SMC Thus, first of all, the drawing which is to be the subject of this computer application should be simple enough to process with the existing computer systems for evaluation, while we develop expert systems for interpretation. For this reason we selected the Structured Mandala Coloring (SMC) (Curry & Kasser, 2005), which is coloring a given geometric pattern, as an art therapy tool. The most common elements in the SMC are directly related to colors. The C_CREATES can evaluate not only these elements, but also the completeness, the accuracy, and the degree of concentration during the coloring activity, which are the most important elements in the SMC. As far as the SMC is concerned, the system considers almost all elements that human experts would consider, and these elements in turn can be evaluated by the C_CREATES. In addition, coloring mandalas was reported to be useful for reducing undergraduate students’ anxiety (Curry & Kasser, 2005) and also the worry and anxiety of patients (Cornell, 1994). There are studies on estimation of the level of dementia based on the elements in the SMC in Chapter 14 (Kim, Kim, & Kang, 2008). Also, the expert system approach proposed in Chapter 9 (Kim, Kim, et al., 2006) can be applied to the knowledge base of these elements. The system analyzes, interprets, and detects changes of SMC drawings. We collect and classify the knowledge required for the interpretation of colors in the SMC, and for the detection of changes in a series of SMC drawings, which constitutes the knowledge base of the expert system. The system analyzes and evaluates personal characteristics of drawers obtained through questionnaire, and the elements in the SMC, which evoke the relevant knowledge in the knowledge base. Of course, the system provides fundamental quantitative information of analysis and evaluation, and results of interpretation. More importantly, it detects and reports any changes in a series of the SMC drawings. We note that clinicians are under pressure to demonstrate their clients’ progress (Betts, 2006). In art therapy, for instance, an assessment can be administered at the outset of treatment, during the middle phase of treatment, and again upon the termination of services. The drawings can be compared to determine the course of patient treatment. When practitioners and institutions are held accountable for charting and reporting client progress using the system, treatment standards are raised, and this has a trickle-down effect that tends to improve the quality of treatment a client

Chapter 11 An Expert System for Interpreting the SMC Drawings

receives (Deaver, 2002; Gantt & Tabone, 2001).

11.2 Knowledge base Knowledge is expressed in the form of IF - THEN statements. The knowledge base in this expert system contains several kinds of knowledge. We list below a part of the knowledge base with an explanation about the knowledge expression. The knowledge can be classified into categories concerning color-related elements in general drawings, color-related elements in the SMC, the relationship between personal preferences of color and the color used in SMC drawings, and the detection of changes in a series of SMC drawings. We include the knowledge in Fincher’s book (Fincher, 1991), which contains a wealth of useful information concerning mandala creation and interpretation (Pitak-Davis, 1992) and a fairly detailed explanation of the hypothesis of Kellogg (1977) on the use of color. 11.2.1 Knowledge expression Knowledge is expressed in the following IF - THEN format. The form is not strict, but flexible enough for some parts to be changed and even be omitted depending on the properties of the knowledge. IF condition-1, or condition-2, or ... , or condition-n, THEN result-1, and/or result-2, ... , and/or condition-m. (confidence level of results: references). An expression consists of n conditions and m results. An expression without a condition is possible for universally true facts. Each condition can consist of several “and” conditions. The content within the double quotation marks is treated as a condition or a result. Confidence level 1 means “occasionally (probability of up to 0.3),” level 2 “may be (probability of up to 0.5),” ... , and level 5 “certainly (probability of 1.0).” At first, we present two knowledge expressions for universally accepted facts, with no condition: [Knowledge 1] IF THEN “Johannes Itten (1961) said in his representative book of art education, Art of colors, that colors are energy having positive or negative effects regardless of our consciousness.” [Knowledge 2] IF THEN “The expert system has interpreted the colors in your mandala. The interpretation of colors is based on the most general and most commonly acknowledged facts on colors in drawings including mandalas. We note that this interpretation can vary from positive aspects to neg-

171

172

PART TWO ART INTERPRETATION

ative aspects depending on your personal experience, background, culture, etc. Thus, the interpretation may be quite different from the real situation.” 11.2.2 Structure of knowledge base Colors in drawings The majority of elements in the SMC are color-related ones. Rorschach (1951) regards color as a means by which people can reveal their emotions. How people react to and use color can provide important diagnostic information regarding their current emotional status (Lev-Wiesel & Daphna-Tekoha, 2000). Also, the therapeutic value of color has long been recognized (Ghaffurian, 1995). We present some knowledge on colors in the form of natural English statements and its knowledge expressions in the IF - THEN format. “There are reports that child victims of severe sexual abuse (Malchiodi, 1990) and depressed patients (Wadeson, 1980; Gantt & Tabone, 1998) tend to use only one or two colors in their drawings. Victims of trauma express their psychological pain, anxiety, fear, sorrow, loneliness, and hopelessness by selecting certain colors. Children who experienced natural disasters such as earthquakes, hurricanes, and plane crashes tend to use a limited number of colors, not more than two or three, mostly consisting of black, white, and sometimes red (Gregorian, Azarian, DeMaria, & McDonald, 1996).” [Knowledge 3] IF the number of colors used ≤ 2, THEN experience of severe sexual abuse (Malchiodi, 1990), and/or depression (Wadeson, 1980; Gantt & Tabone, 1998). [Knowledge 4] IF the number of colors used ≤ 3 and colors used = black, or white, or red, THEN victims of trauma with psychological pain, anxiety, fear, sorrow, loneliness, and hopelessness. [Knowledge 5] IF the number of colors used ≤ 3 and colors used = black, or white, or red, THEN experience of natural disaster such as earthquakes, hurricanes, or plane crashes (Gregorian et al., 1996). “According to Fincher (1991), when red occupies a large area or appears frequently in a mandala, it shows a healthy existence and energy for change to understand inner wisdom in the positive respect, and anger and pain accompanying hurt and destruction in the negative respect.” [Knowledge 6] IF the main color is red, or the main or subsidiary color is red in 3 mandalas among 4 consecutive mandalas, THEN healthy existence, and/or energy for change to understand inner wisdom in the positive respect, or anger

Chapter 11 An Expert System for Interpreting the SMC Drawings

173

or pain accompanying hurt and destruction in the negative respect (3: Fincher, 1991). Now it is clear how natural English statements are expressed in the IF THEN format. There can be two kinds of knowledge that are contradictory to each other. How a person feels about a color is intensely emotional and personal. For instance, a color which one person experiences as warm may be cool for another. Neither is “right.” Both are correct (Fincher, 1991). [Knowledge 7] IF green, THEN a warm color. [Knowledge 8] IF green, THEN a cool color. More desirable expressions will be: [Knowledge 9] IF green and Italian or American or Goethe or Kandinsky, THEN a warm color to most people (Breaem, 1986). [Knowledge 10] IF green and French or German or Luscher, THEN a cool color to most people (Breaem, 1986). The color red holds different meanings for different people. [Knowledge 11] IF red, THEN a warm color, and energetic color that is physically stimulating to behold (Lusher, 1969). [Knowledge 12] IF red, THEN the physical life of man - lust, blood, and atavistic emotions associated with killing and assertiveness (Kellogg, 1977). [Knowledge 13] IF red, THEN burning and surging emotions (Jacobi, 1979). According to the definition of the mandala, the colors used within it act Colors in mandalas as a window to the center of our mind. We may obtain information on the status of a patient or diagnosis by the interpretation of colors (Jung, 1989). We present several knowledge expressions in Fincher (1991) and Kellogg (1977). Colors have both positive and negative meanings. [Knowledge 14] IF red, THEN the energy needed to survive, and/or being healthy, and/or transforming oneself to greater inner wisdom. [Knowledge 15] IF red, THEN wounds, and/or destructive rage, and/or suffering. The color red may have one of the following traditional meanings for one person, or it may mean something entirely different for another.

174

PART TWO ART INTERPRETATION

[Knowledge 16] IF red, THEN raw energy called libido. [Knowledge 17] IF red and blood, THEN anger, and/or suffering. [Knowledge 18] IF red, THEN a commitment to life, and/or the will to survive, and/or an acceptance of the body. [Knowledge 19] IF red, THEN the fire of emotion, or spirituality, or transformation. [Knowledge 20] IF red, THEN “Please consider the possibility of the arousal of healing, life-giving potential deep in the psyche.” [Knowledge 21] IF red, THEN vital sign of progress toward achieving the magnum opus (great work). The appearance of the color red in mandalas is desirable. [Knowledge 22] IF red, THEN an indicator of “the will to thrive” (Kellogg, 1977). [Knowledge 23] IF a series of mandalas with little or no red, THEN an indication of passivity, or lack of self (3). We present some knowledge expressions on colors which occupy only a small part of the knowledge base. [Knowledge 24] IF main color = red, and percentage of red ≥ 0.5, THEN need to change personality of violence, aggressiveness, or a person of passion, or anger. [Knowledge 25] IF main color = yellow, and percentage of yellow ≥ 0.3, THEN a broad-minded person. [Knowledge 26] IF main color = brown, and percentage of brown ≥ 0.3, THEN a shy person. [Knowledge 27] IF main color = red, or main color = orange, or main color = yellow, subsidiary color = red and percentage of subsidiary color ≥ 0.2, or subsidiary color = orange and percentage of subsidiary color ≥ 0.2, or subsidiary color = yellow and subsidiary color ≥ 0.2, THEN “You may be outgoing, active and liberal in expressing your feelings. Received affection from parents and may depend on others. Good human relationships, somewhat self-centered, but cooperative, adaptable to new environments.” [Knowledge 28] IF percentage of red ≥ 0.5, THEN “There may be a need to question possibility of

Chapter 11 An Expert System for Interpreting the SMC Drawings

175

emotional or psychological abnormal state.” [Knowledge 29] IF percentage of red ≤ 0.03, THEN “There may be a need to check passiveness, deficit of desire or confidence to do or to say something.” [Knowledge 30] IF complementary color = blue and orange, THEN “There may be a conflict between the desire to have some relationships and the effort required to form them (Fincher, 1991).” [Knowledge 31] IF percentage of cool colors ≥ 0.7, THEN “You may have a very cool and rational personality in terms of problem solving.” [Knowledge 32] IF the colors red, blue, green, and yellow are balanced, THEN harmony of the inner person. [Knowledge 33] IF purple (= red + blue) THEN energy (red) aligned with the archetype of the mother (blue). [Knowledge 34] IF orange, THEN self-assertion, pride, and ambition, and concerns about one’s power, or one’s lack of power. [Knowledge 35] IF orange, THEN energetic striving, a strong sense of identity, healthy assertiveness, a willful use of power, a hostile attitude toward authority, or no self-discipline. [Knowledge 36] IF a great deal of orange, THEN an ambivalent feeling about maleness and about ego strivings. The appropriate significance of mandalas for a client can be determined by Personal referring to the patterns and meanings suggested by the mandala as a whole characteristics and his or her individual characteristics such as age, gender, family, environment, experience, preference, culture, etc. We present several knowledge expressions for the interpretation of the psychological or emotional state by connecting the personal characteristics obtained via questionnaire with the elements in the SMC. “When the personal color representing his or her happiness appears in the mandala, he or she may feel happy.” [Knowledge 37] IF input-color of happiness = main color, THEN current emotional status = happy (1). We present knowledge which can be used for checking the answers to the questionnaire. [Knowledge 38] IF input-current state of emotion ≠ input-current state of feeling, THEN “There is a contradiction in your answers to the

176

PART TWO ART INTERPRETATION

questionnaire. Please answer the question more sincerely.” [Knowledge 39] IF main color = input-color of happiness, THEN “You seem to be happy now.” [Knowledge 40] IF input-color of myself = input-color of father and inputcolor of myself ≠ input-color of mother, THEN “You seem to be more familiar with your father than with your mother.” [Knowledge 41] IF more red than usual and women, THEN going through menses (Fincher, 1991). [Knowledge 42] IF a great deal of orange and female, THEN attitude towards men, an attachment to father, and a reflection of great self-esteem, ambition, and the like (Kellogg, 1977). Changes Bonny and Kellogg (1977) also employed the mandala to measure changes in drawings in art therapy, as well as to diagnose. There is evidence that the SMC is not only a valuable diagnostic tool, but is also a tool to check therapist’s immediate impression. In addition, it can, at times, be a source of valid predictions that may warn us of pitfalls and guide us in the direction of constructive therapeutic maneuvers. We present several knowledge expressions for detection of changes in a series of SMC drawings. [Knowledge 43] IF number of used colors of previous mandala x 1.3 ≤ number of used colors of present mandala, or number of clusters of previous mandala x 1.3 ≤ number of clusters in present mandala, or completeness of previous mandala x 1.3 ≤ completeness of present mandala, accuracy of previous mandala x 1.3 ≤ accuracy of present mandala, or concentration of previous mandala x 1.3 ≤ concentration in present mandala, THEN “You color this mandala with more perseverance and you seem to be happier.” [Knowledge 44] IF main color of previous mandala = red and main color of present mandala = blue, THEN “You seem to be calmer.” [Knowledge 45] IF percentage of warm colors of previous mandala x 1.3 ≤ percentage of warm colors of present mandala, THEN “You are more liberal and active in the expression of your feelings.”

11.3 An expert system

Chapter 11 An Expert System for Interpreting the SMC Drawings

177

Figure 11.1 The structure of the expert system for art therapy.

The expert system in this study has the structure of Figure 11.1, which is the same as the one proposed in Chapter 9. It has following several functions concerning the SMC. The user interface is a mechanism through which the user and the expert system communicate. We establish a user friendly interface to make the system easier to use. First, a client answers the questionnaire in Figure 11.2 by selecting one of the given choices. The client inputs his or her image files of mandalas as the system instructs, then he or she receives the results of the analysis and interpretation of the system. The C_CREATES is used for the analysis and evaluation of the elements in the SMC. The system evokes relevant knowledge in the knowledge base corresponding to the answers to the questionnaire and elements in the mandalas. The system provides the basic information and statistics as shown in Figure 11.3 and the interpretation and changes detected as shown in Figure 11.4. Knowledge from each case is accumulated in the system knowledge base. Consequently, as the system progresses, it becomes updated and more accurate. For example, the average standard deviation of number of used

178

PART TWO ART INTERPRETATION

Figure 11.2 A questionnaire.

colors are 4.55 and 2.51, respectively, and those of accuracy are 58.8 % and 22.1 %, respectively, and those of completeness are 0.49 and 0.21, respectively. These values are updated. A kind of machine learning takes place.

11.4 Case study We explain the procedure of system operation and usability with the two

Chapter 11 An Expert System for Interpreting the SMC Drawings

179

Figure 11.3 Analysis results of a SMC sample.

SMC drawings colored by a same person over a course of time shown in Figure 11.5. The client answers the questions in Figure 11.2 as follows. (1) Favorite col- Questionnaire or: red, Least favorite color: black (2) Favorite primary color: red, Favorite secondary color: orange, Favorite achromatic color: white (3) Face representing present feeling: happy (4) Color of myself: red, Color of my father: blue, Color of my mother: green (5) Present feeling: very happy (6) Color when sad: gray, Color when angry: black, Color when happy, red.

180

PART TWO ART INTERPRETATION

Figure 11.4 Interpretation of a SMC sample.

Analysis of elements The analyses of the elements in the two SMC samples produced at an interval are summarized in Table 11.1. Examining the differences in the two mandalas, we can see improvement in all the measures of accuracy, number of used colors, number of clusters, completeness, and degree of concentration, and thus conclude that there have been positive changes in the psychological states. Also, there have been increasing percentages of primary colors and warm colors, and thus we estimate that there is a change of emotional activeness.

Chapter 11 An Expert System for Interpreting the SMC Drawings

181

Figure 11.5 Two SMC samples colored at an interval.

This case evoked the following eight items among the knowledge in this Related knowledge study: evoked [Knowledge 2] on note, [Knowledge 11 - 22] on red color, [Knowledge 24] and [Knowledge 27] on main color and subsidiary color, [Knowledge 30] on complementary colors, [Knowledge 34 - 35] on orange color, [Knowledge 37] and [Knowledge 39] on color of happiness, [Knowledge 40] on colors of family, and [Knowledge 42] on color and gender. The answers to the questionnaire and the conditions (IF) deduced from the Statistics and mandala provided the statistics in Figure 11.3 and the interpretation in Fig- interpretation ure 11.4.

11.5 Discussion The Jungian school of thought has been often criticized for being too mystical and unscientific. Jung’s theory has been often criticized for being too vague. Cognitively-oriented art therapists sometimes shy away from Jungian theory claiming that it is too complex and difficult to understand and thus better left to the artistic and the religious. The plight of art therapy Table 11.1 Results of element analysis of the SMC samples in Figure 11.5

_____________________________________________________________________________________________

Number Number Concent- Primary Secondary Warm Cool Complete- ration colors colors Figure Accuracy of of (%) ness colors clusters (rank) (%)

_____________________________________________________________________________________________ 11.5A 60.1 11.5B 75.6

4 7

19 15

0.45 45.07 33.6 65.3 38.2 29.3 0.87 27.85 59.4 32.7 63.2 18.6

_____________________________________________________________________________________________

182

PART TWO ART INTERPRETATION

has been somewhat similar to that of Jungian theory, due to the limited amount of scientific research presently existing in this relatively new field (Slegelis, 1987). Work remains to be done on systematizing the methods of interpretation, so that the process depends less on special intuitive abilities and becomes more accessible to art therapists wishing to work in this area (DiLeo & Kellogg, 1977). As a method systemizing knowledge in art therapy, we constructed an expert system for interpretation of SMC drawings, which is applicable to real cases. We note that there are studies on the computer evaluation of various elements (Kim, Kang, & Kim, 2009) in Chapter 2 and the estimation of the level of dementia (Kim, Betts, Kim, & Kang, 2009) using the SMC in Chapter 14. The user interface system is designed for the user to easily communicate with the computer: input of color preferences via answering a questionnaire to the system, output of analysis results, and interpretation presented in the form of tables and graphics. We tried to develop this system for real-life applications. However, we admit that it is still at the prototype stage. When it progresses to a reasonable level, we will make the system available to the public. We systemized the knowledge concerning coloring. We included the knowledge from the literature of Fincher (1991) and Kellogg (1977). Knowledge collected afterwards can be accumulated according to the knowledge base structure. We emphasized that the interpretation of mandalas does not depend only on the information contained within the drawing, but also on the personal experience, environment, cultural background, etc., of the drawer. However, we admit that this is not sufficient knowledge to consider more diverse aspects of individual situations and backgrounds. When we develop more knowledge related to individual situations in a systematic manner, the knowledge base will evolve and contain more useful information. We anticipate active participation of experts in related fields. However, the basic information in this system can still be a useful aid to the decision-making process of human experts. Monitoring a series of SMC drawings will be especially useful for detecting changes in a drawer’s psychological states over the course of a therapy. The changes can be stored in the computer, like a patient’s medical chart in a hospital. As we mentioned, there is no consensus among therapists regarding the emotional meanings of specific colors (Furth, 1988; Golomb, 1999; Leavitt & Schimmel, 1991). Apart from the meanings described in [Knowledge 34], [Knowledge 35], and [Knowledge 36], the color orange is reported to indicate adaptability, brightness, timidity, anxiety, unrest, imagination, and evasion all at once (Alschuler & Hattwick, 1947) and stability and imagination (Kim & Lim, 1972). Therefore, use of orange can lead to the contradictory results, such as adaptability versus evasion, brightness versus timidity, and unrest versus safety. An expert system can be an approach to solve these problems of inconsistency by establishing a relevant knowledge base. At

Chapter 11 An Expert System for Interpreting the SMC Drawings

least, the expert system can tell contradictions in itself. Also, the system can tell each knowledge with the confidence level as proposed in the previous chapter. Or, the system can tell knowledge with probability as will be seen in the last chapter of this book. The cases accumulated in this system are SMC drawings colored by elderly persons who are either normal or have suspected dementia. In general drawings or unstructured mandalas, art therapists usually examine form, symbols, lines, colors, and movements, which, for the time being, a computer system cannot be expected to detect and analyze. However, as for the SMC, the client does not draw any forms or symbols, but only color given structured patterns, for which the C_CREATES is able to do most of the work necessary for analysis and evaluation. Thus, the SMC can be an appropriate art therapy tool for computerized evaluation of artwork produced by clumsy hands of people suffering from dementia. This expert system is the first computerized art therapy tool of the SMC. As computer technology advances further enough to identify forms, lines, symbols, etc., the system can be extended to other art therapy tools such as the DAP, the House-Tree-Person (HTP) (Buck, 1949), the Diagnostic Drawing Series (DDS) (Cohen, 1986/1994), the Person Picking an Apple from a Tree (PPAT) (Gantt, 1990), etc. (Brooke, 1996). We also anticipate that this expert system will be extended to psychological disorders such as depression, ADHD, multiple personality, dissociative identity disorder, etc. although the same approach in this chapter can still be applied.

183

Computerized Kinetic Family Drawing Using Given Patterns (p_KFD)

Chapter 12 Summary This chapter presents the development of the Computer System for the Kinetic Family Drawing with Patterns (p_KFD) (Kim, Han, Kim, & Oh, 2011). The p_KFD consists of four stages. The system provides client with questionnaires on family relationship and color preference, and his or her answers constitute a fact base. The system provides the client with various patterns of family members and backgrounds which possibly compose a KFD drawing. The client selects several among them, arranges, expands or contracts, and colors them. The system evaluates elements and detects the changes of evaluations in the p_KFD drawings. In addition, the evaluation and detection of changes also constitute the fact base. The system interprets one or several p_KFD drawings by invoking knowledge in the knowledge base corresponding to the facts. The knowledge base of the p_KFD is continuously being accumulated and updated. The system automatically provides art therapists not only with an objective evaluation of the elements, but also with information about client’s psychological states and treatment process. It is expected that the p_KFD with patterns will inherit and complement the validity of the traditional KFD with free drawings.

12.1 The Kinetic Family Drawing (KFD) as a subject of computerizing p_KFD This study delineates the development of a computer system that can operate in real-time via the internet for the KFD developed by Burns and Kaufman (1972). The system deals with the KFD drawings comprised of given patterns, not freely drawn ones. We designate the system as the p_ KFD (Patterned KFD) (Kim, Han, Kim, & Oh, 2011). Despite the remarkable progress of computer technology, at present, the computer is unable to identify forms in free drawing and, therefore, is unable to determine the actions of forms and the spatial relationships among them in the freely drawn 184

Chapter 12 Computerized Kinetic Family Drawing Using Given Patterns (p_KFD)

KFDs. Thus, for the system to identify the subjects in a drawing, the system prepares in advance various kinds of patterns of family members and backgrounds which will possibly compose a KFD drawing, and provides them to the client, who will then select several among them, arranges, expands or contracts, and colors them. Then, the system can identify the forms and evaluate the placement, size, distances between forms, and various other elements. We have examined all the elements for evaluation in the KFD available in literature including Burns and Kaufman (1972), McPhee and Wegner (1976), Myers (1978), and Reynolds (1978), and all the elements in the Descriptive Assessment of Psychiatric Artwork (DAPA) (Hacking, 1999), the DDS, the PPAT, and the C_CREATES. After consolidating those elements which have different names but are the same in the substance, we come up with 127 elements. For the evaluation of these elements, there exist a number of positive reports and papers showing high inter-rater reliability. Mostkoff and Lazarus (1983) mentioned the possibility of creating an objective evaluation system, and McPhee and Wegner (1976), Cummings (1980), and Elin and Nucho (1979) reported high inter-rater reliability. Since most elements are evaluated by the human rater’s intuition, judgment, and subjective determination, however, certain inconsistencies are inevitable. Even for those elements using certain selected objective criteria, such as rulers or grids made of tracing paper marked off in millimeters, the process of evaluating them is time consuming and the problem of inaccuracy still remains. However, the use of the built-in functions of the computer, the C_CREATES, and the p_KFD can automatically, accurately, precisely, quantitatively, and objectively evaluate all the 127 elements, achieving absolute consistency in the evaluation. As a subject of computerizing by providing patterns, we choose the KFD, since it is one of the most widely used art therapy tools and its reliability and validity have been proven. We expect the p_KFD with given patterns to inherit the applicability, usefulness, and validity of the traditional KFD with free drawings. Burns and Kaufman (1972) have argued that the KFD reflected emotional disorders faster than other methods did and offered art therapists information not only about the adaptive and defensive functions of children, but also about their family dynamics. There is a good amount of literature reporting the validity of the KFD in its application for various purposes. For example, the studies of Schornstein and Derr (1978), Hackbarth, Murphy, and McQuary (1991), Reddy, Bhadramani, and Samiullah (2002), and Veltman and Browne (2003) have shown that the KFD could successfully identify cases of child abuse or neglect, or distinguished children who were suffering from child or neglect from normal children. Unlike these studies that reported positive results using the KFD, however, there are other studies that raised questions about its validity. Cummings (1980) suggested that KFD drawings might be sensitive to the transition in

185

186

PART TWO ART INTERPRETATION

children’s personality states. Mostkoff and Lazarus (1983) also asserted that the KFD was sensitive to children’s mood changes and, therefore, might not be an accurate measure of personality traits or characteristics. There are even some studies reporting contradictory results. Monahan (1986) noted that psychological symptoms were found more frequently in the KFD drawings created by high-achieving children than in those by lower-achieving children. These results failed to support the hypothesis of Burns and Kaufman (1972), that the KFD can determine the extent of a child’s adjustment. There are other studies questioning validity or reporting contradictory results but we omit them here. For this problem, we expect that the p_KFD could be a helpful tool to increase the validity of knowledge. This will be revisited in discussion and conclusion section. Four stages of The p_KFD consists of four stages. The first stage of the questionnaires p_KFD aims to gather information about the client’s environment and history, which cannot be obtained from drawings. The system provides two questionnaires regarding the client’s family relationships and color preferences, the answers of which constitute a Fact Base (FB). In the second stage of KFD composition and coloring, the client selects a few among various given patterns, and arranges, rotates, expands or contracts, and colors them. In the third stage of analysis and evaluation of elements in the composed KFD, the system evaluates the elements of a KFD drawing or detects changes of evaluation in several KFD drawings by using built-in functions of the computer and some existing computer systems for art therapy. The results of the evaluation and detection of changes also constitute an FB. In the final stage of interpretation, the system provides information about the client’s psychological states or treatment process by invoking the knowledge in the Knowledge Base (KB) corresponding to the facts in the FB. The KB, which consists of the expertise and experience of experts and relevant literature, is continuously accumulated and updated. The current system remains so far as a prototype model, even though all the features have been designed. The procedure of the p_KFD in real-time on the internet is as follows. The user reads instruction on how to operate the system (estimated time is 5 to 10 minutes). The system offers questionnaires and collects the user’s answers (5 to 10 minutes). With well-designed questionnaires, the system can obtain valuable information about the client’s environment and history. Depending on the answers regarding family members, the system offers various types of patterns. The user selects a few among these patterns, arranges them, and expands or contracts their sizes (15 to 20 minutes). In this process, the system can evaluate the client’s sense of balance. Next, the user colors the patterns on the computer using computer brushes with three different thicknesses (20 to 25 minutes) or the client can print the completed KFD with the patterns on an A4-sized paper and then colors them using crayons or markers, makes a digital copy using

Chapter 12 Computerized Kinetic Family Drawing Using Given Patterns (p_KFD)

a scanner or digital camera, and sends it to the system. By analyzing the client’s coloring work, the system can evaluate three elements: the degree of concentration, completeness, and accuracy. The system evaluates all the elements of a KFD drawing or detects changes of evaluations in several KFD drawings, interprets, and informs the clients of the results (in less than a minute). The system can provide art therapists with various valuable information about element evaluation such as the number of used colors and the area colored. It can also show whom among the family members the client omitted in the picture, whether the family members are represented in a big or small sizes, or how far each family member is placed in the picture, etc., by evaluating the various elements in the KFD drawings with patterns. This way, it can help art therapists in their decision-making on the client’s psychological status and treatment process by interpreting one or more KFD drawings. We hope this kind of interdisciplinary work incorporating art therapy and computer science can not only enhance the use of various art therapy tools, but also pave the way for the development of new technology for the analysis and application of art therapy, as well as its theoretical understanding.

12.2 Questionnaires with fact base In the first stage, the system provides two questionnaires. One is used to obtain information on the client’s gender, age, and family relationship as shown in Figure 12.1. From the client’s answers to this questionnaire, we constitute a questionnaire-related FB, such as (QF1) The client is a 7-year old girl. (QF2) She has both parents and one brother. The second questionnaire is used to obtain information on the client’s color preferences, such as his or her favorite color, most disliked color, father’s favorite color, mother’s favorite color, etc. as shown in Figure 12.2 (The same Figure for the SMC in Chapter 11). From the client’s answers, we constitute an FB, such as (QF3) The client’s favorite color is red. (QF4) The client’s most disliked color is black. (QF5) Her father’s favorite color is blue. (QF6) Her mother’s favorite color is green. Art therapists need sufficient information about the client’s background in order to interpret his or her drawing accurately. Cultural traditions and behavioral patterns of society in which he or she grew up need to be taken into consideration. Documenting interactions and dialogues between the client and art therapists can also lead to critical information in interpreting

187

188

PART TWO ART INTERPRETATION

Figure 12.1 Questionnaire on family relationships.

a drawing. Carefully-designed questionnaires can facilitate these processes.

12.3 Composition and coloring In the second stage, as can be seen in Figure 12.3, the system prepares in advance various kinds of patterns of family members and backgrounds, which would compose a KFD drawing, and provides them to the client. Figure

Chapter 12 Computerized Kinetic Family Drawing Using Given Patterns (p_KFD)

189

Figure 12.2 Questionnaire on color preferences.

12.3 shows a computer screen on which patterns are provided to a 7-year old girl who has parents and an elder brother. The screen is divided into four windows; the working window which forms the largest part of the screen, the family window on the right, the background window on the lower part of the screen consisting of four sub-windows, viz. the compartment, form, shape, and place, and the color window on the left. In the family window, when the father is selected, various patterns pertaining to him are shown: reading, watching TV, talking, playing with his children, working, scolding, drinking, etc. The typical activities of the mother include cooking, washing

190

PART TWO ART INTERPRETATION

Figure 12.3 Modification of family member pattern.

dishes, making the beds, playing with her family, sweeping, ironing, and planting. For a male child, the patterns include playing, eating, throwing a ball, riding a bicycle, and watching TV, and for female child, playing, eating, walking, riding a bicycle, and washing dishes (Burns, 1982). By selecting patterns in the compartment window, the client can divide the paper into several parts. The form window includes the patterns of triangles, squares, circles, arrows, lines, dotted lines, etc. The symbol window includes a vacuum cleaner, a table, chairs, a piano, washing machines, newspapers, books, scissors, knives, a dresser, dolls, a bed, an iron, a sofa, a refrigerator, a lawn mower, a garbage can, balls, a skipping rope, a kite, a ladder, a car, a train, puddle, a house, a building, a butterfly, a cat, a snake, a dog, flowers, trees, leaves, logs for animals and plants, rain, snow, the sun, the moon, stars, clouds, hail, etc. The place window includes a room, a bedroom, a kitchen, a living room, a study room, a bathroom, indoors, the front of a house, a school, the front of a building, downtown, shopping centers, restaurants, hospitals, a swimming pool, a church, outdoors, a playground, inside a car, a garden, a fishing place, a mountain, a sea, a road, etc. The client clicks on each window with a mouse to select a pattern, and moves the pattern to the working window by dragging it. As can be seen in Figure 12.3, the pattern of each family member is decomposed into several parts, and the client expands or contracts each part through the repeated use of the “↑,” “↓,” “→,” and “←,” keys, which correspond to vertical expansion and reduction, and horizontal expansion and reduction, respectively.

Chapter 12 Computerized Kinetic Family Drawing Using Given Patterns (p_KFD)

191

Figure 12.4 A KFD drawing composed through the system.

The system does not offer the function of expanding or reducing the pattern’s size at the same horizontal-to-vertical ratio, so that the user’s sense of balance and functions of fingers can be evaluated. The pattern can also be rotated in three axes: The user can repeatedly push “” to rotate the pattern on the x-axis, “ [ ” and “ ] ” to rotate it on the y-axis, and “ { ” and “ } ” to rotate it on the z-axis. The pattern on the working window can also be deleted by dragging it from the working window to its original window. Then, the client colors the patterns by choosing one of 15 colors and a brush with three thicknesses in the color window, so that the system can evaluate the completeness, accuracy, and degree of concentration in the client’s coloring work. This is how the client composed the KFD in the working window in Figure 12.4, in which her brother is watching TV, her father is reading the newspaper lying on a sofa, her mother is cooking, and the client herself is studying. We note that compartmentalization exists between every pair of the family members and that her mother is turning her back to the client.

12.4 Evaluation of elements and detection of changes 12.4.1 Evaluation of elements in the KFD In the third stage, the system automatically evaluates the elements in the KFD. We examined as many elements as possible in various literatures,

192

PART TWO ART INTERPRETATION

including 30 elements in the study of Burns and Kaufman (1972), 7 in McPhee and Wegner (1976), 20 in Myers (1978), 32 in Reynolds (1978), 23 elements in the DDS, the 27 in the PPAT, and 19 in the C_CREATES (Kim, Bae, & Lee, 2007; Kim, 2010). After consolidating those elements which have different names but are actually the same, such as “space usage” in the DDS and “space” in the Formal Elements Art Therapy Scale (FEATS) (Gantt & Tabone, 1998), and “idiosyncratic color” in the DDS and “color fit” in the FEATS, we found that there are 127 unique elements. By using the basic built-in functions of the computer and developing computer algorithms, the system can evaluate all but 5 of the 127 elements. These 5 elements such as folding, drawing on the back, etc. are irrelevant and thus omitted in the p_KFD. And 2 new elements, the sense of balance and finger functions, can be added. Over all, the system rates 124 elements, which are classified into five categories: 33 form-related elements, 58 action-related elements, 14 color-related elements, 16 space-related elements, and 3 performance-related elements. The evaluations of the elements constitute an FB in the same manner as the answers to the questionnaires do. Form-related FB The 33 form-related elements include family members or omission of family members, body, objects, other forms, etc. The results of these 33 form-related elements constitute a form-related FB. For example, (FF1) There is a light. (FF2) There is a TV. Action-related FB Likewise, the results of the 58 action-related elements, including the action of family members, communication, cooperation, etc., constitute the action-related FB. For example, (AF1) Her father is reading the newspaper lying on a sofa. (AF2) Her mother is cooking in the kitchen with her back to the client. (AF3) The client is studying. (AF4) Her brother is watching TV. Color-related FB The results of the 14 color-related elements, including the number of used colors, the list of colors, the area of each color, the number of clusters, etc., constitute the color-related fact base. (The elements of drawings in Part One will be mentioned also in italics.) For example, (CF1) The number of used colors is five. (CF2) The colors used are black, red, green, blue, and yellow in the order of the largest colored area. (CF3) The colored areas consist of 36,256 (black), 32,086 (red), 24,896 (green), 19,821 (blue), and 14,341 (yellow) pixels. (CF4) The number of clusters is 41. We note that the C_CREATES can be used in the evaluation.

Chapter 12 Computerized Kinetic Family Drawing Using Given Patterns (p_KFD)

193

The results of the 16 space-related elements, including the compartment, Space-related FB barrier, location, length, area, distance between family members, space usage, etc., constitute the space-related FB. We note that the sizes can be evaluated not only by their length but also by their area. For example, (SF1) Her father is compartmentalized. (SF2) Her mother is compartmentalized. (SF3) The client is compartmentalized. (SF4) There is a barrier between the client and her father. (SF5) There is a barrier between the client and her mother. (SF6) There is a barrier between the client and her brother. (SF7) There is a barrier between her father and her mother. (SF8) There is a barrier between her father and her brother. (SF9) There is a barrier between her mother and her brother. (SF10) The client is located in the lower part of the drawing. (SF11) Her father is located at the top of the drawing. (SF12) The length of her father is 327 pixels (11.45 cm). (SF13) The area of her father is 25,594 pixels (31.35 cm2). (SF14) The length of her mother is 5.60 cm. (SF15) The area occupied by her mother is 1,054 cm2. (SF16) The length of her brother is 6.30 cm. (SF17) The area occupied by her brother is 12.13 cm2. (SF19) The length of the client is 5.53 cm. (SF20) The area occupied by the client is 9.45 cm2. Finally, the results of the 3 performance-related elements including com- Performance-related pleteness, accuracy and degree of concentration in coloring the patterns, FB constitute the performance-related FB. For example, (PF1) The completeness is 70.8%. (PF2) The accuracy is 54.1%. (PF3) The degree of concentration is 18%. Some of the facts are summarized and provided in the forms of Figure 12.5 and Figure 12.6. 12.4.2 Detection of changes in two KFD drawings Now, suppose the client composed another KFD drawing, as shown in Fig- Changes-related FB ure 12.7, which shows a traditional Korean family watching TV on the floor without chairs or a sofa. In the same manner, the system interprets this second KFD drawing. Also, from the evaluation of the elements in the two KFD drawings, the system can detect any changes in the facts and construct a changes-related FB. For example, (HF1) Her father is lying down and reading the newspaper in the 1st session and is watching TV in the 2nd session. (HF2) Her father is compartmentalized in the 1st session and not in the 2nd session.

194

PART TWO ART INTERPRETATION

Figure 12.5 Results of the analysis of the color-related elements.

(HF3) Her father is located at the top of the drawing in the 1st session and not in the 2nd session. (HF4) The length of her father is 327 pixels (11.45 cm) in the 1st session and 269 pixels (9.41 cm) in the 2nd session. (HF5) The area of her father is 25,594 pixels (31.35 cm2) in the 1st session and 28,936 pixels (35.45 cm2) in the 2nd session. (HF6) Her mother turned her back to the client in the 1st session, but not in

Chapter 12 Computerized Kinetic Family Drawing Using Given Patterns (p_KFD)

195

Figure 12.6 Results of the analysis of the space-related elements.

the 2nd session and is ironing. (HF7) Her mother is compartmentalized in the 1st session, but not in the 2nd session. (HF8) The length of her mother is 5.60 cm in the 1st session and 8.89 cm in the 2nd session. (HF9) The area of her mother is 1,054 cm2 in the 1st session and 13.67 cm2 in the 2nd session. (HF10) Her brother is watching TV in both the 1st and 2nd sessions.

196

PART TWO ART INTERPRETATION

Figure 12.7 A KFD drawing in the 2nd session.

(HF11) The length of her brother is 6.30 cm in the 1st session and 8.26 cm in the 2nd session. (HF12) The area of her brother is 12.13 cm2 in the 1st session and 18.06 cm2 in the 2nd session. (HF13) The client is studying in the 1st session and watching TV in the 2nd session. (HF14) There is a light in the 1st session, but not in the 2nd session. (HF15) The client is compartmentalized in the 1st session, but not in the 2nd session. (HF16) The client is located at the lower part of the drawing in the 1st session and at the center in the 2nd session. (HF17) The length of the client is 5.53 cm in the 1st session and 8.64 cm in the 2nd session. (HF18) The area of the client is 9.45 cm2 in the 1st session and 24.37 cm2 in the 2nd session. (HF19) There is a barrier between the client and her father in the 1st session. The distance between them in the 2nd session is 13.4 cm. (HF20) There is a barrier between the client and her mother in the 1st session. The distance between them in the 2nd session is 7.3 cm. (HF21) There is a barrier between the client and her brother in the 1st session. The distance between them in the 2nd session is 7.3 cm. (HF22) There is a barrier between her father and her mother in the 1st session. The distance between them in the 2nd session is 19.1 cm.

Chapter 12 Computerized Kinetic Family Drawing Using Given Patterns (p_KFD)

197

(HF23) There is a barrier between her father and her brother in the 1st session. The distance between them in the 2nd session is 6.2 cm. (HF24) There is a barrier between her mother and her brother in the 1st session. The distance between them in the 2nd session is 13.5 cm. (HF25) The number of used colors is 5 in the 1st session and 7 in the 2nd session. (HF26) The colors the most used in the 1st session are black, red, green, blue, and yellow in the order of the largest area colored, and those in the 2nd session are blue, orange, black, yellow, red, green, and purple. (HF27) The areas colored in the 1st session are 36,256 (black), 32,086 (red), 24,896 (green), 19,821 (blue), and 14,341 pixels (yellow), and those in the 2nd session are 255,792 (blue), 30,085 (orange), 26,338 (lack), 19,685 (yellow), 8,780 (red), 6,774 (green), and 4,279 pixels (purple). (HF28) The numbers of clusters in the 1st and 2nd sessions are 41 and 37, respectively. (HF29) The completeness in the 1st and 2nd sessions are 70.8% and 74.7%, respectively. (HF30) The accuracies in the 1st and 2nd sessions are 54.1% and 78.6%, respectively. (HF31) The degree of concentration is 18% in the 1st session and 62% in the 2nd session.

12.5 Interpretation with knowledge base In the final stage of interpretation, a KB is constructed in advance from the expertise and experience of experts and literatures. Knowledge in the KB is expressed in an IF-THEN format. The knowledge is also classified into 5 categories. For example, [FK1] IF the client is omitted from the KFD, Form-related KB THEN he or she is a sexually abused child (Cohen & Phelps, 1985). [FK2] IF the client draws a family member with his or her back to the client, THEN he or she has a negative attitude and suppressed anger towards that family member. [FK3] IF there are objects such as a light, a lamp, a TV, or the sun, THEN the client desires love or affection. [FK4] IF there is a form with a sense of direction lost, THEN the client is neglected by his or her family. [AK1] IF a boy draws himself riding a bicycle or a horse, Action-related KB THEN he has learning problems, anxiety, fear, attention deficits, hy-

198

PART TWO ART INTERPRETATION

peractivity, or conduct disorder (Stawar & Stawar, 1987). [AK2] IF a boy draws himself playing, THEN he is normal (Stawar & Stawar, 1987). [AK3] IF the family members are doing a same action, THEN there is good communication and cooperation among them. Color-related KB [CK1] IF two or less colors are used, THEN the child is severely sexually abused or depressed (Malchiodi, 1990; Wadeson, 1980; Gantt & Tabone, 1998). [CK2] IF a limited number of used colors, not more than two or three, mostly consisting of black, white, and sometimes red is used, THEN the client experienced a natural disaster such as earthquakes, hurricanes, and plane crashes (Gregorian, Azarian, DeMaria, & McDonald, 1996). [CK3] IF the percentage of red color is over 50%, THEN the child is in an abnormal psychological and/or physical state, often thought as violence, passion, aggression, and anger (Hammer, 1953, 1969; Klepsch & Logie, 1982; Precker, 1950). [CK4] IF the percentage of yellow color is over 50%, THEN the child is often perceived as a likable person. (Hammer, 1953, 1969; Klepsch & Logie, 1982; Precker, 1950). [CK5] IF the percentage of brown color is over 50%, THEN the child is often thought of timidity (Hammer, 1953, 1969; Klepsch & Logie, 1982; Precker, 1950). [CK6] IF the color that is most used is the same as the color of the mother, THEN the child is closer to his or her mother than to his or her father. [CK7] IF black is the color the most used, THEN the child is suffering from depression or despair. [CK8] IF the sun is colored black, THEN the child has experienced trauma. [CK9] IF a red house is drawn, THEN the client is a survivor of sexual abuse (Cohen & Phelps, 1985). Space-related KB [SK1] IF the client is isolated from his or her family members, THEN the child has alcoholic (drug-abusing) parents (Holt & Kaiser, 2001). [SK2] IF there is a barrier between the family members, THEN there are problems in the communication or relationship between them. [SK3] IF a boy draws himself very close to his mother, THEN he has learning problems, anxiety, fear, attention deficit, hyperactivity, or conduct disorder (Stawar & Stawar, 1987). [SK4] IF there are compartmentalized family members,

Chapter 12 Computerized Kinetic Family Drawing Using Given Patterns (p_KFD)

199

THEN the child is estranged from his or her family members, or his or her family members are estranged from one another in terms of their communication and/or relationships. [SK5] IF a family member is drawn very big, THEN he or she is either admired or authoritative. [SK6] IF the distance between two family members is close, THEN they are very intimate. [SK7] IF the distance between two family members is far, THEN they are emotionally distant. [SK8] IF a family member is drawn in the lower part of the picture, THEN he or she is depressed or is less active than usual. [SK9] IF a family member is located at the top of the drawing, THEN he or she is a key member of the family and either admired or authoritative. [SK10] IF there is no compartment in the family, THEN the family members communicate well with one another and have good relationships with each other. [PK1] IF the completeness is 60% or less, THEN the child easily feels bored in his or her everyday life. [PK2] IF the accuracy is 60% or less, THEN the child has an attention deficit disorder. [PK3] IF the degree of concentration is 20% or less, THEN the child has an ADHD.

Performance-related KB

There is also a changes-related KB. For example, Changes-related KB [HK1] IF a family member has disappeared or has newly appeared, THEN special caution is required with regard to the interpretation of that family member. [HK2] IF the number of family members doing the same action has increased, THEN the home atmosphere has improved. [HK3] IF a static action changes into a dynamic one, THEN the client has stabilized emotionally. [HK4] IF the number of used colors has increased by two or more, THEN the client’s depression symptoms have been alleviated and he or she has become more cheerful. [HK5] IF the color which is most used has changed from cool to warm, THEN the client has become freer in expressing his or her emotions. [HK6] IF the color which is most used has changed from a primary to a secondary one, THEN the client has two types of psychological states and is undergoing a big change. [HK7] IF the color which is most used changes,

200

PART TWO ART INTERPRETATION

THEN the corresponding interpretation will likewise change. [HK8] IF the compartmentalization has disappeared, THEN the communication and relationships between the compartmentalized family members and the other family members have improved. [HK9] IF the direction of the face of a certain family member has changed from turning her back to the client to facing him or her, THEN the client’s negative attitude and suppressed anger towards him or her have been resolved, and the client has begun thinking about him or her more positively. [HK10] IF a family member in the lower part of the picture has moved to the center, THEN this person, who is depressed and less active than usual, has been recognized as a central family member by the family. [HK11] IF the placement of a family member changes, THEN the corresponding interpretation will also change. [HK12] IF the completeness has increased by 20% or more, THEN the client’s aversive character has improved. [HK13] IF the accuracy has increased by 20% or more, THEN the client’s attention deficit disorder has been alleviated. [HK14] IF the degree of concentration has increased by 20%, THEN the client’s emotional state has stabilized. The system interprets a KFD drawing by activating the knowledge corresponding to the facts derived in the previous stages. For example, from the fact SF4 in the FB which indicates that there is a barrier between the client and her father, the system activates the knowledge SK2, in the KB that IF there is a barrier between family members, THEN there are problems in the communication or relationship between them. In other words, the system interprets that there are problems in the communication or relationship between the father and the client. Also, from the fact AF2 in the FB which shows that the mother is cooking in the kitchen with her back to the client, the system invokes the knowledge FK2, in the KB that IF the client draws a family member with his or her back to the client, THEN the client has a negative attitude and suppressed anger towards that family member. In other words, the system interprets that the client has a negative attitude and suppressed anger towards her mother. Likewise, from the fact SF4, the system activates the knowledge SK9. In other words, the system interprets that her father is a key member of the family, and either admired or authoritative. Now, we have these interpretations. One KFD drawing (I1) There are some problems in the communication or relationship between the client and her father. (I2) There are some problems in the communication or relationship between

Chapter 12 Computerized Kinetic Family Drawing Using Given Patterns (p_KFD)

201

the client and her mother. (I3) There are some problems in the communication or relationship between her father and her mother. (I4) There are some problems in the communication or relationship between the client and her brother. (I5) The client is depressed, less active than usual, or in a state of psychological shock or despair. (I6) Her father is the central member in her family. (I7) The client has a negative attitude and suppressed anger towards her mother. (I8) Her father is a key member of the family and either admired or authoritative. (I9) The client easily feels bored in her everyday life. (I10) The client is suffering from ADHD. The system also activates the knowledge in the changes-related KB corresponding to the fact in the changes-related FB. Art therapists are interested in the treatment process (Betts, 2006). The assessment of elements of a drawing is done in various steps: the beginning, the middle, and the end of the treatment. Finger (1997) showed that KFD is useful in determining whether there is any change before and after the treatment. For example, the system interpreted the above two KFD drawings. (DC1) It can be said that the communication and relationship between the Two KFD drawings client and her father has improved. (DC2) It can be said that the communication and relationship between her father and her brother, and the overall mood between them have improved. (DC3) It can be said that the communication and relationship between the client and her brother have improved. (DC4) The client’s negative attitude and suppressed anger towards her mother have been resolved, and she has begun thinking about her mother in a more positive light. (DC5) The client’s desire for love or affection has diminished. (DC6) The client’s symptoms of depression have been alleviated, and she has become more cheerful. (DC7) The client’s depression or despair appears to have been alleviated. (DC8) The client’s characteristic of becoming easily bored has changed. (DC9) The symptoms of the client’s attention deficit disorder have been alleviated. (DC10) The client’s emotional state has stabilized.

12.6 Discussion and conclusion

202

PART TWO ART INTERPRETATION

As a media for art therapy, we have provided given patterns intended for coloring, instead of having the client draw something. The SMC is a typical type of this pattern coloring whose pattern is given as a geometric one within a circle. Another kind of pattern coloring is the Face Stimulus Assessment (FSA) (Betts, 2003) whose pattern is a human face. Also, the p_KFD in this chapter is a pattern coloring where patterns are family members doing something. In this manner of providing patterns all art therapy tools such as the DAP, the HTP, the PPAT, and even free drawing can be computerized. The prototype computer system, the p_KFD, can automatically provide art therapists with accurate, precise, objective, quantitative, and consistent evaluation with 100 % reliability for the 124 elements in hundreds of the KFD drawings in a minute by using built-in functions of the computer and the C_CREATES for art evaluation. Also, the p_KFD aids art therapists in making their decision on art therapy treatment for clients by interpreting one or several KFD drawings. It is expected that the p_KFD with patterns would inherit the original applicability, usefulness, and validity of the traditional KFD with free drawings. The system can evaluate the sizes of forms, not only in terms of their length, but also in terms of their area. Moreover, the p_KFD evaluates the client’s sense of balance in the process of composing a drawing, and the completeness, accuracy, and degree of concentration in the coloring patterns. These are elements in the C_CREATES which are newly added to the traditional KFD. The p_KFD can be used as an experimental tool to set the criteria for evaluation and interpretation of the traditional KFD. For example, it could find and specify how small the resulting picture should be in the evaluation to be judge one as having low self-esteem, or how large the resulting distance between family members should be to be seen as a sign of a child’s psychological sense of alienation. Since the p_KFD offers a new, easy, and interesting way of composing a drawing and coloring, compared to the traditional KFD, it may be more appropriate for clients who have severe developmental, psychological, or physical disorders. Also, the p_KFD could offer the emotional stabilization and the improvement of the client’s psychological state during the work of selection, arranging, modifying, and coloring given patterns, which are reported in coloring structured mandalas by Curry and Kasser (2005). There is a huge amount of expertise, experience, case studies, and results of experts who are world renowned authorities in art therapy, reports, papers, books, and literatures. Millions of items can be stored in a KB as a kind of dictionary, and from this, the p_KFD can retrieve each item in a second. In conclusion, as a worldwide authority of art therapy, the p_KFD can not only interpret a single KFD drawing, but also detect any changes, trends, or process of art therapy treatment from a series of KFD drawings. In doing this, the system never shirks its duty, is never fatigued with its responsibility, and never makes a mistake of omission. We hope the p_KFD

Chapter 12 Computerized Kinetic Family Drawing Using Given Patterns (p_KFD)

can be developed as a standardized instrument. With its reliability, validity, and expert system functions, the system can be a new methodology to be extended to the Human Figure Drawing (HFD) (Koppitz, 1968), the DAP (Neglieri, McNeish, & Bardos, 1991), the HTP, the PPAT, etc., with patterns until computer technology progresses to the point where it is able to identify the forms and symbols in a free drawing. It has sometimes been reported that the usefulness of projective techniques such as the KFD, the HFD, the DAP, and the HTP and the interpretation of art is doubtful, due to their lack of reliability and highly questionable validity. This might be due to the fact that some elements are too complex for raters to evaluate perfectly and that people and art are too complex for drawings to perfectly embody a given interpretation (Cohen, 1986/1994). In particular, the interpretation of children’s drawings is confounded by maturational issues (Kahill, 1984). However, we cannot deny the contribution and usefulness of projective techniques. From the viewpoint of artificial intelligence, the field of art therapy is a so-called ISP where numerous studies and methods have produced results that are diverse, lacking for consistency and sometimes even contradictory (Kim, 2008b). There are many other important disciplines that are associated with the ISP, including clinical diagnosis, weather forecasting, and demand forecasting. Usually, there are several typical problems in the ISP: First, which item should be included in the KB and which item should not be; Second, some items have very high validity while others have much less validity; Third, the consideration of individual situations. The approach of an expert system in artificial intelligence is known to be a very useful solution to these kinds of problems. For the first problem, the system may be used in experiments for the validity of knowledge. For the second problem, the system in Chapter 10 can give some sort of validity measure to each item. For example, 0 for no evidence, 1 for weak evidence, 2 for possibility, 3 for moderate evidence, 4 for strong evidence, and 5 for absolute evidence. Also, the Bayesian network (Charniak, 1991) approach in the last chapter could be adopted. For the third problem, the system can consider a client’s individual characteristics, environment, background, history, etc., with suitably designed questionnaires and the exceptional memory capability of the computer.

203

Computerized Structured Mandala Coloring (c_SMC) for Differentiation and Identification of Psychological States Using Statistical Methods Chapter 13 Key points Rubin (1986) has reported that it was not possible to identify schizophrenic subjects from normal subjects only from a drawing. Only three out of 40 judges could identify beyond chance the normal children’s artworks from schizophrenic children’s. Veltman and Browne (2001) have reported that two developmental psychologists failed to identify abused children from non-abused children using either the Favorite Kind of Day (Manning, 1987) or the Kinetic Family Drawings (Burns & Kaufman, 1972). Numerous studies have warned against using only projective methods to identify abused, neglected, or indifferently treated children (Veltman & Browne, 2000a, 2000b). The Computerized Structure Mandala Coloring (c_SMC) developed in this chapter (Kim, Betts, et al., 2009) applies the C_CREATES along with statistical methods of factor analysis and regression analysis to differentiation and identification of psychological states. While it relies on projective methods, it is nevertheless capable of not only differentiating the four groups of non-patients, anxious patients, depressed patients, and schizophrenic patients, but also identifying each group on the basis of only a single SMC drawing. The methodology of the c_SMC can also be applied to patients with psychological states or disorders not mentioned above. Moreover, it can be used for any types of drawings with patterns other than the structured mandala.

13.1 The Structured Mandala Coloring (SMC) as a subject of computerization There are two types of mandala used as art therapy tools: drawing free figures in a circle (unstructured mandala) and coloring a given pattern in a circle (structured mandala). This chapter focuses on structured mandalas colored by four different groups: non-patients, anxious patients, depressed patients, and schizophrenic patients. The main reason for choosing the structured mandala is that is that all important elements it contains are suitable 204

Chapter 13 c_SMC for Differentiation and Identification of Psychological States

205

for analysis and evaluation by the C_CREATES. The detection of forms, symbols, lines, and movements of unstructured mandalas or ‘free’ drawings is beyond the capacity of current computer technology (Kim, Kang, & Kim, 2009). The SMC can be considered as prestructured art in the sense that one simply colors a given pattern and it can be utilized in a short-term art therapy environment (Vick, 1999). Curry and Kasser (2005) reported that coloring a structured mandala and coloring a plaid design are equally effective for anxiety reduction. Vennet and Serice (2012) concluded that coloring a structured mandala reduces anxiety to a significantly greater degree than coloring a plaid design or coloring on blank sheet. In the preceding chapter, we developed an expert system that interprets drawers’ relationships with their families in relation to the colors they used in the SMC (Kim, Kim, & Kim, 2008). With regard to the elements in the SMC, we evaluated completeness and accuracy using the C_CREATES in Chapter 2, and degree of concentration in Chapter 7 (Kim, Kang, & Kim, 2009). In addition, we will analyze the correlation between the elements in the SMC and the levels of dementia in the next chapter (Kim, Betts, et al., 2009). One of the purposes of art therapy assessments is to gather information Computerization of on patients’ clinical state and to provide a psychiatric diagnosis. The DDS SMC is designed to gather clinical information on a client in a single session. The FEATS examines non-symbolic aspects of art and demonstrates how the structural characteristics (elements) of a drawing furnishes information about a person’s clinical state and gives a psychiatric diagnosis. This chapter examines information that can be derived from a SMC drawing (or simply mandala hereafter) colored by patients with the three aforementioned psychological disorders, i.e. anxiety, depression, and schizophrenia. We develop a system named Computerized Structured Mandala Coloring (c_SMC) (Kim, Betts, et al., 2009) that evaluates the elements in the SMC, differentiates the above four groups, and identifies the group that drew a given mandala. Here when we say an element of drawings “differentiates” group A and group B, it means that its evaluation results are different statistically at the given significance level between the two groups, and when an element “identifies” group A from group B, it means that from its evaluation results we can determine whether the drawing is made by group A or group B with Type I and Type II errors. The c_SMC evaluates 32 elements of a structured mandala. Some of these elements are included in the C_CREATES and others can be directly generated by the elements in the C_CREATES. In contrast to the elements of an unstructured mandala, the elements of a structured mandala, in which geometric lines are given, can be automatically evaluated by the C_CREATES. Through the application of digital image processing techniques, the C_CREATES can automatically and quantitatively analyze the 32 elements

206

PART TWO ART INTERPRETATION

in the c_SMC, including number of used colors, list of colors, area of each color, the number of clusters, length of edges, main color, completeness, accuracy, etc., all of which are either explicitly or implicitly related to color. Differentiation and Most art therapists would likely to be at a loss if they were presented with identification of the only one SMC drawing and asked to guess which patient group the drawer c_SMC belongs to. Therapists must consider other possible supplementary information, such as the drawer’s relationships with family or others; cultural environments, which affect their values, languages, and traditions; and the behavioral patterns of the society in which he or she was raised or currently resides. They are discouraged to rely exclusively on the singular aspect of a single drawing in making decision. Thus, most art therapy assessments have been limited to finding statistically significant elements that differentiate the groups. For example, Reddy et al. (2002) pointed out that in the KFD those done by neglected children reveal differences from those done by children raised in a normal family settings. The results showed that they differed in drawing their family members. Children from a neglected family were more likely to put the family members ‘irregularly’ and are more ‘reluctant’ to draw them. The validity of the DDS and the FEATS has been tested on groups of patients with various psychiatric symptoms. For example, the elements of light pressure, unusual placement, water images, etc., in the DDS showed differences in the groups of non-patients, dysthymia patients, depressed patients, and schizophrenic patients (Cohen, Hammer, & Singer, 1988). Three elements, prominence of color, details of objects and environment, and line quality, in the FEATS showed differences between children with ADHD and children with no learning or behavioral problems (Munley, 2002). Rubin (1986) reported that it was not possible to identify (distinguish) the schizophrenic subjects from the normal subjects only from evaluating their drawings. Only three out of 40 judges could identify beyond chance the normal children’s artworks from schizophrenic children’s. Veltman and Browne (2001) reported that two developmental psychologists failed to identify abused children’s Favorite Kind of Day (FKD) (Manning, 1987) and KFD drawings from non-abused children’s. From the samples of 23 non-abused children and five abused children, which is a relatively small sample size, the Type I error (mistaking non-abused children’s drawings for abused children’s) for the FKD and the KFD is 1/23 and 4/23, respectively, and the Type II error (mistaking abused children’s drawings for non-abused children’s) is 4/5 and 3/5. Veltman and Browne (2003) argued that the KFD should not be used for the diagnosis of physical, sexual, or emotional abuse, and should instead be considered only as part of a more formal and in-depth investigation. Numerous studies have warned against using only projective methods to identify abused, neglected, or indifferently treated children (Veltman & Browne, 2000a, 2000b). Nevertheless, the c_SMC proposed in

Chapter 13 c_SMC for Differentiation and Identification of Psychological States

this chapter does not only differentiate several groups based on significant elements in the c_SMC, but also identifies a group from another group on the basis of only one given SMC drawing. First, the c_SMC finds, through the factor analysis, statistically significant elements in the c_SMC that differentiate two groups in the following four cases: (1) non-patient and patient, (2) non-patient and anxious patient, (3) non-patient and depressed patient, and (4) non-patient and schizophrenic patient. Here, the patients are separated into three groups, each suffering from anxiety, depression, or schizophrenia. Next, using the regression analysis, the c_SMC estimates the probability that one group, not the other, colored the mandala, and based on this probability, identifies which group colored the mandala. The c_SMC provides art therapists with quantitative results of element evaluation, group differentiation, and group identification, which are useful for objective decision-making.

13.2 Methods 13.2.1 Sampling We obtained 201 samples of structured mandalas with the same pattern, as shown in Figure 2.1a in Chapter 2, from people without symptoms of psychological disorders, 100 from anxious patients, 94 from depressed patients, and 100 schizophrenic patients. As the structured mandala pattern has moderate complexity which is neither too simple for non-patients nor too complex for patients, it is expected to reflect differences in elements among these four groups. The participating patients were diagnosed by a psychiatrist based on DSM-IV-TR (2000) criteria and the non-patients were employees of a hospital with no current or past mental illness as determined by the same psychiatrist based on Symptom Checklist-90-Revision (Derogatis, 1992). In accordance with the regulations of Korean Art Therapy Association (2013), we obtained informed consent from the participants after explaining the purposes of the study, procedures and expected effects in a plain layman’s language. The study was approved by the Institutional Review Board of the hospital. We conducted a number of single sessions to collect the SMC samples over a two-year period. Participants were provided with a box of crayons manufactured by Titi Co. Ltd containing 12 colors: red, orange, yellow, light green, green, azure, blue, purple, brown, reddish brown, white, and black. In the rare cases in which white was used, it was considered as no color used for the sake of computer analysis. A session began with the instruction: “Please express your emotional state in colors. Please do not chat or discuss with other people.” All the participants were given 30 minutes to complete the task 13.2.2 Measurements

207

208

PART TWO ART INTERPRETATION

In a SMC sample there are 32 elements that can be analyzed by the C_CREATES. They consist of the number of used colors, number of clusters, warm / cool colors, primary / secondary colors, complementary colors, completeness, length of edges, accuracy, 11 main colors of crayons, and areas of 11 colors. We remind the readers of the definitions of the following terms: A cluster is a contiguous area in the same color, not separated by different colors. An edge consists of pixels in which the color differs from that of its neighboring pixels. A main color is the color which is believed to be most meaningful in a drawing. However, in the case of structured mandala, we consider the color with the largest colored area as the main color. Completeness refers to the colored portion of a SMC drawing, and accuracy refers to the preciseness of coloring a structured mandala in accordance with the given pattern. 13.2.3 Data analysis One can use factor analysis to find statistically significant elements that differentiate the patient group and the non-patient group, such as number of used colors. The average numbers of colors used by non-patients and patients were 10.93 (1.22) and 7.84 (2.27), respectively, which indicate a statistically significant difference at the level below 0.05. The numbers in the parentheses are standard deviations. For more understanding of basic statistics, readers are advised to consult textbook such as Probability and statistics for engineers and scientists (Walpole & Myers, 2006). Logistic regression Regression analysis is used to identify one of the two groups. A regression analysis is a statistical methodology that utilizes the relation between two or more quantitative variables so that one variable can be predicted in relation to another or others (Kutner, Nachtsheim, Neter, & Li, 2004). The independent variables are the 32 elements mentioned above. The dependent variable, for example, of group A and group B is defined either as 0 in case of group A or as 1 otherwise. In this chapter, the logistic regression model is used, of which the regression function is (1) Y = exp(Z) / {1 + exp(Z)}, in which (2) Z = a1X1 + a2X2 + ... + amXm. Here, Y is the dependent variable, Xi s are the independent variables, ais are the parameters to be estimated, and exp denotes the exponential function. The stepwise regression is used to select the independent variables that “well” explain the dependent variable. We note that in the case in which the dependent variables are either 0 or 1 as in this case, the estimated value of Y can be interpreted as the probability that it belongs to group B. We identify group B when this probability is greater than 0.5 and identify group A otherwise.

Chapter 13 c_SMC for Differentiation and Identification of Psychological States

13.2.4 System validity The study of Veltman and Browne (2003) focused on the “blind” identification of abused children based on a single drawing without any supplementary information. To measure the power of the identification method in this study, an experiment of 5-fold cross validation (Witten, Frank, & Hall, 2011) has been performed. The 5-fold cross validation divides the sample into five sub-samples, from four of which an identification method is developed (training data). Then the developed method identifies the group for the remaining one sub-sample (test data). Thus, it is the equivalent to performing a “blind” identification on the test data. The experiments are performed five times, changing the training data and the test data.

13.3 Results 13.3.1 Differentiation of groups The C_CREATES evaluated 32 elements of each SMC samples. After the adjustment of a SMC drawing for evaluation, the time taken for the PC with an Intel(R) Core(TM) i5 650 @ 3.20GHz CPU, common specifications in 2010 in which this study was performed, to process one SMC drawing was 0.78 seconds on average. The following 17 elements were observed to significantly differentiate the patient group and the non-patient group at the significance level of below 0.05: number of used colors (0.00), number of clusters (0.00), ratio of warm colors (0.00), ratio of cool colors (0.00), ratio of primary colors (0.00), ratio of secondary colors (0.00), complementary colors (0.00), length of edges (0.00), completeness (0.00), accuracy (0.00), yellow (0.00), green (0.01), azure (0.02), ratio of yellow (0.00), ratio of green (0.00), ratio of brown (0.03), and ratio of reddish brown (0.00). The numbers in parentheses are significance levels. For example, the averages of elements, accuracy, length of edges, and ratio of yellow are 0.780 (0.16), 9,614 (2,038), and 0.17 (0.13), respectively, for the non-patients and 0.399 (0.26), 6,168 (3,362), and 0.09 (0.12), respectively, for the patients. The numbers in parentheses are standard deviations. The following twenty elements were observed to significantly differentiate the anxiety patient group and the non-patient group: number of used colors (0.00), number of clusters (0.00), ratio of warm colors (0.00), ratio of cool colors (0.01), ratio of primary colors (0.00), ratio of secondary colors (0.01), length of edges (0.00), completeness (0.00), accuracy (0.00), yellow (0.00), green (0.00), azure (0.00), brown (0.03), ratio of orange (0.00), ratio of yellow (0.00), ratio of green (0.00), ratio of sky blue (0.00), ratio of purple (0.03), ratio of brown (0.01), and ratio of reddish brown (0.00).

209

210

PART TWO ART INTERPRETATION

The following fifteen elements were observed to significantly differentiate the depressed patient group and the non-patient group: number of clusters (0.00), ratio of warm colors (0.00), ratio of cool colors (0.00), ratio of primary colors (0.00), ratio of secondary colors (0.00), length of edges (0.00), completeness (0.00), accuracy (0.00), yellow (0.03), ratio of yellow (0.00), ratio of green (0.02), ratio of purple (0.02), ratio of brown (0.00), ratio of reddish brown (0.00), and ratio of black (0.01). The following fifteen elements were observed to significantly differentiate the schizophrenic patient group and the non-patient group: number of used colors (0.00), number of clusters (0.00), ratio of warm colors (0.00), ratio of cool colors (0.01), ratio of primary colors (0.00), ratio of secondary colors (0.01), complementary colors (0.00), length of edges (0.00), completeness (0.00), accuracy (0.00), yellow (0.00), green (0.01), ratio of yellow (0.00), ratio of green (0.00), and ratio of black (0.03). 13.3.2 Identification of group Equations (2), which determine regression function (1) to identify one of the two groups were as follows: for the patient group, Z = 7.586 - 0.226 x Number of used colors (0.012) - 6.633 x Accuracy (0.000) - 5.252 x Ratio of yellow (0.000), (3) where the values in the parentheses indicate the p-values of the variables, for the anxious patient group, Z = 4.725 + 9.649 x Ratio of cool colors (0.005) - 10.285 x Accuracy (0.000) + 3.581 x Orange (0.007) - 8.849 x Ratio of yellow (0.001) + 8.640 x Ratio of green (0.000), for the depressed patient group, Z = 4.463 + 0.030 x Number of clusters (0.001) + 8.187 x Ratio of cool colors (0.014) - 0.001 x Length of edges (0.000) - 5.581 x Accuracy (0.000) - 2.822 x Black (0.017) + 6.507 x Ratio of green (0.004) + 11.753 x Ratio of black (0.000), and for the schizophrenic patient group, Z = 7.942 - 0.359 x Number of used colors (0.000) + 0.029 x Number of clusters (0.001) - 0.0005 x Length of edges (0.000) - 4.260 x Accuracy (0.000) - 4.809 x Ratio of yellow (0.000). Based on the regression equation (1) and equation (3) identifying a group, the probability of being a patient increases as number of used colors, accuracy, and ratio of yellow all decreases. There are reports that child victims of severe sexual abuse (Malchiodi, 1990) and depressed patients (Gantt & Tabone, 1998; Wadeson, 1980) tended to use only one or two colors in their drawings. Children who have experienced a natural disaster,

Chapter 13 c_SMC for Differentiation and Identification of Psychological States

211

Figure 13.1 Type I and Type II errors in identifying non-patients and patients based on probability of logistic regression function.

such as an earthquake, hurricane, or a plane crash, tended to use a limited number of colors (generally no more than two or three), and these colors mostly include black, white, and sometimes red (Gregorian, Azarian, DeMaria, & McDonald, 1996). Pianetti, Placios, and Elliott (1964) observed that drawings by patients with chronic schizophrenia showed a lack of detail and small area colored, and that Alzheimer’s disease patients showed significantly fewer number of used colors than control subjects, i.e. healthy people (Rankin, Liu, Howard, Slama, Hou, Shuster, & Miller, 2007). Kim, Betts, et al. (2009) are the first to mention in their study the accuracy in the SMC and report that the degree of dementia increases as accuracy decreases. The same can be inferred in the case of psychological disorders. The color yellow is generally perceived to be lively and energetic and to elicit positive emotions, including happiness and excitement, because it is typically associated with the sun, blooming flowers, and summertime (Kaya & Epps, 2004). When the probability of regression function (1) is less than 0.5, a non-pa- Type I and Type II tient is identified. Otherwise, a patient is identified. When this criterion of errors

212

PART TWO ART INTERPRETATION

Table 13.1 Results of identification in four cases by logistic regression _____________________________________________________________________________________________

Actual Non- Non- Non- Non- Patient Anxiety Depress. Schizo. Estimated patient anxiety depress. schizo.

__________

________________

_________________

_________________

________________

Non-patient 164 37 188 13 194 7 190 11 Patient 54 240 15 85 24 70 34 66 __________ ________________ _________________ _________________ ________________ Type I error 0.18 0.06 0.03 Type II error 0.18 0.15 0.26 0.34

0.05

_____________________________________________________________________________________________ 0.5 is adjusted to a lower or higher value, Type I error increases or decreases, respectively, and Type II error decreases or increases, respectively. Adjusting the criterion will yield a more desirable balance between Type I and Type II errors. These relations are illustrated in Figure 13.1. The results of identifications including Type I and Type II errors are summarized in Table 13.1. Type I errors in the four cases were 0.18 (37/201), 0.06 (13/201), 0.03 (7/201), and 0.05 (11/201), respectively, and Type II errors were 0.18 (54/294), 0.15 (15/100), 0.26 (24/94), and 0.34 (34/100), respectively. 13.3.3 System validity Type I errors of the experiment of 5-fold cross validation in the four cases were 0.19, 0.18, 0.33, and 0.26, respectively, and Type II errors were 0.22, 0.08, 0.11, and 0.11, respectively, showing nearly identical results to the ones in Table 13.1. These results confirm the validity of the method presented in this study. When we identify a group using additional information, such as gender, age, and educational background, the regression function will change, and a more accurate identification can be obtained. In this case, Type I errors in the four cases were 0.14, 0.04, 0.05, and 0.07, respectively, and Type II errors were 0.09, 0.12, 0.21, and 0.15, respectively, which suggests that more accurate identifications can be made in most cases when additional factors are considered. Applying neural networks (Haykin, 2008), the c_SMC attempted to identify the four groups based on their mandala. Although the specific results are

Table 13.2 ________________________________________________________________________ Identification of four Actual Non-patient Anxiety Depression Schizophrenia Error groups by neural networks

Estimated ________________________________________________________________________ Non-patient Anxiety Depression Schizophrenia

136 12 21 22

20 44 28 19

23 27 19 14

22 17 26 45

0.32 0.56 0.80 0.55

________________________________________________________________________

Chapter 13 c_SMC for Differentiation and Identification of Psychological States

213

Figure 13.2 SMC samples in four different cases.

not shown here, we could not obtain a satisfactory identification. The results of 5-fold cross validation when the neural networks were applied are shown in Table 13.2. The hit probabilities, i.e. correct identifications, for the four groups were 0.68, 0.44, 0.20, and 0.45, respectively.

214

PART TWO ART INTERPRETATION

Table 13.3 Identification of groups based on the SMC samples _____________________________________________________________________________________________





Mandala A B C D E F G H

_____________________________________________________________________________________________ Number of colors 5 5 10 8 5 6 6 3 Accuracy 0.90 0.54 0.89 0.93 0.52 0.27 0.89 0.37 Ratio of yellow 0.26 0.14 0.27 0.13 0.00 0.09 0.28 0.00 _____________________________________________________________________________________________ Probability 0.29 0.89 0.12 0.25 0.95 0.98 0.24 0.99 _____________________________________________________________________________________________ Identification Non. Patient Non. Non. Patient Patient Non. Patient _____________________________________________________________________________________________ Actual

Non.

Non.

Non.

Non. Patient Patient Patient Patient

_____________________________________________________________________________________________

Example Eight SMC samples identifying patient groups from the non-patient group are shown in Figure 13.2. Number of used colors, accuracy, and ratio of yellow to other colors are evaluated by the C_CREATES and the results are shown in Table 13.3. The probabilities of patient groups by regression function (1) were given by equation (3) based on these evaluations. The results of identifications are presented according to the probability of either greater than or not greater than 0.5. In addition, the actual classifications identified whether the mandala was drawn by the patient group or the non-patient group. As can be seen in Table 13.3, Type I error was committed in mandala B and Type II error in mandala G.

13.4 Discussion and conclusion The c_SMC, which utilizes the C_CREATES and the statistical methods of factor analysis and regression analysis, has been developed to gather information on subjects’ psychiatric conditions from their SMC works. A total of 495 samples were obtained from four groups: non-patients, anxious patients, depressed patients, and schizophrenic patients. The c_SMC automatically evaluated 32 elements of the mandalas, found 67 elements that differentiated the non-patient groups and the patient groups, and successfully identified the group only based on the samples. The sample size is large enough to show the result that there exist small variations among Type I errors and Type II errors in the 5-fold cross validation. A sufficient number of statistically significant elements were found through factor analysis. In addition, the regression variables had high significance. This study, therefore, proves the reliability, validity, and usefulness of the c_SMC as a tool for art interpretation. Previous studies of art-based diagnosis, identifying the group based on a

Chapter 13 c_SMC for Differentiation and Identification of Psychological States

given drawing, have been unsuccessful in diagnosing the drawer’s psychological states or disorders. Caution is required when predetermined methods of psychological diagnosis and treatment are applied, especially in regions where considerable cultural differences may exist (Wegmann & Vusenbrink, 2000). Thus, most studies have been limited to finding elements that significantly differentiate the groups. However, the method of using the logistic regression model in this chapter of a mandala-based diagnosis has demonstrated satisfactory Type I and Type II errors in identifying a group. The adjustment of criteria of probability of 0.5 would yield a more desirable balance between the two types of errors. To conclude, the c_SMC has succeeded in acquiring valuable information about the psychiatric conditions of the participant from only one given SMC drawing. The method of evaluating elements, degree of concentration, completeness, and accuracy is not limited to the pattern coloring of structured mandala, but can also be applied to the coloring of any given patterns. Also, this study can be extended to the analysis of psychological disorders such as DID, ADD, and ADHD, as well as all age groups including children, adolescents, and elderly persons. While the samples used in this study were collected from Koreans with similar cultural backgrounds, the c_SMC methodology can also be applied to samples collected from patients of other nationalities or cultures. The impact of the cultural background on the differentiation and the identification of the psychological states can be discussed in future studies. One major challenge in applying computer technology to art therapy assessments is the fact that we are in a preliminary stage of technology development for understanding arbitrary drawings. Currently, there are only few software packages that can recognize characters or scenes in photos. Conversely, the state-of-the-art computer technology can fully analyze every element in the c_SMC. In this chapter, 32 elements of the 495 SMC samples have been evaluated by the c_SMC. Considering the unfeasibility of manual evaluations of this scope, we propose that the SMC is one of the most suitable media of art therapy to be computerized. Aside from the color-related elements in the SMC, the analysis of the elements in unstructured mandala drawings, such as forms, symbols, lines, and movements requires further study. We hope that many experts in related fields will participate in future study and that notable progress will be made in the application and theory of art therapy.

215

Statistical Models for Estimating Level of Psychological Disorder

Chapter 14 Abstract The mandala has been widely accepted as an effective art therapy tool for determining drawer’s emotional or psychological states and for treating associated disorders. In this chapter, we devise a computer system to estimate the level of dementia based on the element evaluations of the c_SMC developed in the previous chapter. Toward the end, we implement a regression model whose dependent variable is the level of dementia scored by a test, and whose independent (explanatory) variables are the elements of the structured mandala. All of these dependent variables are analyzed using the C_CREATES. The consistency between the test score and the computer system’s estimated score is verified statistically. Also, another regression model is formulated to estimate the probability of severe dementia. This model classifies the level of dementia into one of two categories, severe or not, by representing it as an indicator dependent variable. In both regression models developed, the important independent variables are selected by the stepwise regression, and the relative magnitudes of their effects are compared by the standardized regression. The developed system is found to provide art therapists with useful information on the stages of a patient’s advancing dementia. The proposed methods can be applied to estimate the level and the severity of other psychological disorders or states.

14.1 Regression model to estimate degree of dementia using structured mandala Jung (1973) claimed that mandala has a calming and centering effect upon its makers who have psychological states of disorientation or panic. The magic of the mandala can serve individuals by giving needed stability to inner confusion of feelings and by restoring an existing order of the self (Ireland & Brekke, 1980). Art therapists today often employ mandala as a tool for self-awareness, conflict resolution, and various art psychotherapeu216

Chapter 14 Statistical Models for Estimating Level of Psychological Disorder

217

tic practices in various situations (Slegelis, 1987). On the other hand, the Jungian method has also been subject to critique. Slegelis stated that the Jungian school of thought was often criticized for being too mystical and unscientific. Reitman (1951) claimed that the Jungian method was a priori and not empirically justified. Similarly, Slegelis offered a skeptical perspective on art psychotherapy in relation to the Jungian method: The plight of art psychotherapy has been somewhat similar to that of Jungian theory due to the limited amount of scientific research presently existing in such a relatively new field. The difficulty of reducing artwork to measurable data is but one of the dilemmas of such research. Art psychotherapy, too, struggles to be heard as a credible approach to psychological healing. While art therapists see the positive results of their work on an ongoing basis, they are frequently not taken seriously by their colleagues of verbal therapy persuasion. (p. 310) Thus, work remains to be done in systematizing art interpretation so that the process will less depend on special intuitive abilities and become more accessible to art therapists (DiLeo & Kellogg, 1977). A computer system approach to interpreting artwork can be a solution to the skepticism regarding vague, unscientific, or unmeasurable characteristics of drawings. The purpose of this chapter is to develop a computer system that analyzes how the level of emotional, psychological, or physical disorder might be embodied in drawings. In other words, the computer analysis and evaluations of formal elements of drawings can quantitatively evaluate the therapeutic process of treating disorders. We use the Computerized Structured Mandala Coloring (c_SMC) developed in the previous chapter. The Structured Mandala Coloring (SMC) is widely accepted to be effective in alleviating emotional or psychological disorders (Curry & Kasser, 2005). We note again that the simplicity of the SMC is deemed most conducive to the purposes of this research given that the application of computer technologies to art therapy is still at a preliminary stage. In contrast to the complicated process of evaluating elements of unstructured mandala using computer technologies, such as the compositional patterns, thematic imagery (Cox & Cohen, 2000), movement (Kellogg, 1977) as well as forms, symbols, lines, and shapes, the elements in the SMC are mostly color-related for which computer systems have already been developed (Kim, Bae, & Lee, 2007; Kim, Kang, & Kim, 2008). The C_CREATES evaluates not only basic elements of the principal / sub- Independent sidiary colors, the primary / secondary colors, the warm / cool colors, the variables complementary colors, and the completeness, but also applied elements, such as the accuracy and the degree of concentration involved in the activity of coloring a structured mandala, which are deemed important in pattern coloring. We remind the reader of the definitions of three most important

218

PART TWO ART INTERPRETATION

elements in the c_SMC: The completeness refers to the ratio of the area (the number of pixels) colored to the total area, and the accuracy refers to the ratio of the area colored correctly within boundaries to the total area. The degree of concentration is estimated by a function of various formal elements in the C_CREATES, such as the number and the list of used colors, the number of clusters, the length of edges, the completeness, and the accuracy. These three elements are believed to establish the effectiveness of coloring a pattern given in the structured mandala as an art therapy tool. Thus, it is determined that all of the formal elements in the SMC could be evaluated by the C_CREATES. In this chapter, elderly people with suspected dementia are selected as the subjects of the study. The SMC is an appropriate art therapy tool for people with notably diminished functions. Couch (1997) described the appropriateness of mandala for dementia patients: [Mandala] “can indicate thoughts and emotions that patients may be experiencing but unable to express due to the disease process (p. 187);” [Mandala] “can provide order and balance to one whose disordered thoughts, memories, and feelings may be difficult or impossible to access, lost behind the veil of dementia (p. 187),” and “seem to provide a stimulus for art making that dementia patients are able to tolerate, no matter the degree of brain dysfunction” (p. 192). Furthermore, Couch asserted that the symbolism within the circle can serve as a gateway to accessing patients’ remote symbol systems. The level of dementia severity may depend on the degree of concentration involved in the activity of coloring a structured mandala, which, in turn, depends on the completeness and accuracy of the coloring. Various elements involved in c_SMC generally show large variations in the colorings of children or elderly persons, especially the ones over the age of 60 with dementia. The high susceptibility of the elements in c_SMC to the levels of dementia support the appropriateness of the tool (mandala) and the subjects (elderly persons with suspected dementia). Dependent variables We formulate a multiple linear regression model to estimate the level of dementia severity, which is the dependent variable, based on the evaluation of the elements in c_SMC, which are the independent variables. The value of the dependent variable is scored by a test, and the values of the independent variables are evaluated automatically by the system. Several elements that are considered to be important in estimating the level of dementia are selected by stepwise regression and the relative importance of the selected elements is compared by standardized regression. Also, when the level of dementia is classified into one of two categories, severe or not, the probability of severe dementia is estimated by a regression model with an indicator dependent variable.

Chapter 14 Statistical Models for Estimating Level of Psychological Disorder

219

Figure 14.1 Comparison between a normal person and a dementia patient expressed through SMC drawings.

14.2 Methodology 14.2.1 Application to dementia The application of C_CREATES is demonstrated with a few SMC sample. Figure 14.1 shows two SMC samples and its original pattern. The middle mandala is colored by a person with a score of 9 on the MMSE-K (Mini-Mental State Examination-Korean) (Kwon & Park, 1989), corresponding to “definite dementia,” and the right one is by a person with a score of 30, corresponding to “definitely normal person.” The MMSE-K is a modified version of the MMSE developed by Folstein, Folstein, and McHugh (1975). The MMSE-K is composed of twelve questions with a total score of 30: Time orientation (3), place orientation (5), registration (3), recall (3), attention and calculation (5), language (7), understanding (3), and judgment (2). The numbers in parentheses are the scores. A person who scores 24-30 is classified as “definitely normal,” 0-10 as “definitely having dementia,” and 10-23 as “having potential dementia.” The objective and quantitative information on the color-related elements, based on the results of the analysis and rating of the two mandalas in Figure 14.1 is summarized in Table 14.1. The area of each color in number of pixels and the principal / subsidiary colors are deduced with the corresponding area, the primary / secondary colors with their percentages, the warm / cool colors with their percentages, and the complementary colors. Also, we obtain the values for number of clusters, completeness, accuracy, and degree of concentration. In Figure 14.1, less completeness and accuracy are noted in the middle mandala colored by a person with a lower MMSE-K score (more severe dementia) than in the right mandala colored by someone with a higher MMSE-K score. The lower percentage of primary colors or the higher

220

PART TWO ART INTERPRETATION

Table 14.1 The results of analysis and rating of Figures 14.1A and 14.1B _____________________________________________________________________________________________ Accuracy Number Number Complete- Concent- Main Subsidiary Primary Secondary Warm Cool Figure Score (%) of of ness ration colors colors colors colors clusters (rank) (%) _____________________________________________________________________________________________ 14.1A 14.1B

9 30

72.8 91.4

3 9

7 20

0.88 0.98

58.7 orange light-green 5.3 red blue-green

4.6 52.4

56.6 61.2 0.0 14.9 46.9 11.5

_____________________________________________________________________________________________

percentage of warm colors in the middle mandala is deemed to result from the difference in personal preferences. The relationships between the level (score) of dementia and the differences of various elements in the mandalas are investigated. 14.2.2 Model I: Estimation of levels of dementia A multiple regression analysis is applied to estimating the level of dementia scored by the MMSE-K, Y, based on the elements, Xj (j = 1, 2, ... , m), in the c_SMC. The sample size is denoted by n and the values of Y and Xj for sample i as Yi and Xij, respectively. The regression model is formulated as: Yi = a1 Xi1 + a2 Xi2 + ... + ak Xik + εi, i = 1, 2, ... , n. See Chapter 6 for more detailed explanation on the regression model, stepwise regression, and standardized regression. The following independent variables are considered as the potential variables that explain the level of dementia score: number of used colors (X1), principal color (X2), percentage of primary colors (X3), percentage of warm colors (X4), existence of complementary colors (X5), number of clusters (X6), accuracy (X7), completeness (X8), and degree of concentration (X9). The principal color is one of the 15 colors defined by the KIS. To represent each of the 15 colors, the variable X2 is replaced with 14 indicator variables: X21 = 1, in the case of red, X22 = 1, orange, X23 = 1, yellow, X24 = 1, light green, X25 = 1, green, X26 = 1, blue-green, X27 = 1, blue, X28 = 1, blue-purple, X29 = 1, purple, X2,10 = 1, red-purple, X2,11 = 1, pink, X2,12 = 1, brown, X2,13 = 1, white, X2,14 = 1, grey. Otherwise, these variables have values of zero. In the case of the color black, X21 = ... = X2,14 = 0. The existence of complementary colors is represented as X5 = 1 when complementary colors exist, and X5 = 0, otherwise. This is designated as Model I. 14.2.3 Model II: Probability of severe dementia In addition, another regression model is formulated with an indicator dependent variable that represents whether the level of dementia is severe or not. Here, the severity is classified into two groups. The dependent variable is set to Y = 1 when the dementia score is below a certain value of C, and Y = 0, otherwise, where C is the threshold value that determine the severity

Chapter 14 Statistical Models for Estimating Level of Psychological Disorder

221

group of dementia. In this model, Model II, the estimated value of Y can be interpreted as a probability that the level of dementia is severe (Kutner et al., 1996). By applying a stepwise regression, the important independent variables in estimating the dependent variable in Model I and Model II are selected among the 21 variables. Each of the aforementioned procedures can be performed by a statistical software package such as SPSS (Statistical Package for the Social Science) (IBM Corporation, 2015) or SAS (Statistical Analysis System) (SAS Institute, 2014).

14.3 Results and system validity 14.3.1 Selection of independent variables and their effects in Model I A sample of 100 SMC drawings was collected from elderly persons with suspected dementia in a sanitarium. The persons were tested by the MMSE-K and their scores of dementia were obtained. Table 14.2 summarizes the results of the analysis in Model I with the level of dementia scored as a dependent variable and the results of the analysis in Model II with the severity of dementia as the indicator dependent variable. In Model I, five independent variables, light-green (X24), brown (X2,12), green (X25), the number of clusters (X6), and accuracy (X7) (the notations have been changed) were selected as important explanatory variables. The variable, degree of concentration, was omitted in this model. This does not mean that the variable is not important, but implies that it is highly correlated with the variables selected. The regression function derived was as follows: Level of dementia = 4.411 + 14.598 X24: Light green - 16.507 X2,12: Brown - 5.145 X25: Green + 0.718 X6: Number of clusters + 0.095 X7: Accuracy. COD The Coefficient Of Determination (COD) was 2 R = 0.626, which is only a moderately satisfactory value, but indicates that the model is applicable and relevant. The standard deviation of error was estimated to be 6.268. The residuals, which are not presented here, showed the normality, the independency, and the constant variance of the error term εi. In conclusion, the method is appropriate and can be used effectively. The positive or negative signs of the coefficients of the independent vari- Interpretation of ables coincide with our expectations. As light-green (X24), number of clus- coefficients ters (X6), or accuracy (X7) with a plus sign appears or increases, the estimated level of dementia scored increases (the dementia becomes weaker). When, as the principal color, brown (X2,12) or green (X25) with a minus sign appears, the estimated level of dementia scored decreases (the dementia be-

Standardized Sample Value Significance coefficient average of t level (beta) (Standard deviation)

___________________________________________________________________________________________________________________________

Coefficient of determination: R2 = 0.376, Standard error: 0.374 Dependent variable: Sample average = 0.32, Sample standard deviation = 0.47

Model II Constant - 0.299 0.090 - 3.322 0.001 Brown 0.623 0.222 0.228 2.802 0.006 0.03 ( 0.17) Concentration - 0.009 0.001 - 0.607 7.462 0.000 63.49 (30.09) ___________________________________________________________________________________________________________________________

Coefficient of determination: R2 = 0.626, Standard error: 6.268 Dependent variable: Sample average = 15.14, Sample standard deviation = 9.99 ___________________________________________________________________________________________________________________________

Model I Constant 4.411 1.198 3.683 0.000 Light green 14.598 2.885 0.320 5.060 0.000 0.05 ( 0.22) Brown - 16.507 3.729 - 0.283 - 4.427 0.000 0.03 ( 0.17) Green - 5.145 2.466 - 0.132 - 2.086 0.003 0.07 ( 0.26) Number of clusters 0.718 0.159 0.437 4.529 0.000 9.47 ( 6.08) Accuracy 0.095 0.031 0.300 3.805 0.040 42.61 (31.50) ___________________________________________________________________________________________________________________________

___________________________________________________________________________________________________________________________

Non-standardized _________ ____________________ Variables coefficient Standard error

Table 14.2 Results of regression analysis ___________________________________________________________________________________________________________________________

222 PART TWO ART INTERPRETATION

Chapter 14 Statistical Models for Estimating Level of Psychological Disorder

comes more severe). The magnitude of the effects on the level of dementia scored was estimated. For example, it was found that as the accuracy increases by 1%, the level of dementia scored increases by 0.095 on average. Although going outside the lines can be interpreted as an act of creativity and/or rebelliousness, the accuracy of coloring inside the lines (a positive sign) signifies the general tendency of less severe dementia in this study. As such, the regression function discussed herein provides information on general tendency, which may benefit art therapists. Brown, green, and light-green were selected as important colors affecting the level of dementia. Based on the samples of structured mandala colored by elderly persons, it was found to be statistically significant that light green is associated with positive aspects (mild dementia), and brown and green are associated with negative aspects. Using brown as a principal color in a mandala implies a severe level of dementia. Kellogg (1977) described the interpretation of brown in a mandala as follows: When we consider brown, we think immediately, of course, of feces. Brown can reflect contact with fecal material during birth, and can point to a preoccupation with it. It can at times betray masochistic tendencies. On the positive side, the use of brown can indicate tremendous potential - think of the brown earth ready for planting. If brown is placed in a small area at the center of the mandala, it often means that the person has very low self-esteem, feels worthless and dirty. (p. 124) The negative aspect of green and the positive aspect of yellow-green indicate that green and yellow-green imply a severe and a mild level of dementia, respectively. Kellogg (1977) described green and yellow-green as follows: Green connotes a critical area because it is most polarized in both its positive and negative connotations. Green is the color of vegetation. It can therefore represent healthy growth. It can show a capacity for nurturing, both the internalized nurturing of oneself and the nurturing of others. The entire spectrum of green needs to be carefully studied. It is interesting that therapists and nurses frequently use green in mandalas in combination with yellow. I think that, in this combination, yellow represents assertiveness directed toward people and the green represents a nurturing quality. Green is care of others but, first of all, care of one’s self. Green, used too extensively, can show rigidity or a tendency to overly care for, be overly possessive of, or overprotect other people. Excessive use of green deserves attention, especially when found in mandalas of health-care professionals. (p. 124) Fincher (1991) stated in relation to green: Green is a pleasant color to most people... Green is sometimes en-

223

224

PART TWO ART INTERPRETATION

Figure 14.2 Correlation between computer evaluation and MMSE-K test scores.

countered as a symbol of negativity. One may recall that the poisonous venom of snakes is green. The dark green woods of fairy tales is a place of danger. Vegetation, as it rots, becomes a darker and darker green. The human body, too, allowed to pass through natural process of death and decay, assumes a greenish tint. Folk wisdom tells us of the unpleasant state of being “green with envy.”... Dark and light shades of green have special significance when considering the mandala. (p. 58) Kellogg (1978) found that dark green points to threatening aspect of the nurturer and speaks to the memories of “dark forest where the witch lives.” In analytic terms, this refers to the frightening aspects of the public area out of which we were all thrust. Medium green and some lighter shades speak in a positive way of the harmonious blending of active and receptive forces in the psyche, of growth and fertility. There are instances also where the use of a bright chartreuse green, heavy with a yellow hue will reflect a harsh superego an authoritarian kind and may reflect a conflict in the previous developmental stage. (p 76-77) These colors are not subject to cultural differences according to Kellogg and Fincher’s interpretations. It would be too difficult to relate the results of this study to those of the studies by Kellog (1977) and Fincher (1991). Nevertheless, it is important to note that every color can have either a positive or a negative meaning, depending on the context in which it is used (Kellogg, 1977). The analysis of the results of this study demonstrates only a general tendency for color interpretations. In Figure 14.2, the relationship between the level of dementia scored by the MMSE-K and that estimated by the computer system is plotted. The

Chapter 14 Statistical Models for Estimating Level of Psychological Disorder

225

100 SMC samples are denoted by 00, 01, 02, ... , 99. The system’s ability to estimate the level of dementia based on the analysis of formal elements in c_SMC is demonstrated. Findings indicate a large discrepancy between the test scores and the estimated computer system scores on some points. Thus providing further evidence that people are too complex to be interpreted by a single element in a drawing as substantial evidence of their characters (Cohen & Mills, 1994). Nonetheless, the result demonstrates to a general tendency and provides useful information about people with dementia. The regression function in the standardized regression model with standardized variables was derived as Standardized level of dementia = 0.320 X24’- 0.283 X2,12’ - 0.132 X25’ + 0.437 X6’ + 0.300 X7’. For example, as accuracy (X5) (the ratio of the area colored correctly within boundaries to the total area) increases by one standard deviation (31.5 %), the level of dementia severity estimated increases (the dementia becomes less severe) by 0.300 of its standard deviation (0.300 x 9.99 = 3.0) on average. Thus, the more the colors within the boundaries of the circle, the higher the MMSE-K score, and the less severe the level of dementia. Based on these estimated standardized coefficients assuming no inter-correlation among these variables, the important independent variables are listed in descending order: the number of clusters (X6), light-green (X24), the accuracy (X7), brown (X2,12), and green (X25). Light green was found to have the strongest impact on level of dementia, and green, the weakest. 14.3.2 Selection of independent variables and their effects in Model II In Model II, Y = 1 denotes a score below 10 (= C) indicating the level of dementia on the MMSE-K, the case of severe dementia, and Y = 0, otherwise. As important independent variables, brown (X2,12) and the degree of concentration (X9) were selected. The regression function was derived as Probability of severe dementia = - 0.299 + 0.623 X2,12: Brown - 0.009 X9: Concentration. For example, when the values of the other independent variables are held constants, as the degree of concentration is increased by one in rank, the probability of severe dementia decreases by 0.009 on average. When brown appears as a main color, the probability becomes 0.623 higher. This analysis is illustrated in Figure 14.3. The COD was:

R = 0.376. However, this low value is irrelevant when the dependent variable is an indicator type. The standard deviation of error was estimated to be 0.374. The independent variable brown in Model I still remains in Model II, which may reinforce its negative aspect. The removal of light-green may weaken its 2

COD

226

PART TWO ART INTERPRETATION

Figure 14.3 Effects of degree of concentration and brown on the probability of severe dementia.

positive aspect, and the removal of green may weaken its negative aspect. Removing the number of clusters and accuracy and including the degree of concentration instead may be due to a high correlation between them. The positive or negative signs of the coefficients coincide with the researcher’s expectations. The regression function in the standardized Model II was: Standardized probability of severe dementia = 0.228 X2,12’ - 0.607 X9’. For example, as the degree of concentration increases by one standard deviation (30.1 %), the probability of severe dementia decreases by 0.285, which is 0.607 of its standard deviation (0.47) (0.607 x 0.47 = 0.285). In other words, the higher the degree of concentration, the lower the probability of severe dementia. From these standardized coefficients assuming no inter-correlation among these elements, the most important independent variables in the descending order are the degree of concentration and brown. In other words, degree of concentration as a variable has a stronger impact on the level of dementia than brown does. Finally, comparison of the MMSE-K score, the computer system rating and the results of two art therapists’ diagnosis of dementia based on their experience was made. Two art therapists working at a large elderly care facility classified the structured mandalas into “colored by persons definitely dementia” (D), “underdetermined” (U), and “definitely normal” (N). They made decisions only based on mandala colorings without any contact with the subjects. QWK The Quadratic Weighted Kappa (QWK) values among the MMSE-K, the computer system and the two art therapists are given in Table 14.3. The QWK value of κ2 = 0.771,

Chapter 14 Statistical Models for Estimating Level of Psychological Disorder

________________________________________________________________________

227

Table 14.3 Art therapist-1 Art therapist-2 MMSE-K test QWK values among _____________________________________________________________ computer system, MMSE-K, and two art D U N D U N D U N therapists Computer D 23 6 1 26 4 0 23 7 0 system U 2 41 8 13 24 14 8 32 11 N 0 8 11 2 6 11 1 4 14 _____________________________________________________________

Kappa (κ2): 0.706

Kappa: 0.600

Kappa: 0.707

________________________________________________________________________

Art therapist-1 Art therapist-2 _________________________________________

MMSE-K D 19 13 0 24 7 1 test U 6 28 9 14 18 11 N 0 14 11 3 9 13 _________________________________________ Kappa: 0.592

Kappa: 0.534

____________________________________________________

Art therapist-2 ____________________

Art D 25 0 0 Therapist-1 U 16 32 7 N 0 2 18 ____________________ Kappa: 0.771

_______________________________ between two art therapists indicates substantial agreement between them. However, the QWK values of κ2 = 0.592 and 0.534, between MMSE-K and the two art therapists indicate only moderate agreement. The QWK value between the MMSE-K and the computer system was κ2 = 0.707, which is higher than the kappa value between two art therapists. The QWK values of agreements between the computer system and two art therapists were κ2 = 0.706 and 0.600, which implies a substantial agreement between the computer system and the art therapists. This thus proves that the computer rating system can provide reliable information.

14.4 Case studies In Figure 14.4, 6 SMC samples corresponding to levels of dementia (scores)

228

PART TWO ART INTERPRETATION

Figure 14.4 SMC samples colored by people with different levels of dementia.

of 0, 9, 12, 20, 22, and 30, respectively, are shown. The regression functions derived could be used by art therapists without difficulty, since they are only linear equations with statistical concepts of average and standard deviation. Moreover, all the readers in practice do not need to try to understand the method and Table 14.2. In Table 14.4, the scored and estimated values with other ratings of the independent variables are summarized. The general tendency for the dementia to be less severe (as indicated by an increase in the MMSE-K score and estimated score) correlates with higher accuracy. Also, the number of used colors and the number of clusters become larger except for one drawing (Figure 14.4D). The same results were obtained for completeness. The order of the values scored by the test coincided completely with that of the values estimated by Model I. Except for one drawing (Figure 14.4D), the same results were obtained by Model II regarding the estimation of the probability of severe dementia.

14.5 Discussion, conclusion, and further study

Chapter 14 Statistical Models for Estimating Level of Psychological Disorder

229

Table 14.4 Analysis of mandalas in Figure 14.4

_____________________________________________________________________________________________

Dementia Severe dementia Accuracy Number Number Complete- Concent- Main _____ _________ ______ __________ Figure of of ration (%) ness color colors clusters (rank) score estimate yes/no prob. _____________________________________________________________________________________________ 14.4A 14.4B 14.4C 14.4D 14.4E 14.4F

11.4 47.7 73.3 77.8 86.1 93.1



1 2 6 3 7 8

5 9 8 12 13 19

0.10 0.68 1.0 1.0 1.0 0.96

96.7 yellow 72.0 orange 42.0 pink 50.9 grey 25.8 blue-green 15.0 yellow

0 9 12 20 22 30

9.1 15.4 17.1 20.4 21.9 25.9

1 1 0 0 0 0

0.63 0.38 0.00 0.17 0.12 0.17

_____________________________________________________________________________________________ In this chapter, the level of dementia was estimated based on the elements in the c_SMC using regression models. Cohen et al. (1988) used a similar approach of the regression method in analyzing the DDS. However, they incorrectly assumed independence among explanatory variables. For example, the DDS variables, such as color type, blending, and space usage are inevitably correlated among themselves. As the number of colors increases, blending tends to occur more frequently, and space usage tends to occupy a larger area. Moreover, the same elements in the free, tree, and feeling DDS pictures must be generally highly correlated. Thus, all the results of Cohen et al.’s analysis might be misleading. Elements, such as brown, light-green, green, the number of clusters, and accuracy, are identified as important variables for determining levels of dementia. As the number of clusters and the accuracy increase, the level of dementia becomes less severe. When brown or green appear to be a principal color, the level of dementia becomes more severe. When light-green is a principal color, it becomes lower. Assuming no correlation among elements, the order of magnitude of the relative effects on the level of dementia are the number of clusters, light-green, accuracy, brown, and green, in descending order. In the case of identifying severe dementia where the probability of severe dementia is estimated, by the stepwise regression, green and lightgreen were removed from the explanatory color elements while brown was retained. Also, the number of clusters and accuracy were removed, and the degree of concentration was added instead. Finally, brown and the degree of concentration were selected as the “best” explanatory variables. Brown selected in both models may make its negative aspect more certain. Adding the degree of concentration, and removing the number of clusters and the accuracy do not mean that the latter variables are not important, but they imply that the degree of concentration is highly correlated with the two variables. The relatively important elements on the probability of severe de-

230

PART TWO ART INTERPRETATION

mentia were the degree of concentration and brown in the descending order. These interpretations generally coincide with the researchers’ expectation. The contribution of this chapter is that it expresses these interpretations quantitatively. Furthermore, art therapists both in practice and theory can apply this approach and generate their own formulas for certain psychological disorders based on specific art therapy assessments. In Chapter 11, we developed an expert system for the interpretation of SMC (Kim, Kim, & Kim, 2008) based on the results obtained from analyzing and evaluating various elements by the C_CREATES. In this chapter, regression model was developed for the estimation of the level of dementia based on the elements in c_SMC. The c_SMC is simple and thus appropriate as an art therapy tool for elderly persons with dementia. The same may be true for young children, and possibly for high school students or young adults. The c_SMC is a validated tool for estimating the level of dementia with a computer evaluation system. When the level of a psychological disorder can be determined or measured by some tests or psychologists in an art therapy assessment, it can also be estimated by a computer system using the same approach based on the variables of drawings. For instance, the possible psychological disorders that can be applied are: the ADD (Attention-Deficit Disorder), ADHD, or DID (Personality/Dissociative Identity Disorder), and possible art therapy assessments that can be used are: the DAP, KFD, DDS, PPAT, or FSA. Furthermore, art therapists can apply the approach and generate their own formula for certain psychological disorders based on certain art therapy assessments. These studies are expected to contribute to the effective application of structured mandala and its extension to general art therapy practice. Gantt (1998) foresaw the future of art therapy research, in which science is given proper roles in art therapy and skillfully applied to its practice. Kaplan (1998) recognized the need for research in the findings of other disciplines, along with art therapy. Recently, Kapitan (2007) described the attitude that art therapists should have and the direction they should take in the rapidly expanding digital information age: The artistic and technological tools of the age are creating a growing demand for art therapists who can interact not only with new media but also with the values and pathologies of an emerging techno-culture. Art therapists must be willing to move beyond historically validated media and offer our work in new contexts. Art therapists must broaden the definitions of art materials and contexts across a wide spectrum. (p. 50) We hope that many experts in related fields such as art therapy, psychology, psychiatry, art, education, computer science, statistics, etc., will participate in this kind of interdisciplinary study. Computers can provide human experts with objective information that can assist their decision-making.

Chapter 14 Statistical Models for Estimating Level of Psychological Disorder

Also, non-experts such as guardians, teachers, or parents can use the system in ascertaining signs of their clients’ psychological problems at an early stage and, therefore, promptly seek the help of professional art therapists.

231

A Statistical Approach to Comparing the Effectiveness of Several Art Therapy Tools in Estimating the Level of a Psychological State Chapter 15 Abstract In this chapter, we propose a generalized statistical approach to comparing the effectiveness of several art therapy tools in estimating the level of a psychological state. A regression model is developed for each of the selected tools, whose dependent variable (to be predicted) is the level of a psychological state. Various elements of drawings proposed by corresponding tools are considered as independent (predicting) variables, some of which are evaluated by the C_CREATES while others are manually evaluated by the method of corresponding art therapy tools. A set of independent variables which “well” predict the dependent variable is determined by stepwise regression. The effectiveness of the tools is analyzed and measured in a comparable quantitative term so that they can be compared with one another. While this approach is applied to three widely used art therapy tools in estimating the level of dementia in this study, it can be applied to any art therapy tool and to any psychological state.

15.1 A generalized approach to compare effectiveness of several art therapy tools Louw and Ramkisson (2002) identified sexually abused girls using the HTP and the DAP, the results of which indicate that the sexually abused girls differ significantly on the four scales of the HTP and the DAP from the same age group who do not have any history of sexual abuse. Then one might want to know which test, the HTP or the DAP, is more effective in distinguishing (identifying) the sexually-abused from the non-abused. Veltman and Browne (2001) found that the KFD was three times more accurate than the Favorite Kind of Day (FKD) (Manning, 1987) in identifying maltreated children. However, their method is based on small sample and subjective decision. We predicted the existence of severe dementia and estimated the level of dementia in Chapter 14 using the c_SMC. We found that the level 232

Chapter 15 Effectiveness Comparison of Several Art Therapy Tools

233

of dementia correlates with the accuracy in coloring, the number of clusters, and the use of light-green, brown, and green, and that severe dementia is related to the degree of concentration and the use of brown. In this chapter, we develop an approach using a statistical method to com- Dependent variable pare the effectiveness of several art therapy tools in estimating the level of a psychological state, including symptoms or disorders, such as child abuse or neglect, sexual abuse, trauma, schizophrenia, dementia, depression, attention deficit, etc. As a statistical method, a multiple linear regression analysis is adopted, which is a widely used method predicting one variable from the others (Kutner et al., 2005). The dependent variable of this model to be predicted is the level of a psychological state. The level may be determined in scores by formal tests such as the Mini-Mental State Evaluation (MMSE) (Folstein et al, 1975), the Beck’s Depression Inventory (Beck, Ward, Mendelson, Mock, & Erbaugh, 1961), the Symptom Checklist-90-Revisional (SCL-90-R) (Derogatis, 1977), the Myers-Briggs Type Indicator (MBTI) (Myers & McCaulley, 1985), and the Murphy- Meisgeier Type Indicator for Children (MMTIC) (Murphy & Meisgeier, 1987), or by evaluation of experts, such as art therapists and psychologists. The independent variables, which predict the dependent variable, can be Independent the elements of drawings including the number of used colors, space us- variables age, details, and accuracy in coloring, that can be found in the Descriptive Assessment for Psychiatric Art (DAPA) (Hacking, Foreman, & Belcher, 1996), the DDS (Cohen et al., 1988), the FEATS, the revised Face Stimulus Assessment (Hamilton, 2008), and the C_CREATES. Some elements are automatically evaluated by the C_CREATES, which is a computer system that determines unusual placement (Kim, Kang, & Kim, 2008) as covered in Chapter 5, judges main color (Kim, 2008a) in Chapter 4, measures completeness and accuracy (Kim, Kang, & Kim, 2009) in Chapter 2, and rates variety of colors (Kim & Hameed, 2009) in Chapter 3. Others are manually evaluated by art therapists. Among various independent variables, the ones that “well” predict the dependent variable are selected by stepwise regression. Then the effectiveness of tools is analyzed via scatter plots and is measured by the Coefficient Of Determination (COD). Also, the relative importance of selected independent variables is compared by the standardized regression. We develop a generalized approach to compare several art therapy tools in estimating the degree of a psychological state. As an application and case study of this approach, we compare the effectiveness of three art therapy tools in estimating the level of dementia, the PPAT, FSA, and SMC. Samples are collected from 58 patients with suspected dementia in a psychiatric unit, their ages range from 60 to 90. The level of dementia, which is a dependent variable, is measured by the Korean version of the MMSE

234

PART TWO ART INTERPRETATION

(MMSE-K). A total of 64 elements are considered as the independent variables of this model. These elements came from the DAPA, DDS, FEATS, FSA, and the C_CREATES. Some independent variables, such as the number of used colors, number of clusters, length of edge, and area colored in number of colored pixels, are automatically evaluated by the C_CREATES (Kim, Bae, & Lee, 2007; Kim, 2010), while others, such as the color fit, implied energy, integration, and logic, are manually evaluated by human art therapists.

15.2 Approach: Regression model We compare the effectiveness of several art therapy tools in estimating the level of a psychological state or disorder based on the evaluation results of various elements of drawings. Psychological disorder includes child abuse or neglect, sexual abuse, trauma, schizophrenia, dementia, depression, attention deficit disorder, etc. Art therapy tools include free drawings, the SMC, DDS, PPAT, HTP, KFD, DAP, FKD, FSA, etc. Multiple regression analysis is applied to each art therapy tool in order to estimate the level of the psychological disorder (Y, dependent variable) based on the m elements of drawings (Xj, j = 1, 2, … , m, independent variables). In this study, the dependent variable is the level of psychological state. As independent variables, we consider 15 elements in the DAPA, 22 elements in the DDS, 14 elements in the FEATS, 9 elements in the revised FSA (Hamilton, 2008), and 15 elements in the C_CREATES. Some of the variables overlap, such as the integration in the FEATS and DDS, the line quality in the FEATS and DDS, and the space in the FEATS and DAPA, while some of them are the same elements using different names such as the line length in DDS and the length of edge in the C_CREATES, the representational in the DDS and the realism in the FEATS, and space usage in DDS and space in DAPA. With this taken into account, a total of 64 elements are evaluated, with 15 elements automatically evaluated by several computer systems for art evaluations and the remaining 49 elements manually evaluated by human raters. Of course, we do not consider elements that are inappropriate for certain art therapy tools, such as accuracy in the PPAT, line quality in the SMC, and rotation in the FSA. Elements that are considered important in estimating the level of psychological state are selected by stepwise regression, and the relative importance of the selected elements is compared by standardized regression. When we assume no correlation among the elements, the larger coefficients of independent variables in standardized regression can be interpreted as having the greater relative effect on the dependent variable. The whole procedure is performed using a statistical package, SPSS.

Chapter 15 Effectiveness Comparison of Several Art Therapy Tools

235

15.3 Case study 15.3.1 Subjects of psychological disorder and art therapy tools In this chapter, dementia is chosen as a target psychological disorder due to Dementia its prevalence worldwide. Dementia is one of the leading public health issues of the 21st century (Stewart, 2006). The number of Americans affected by Alzheimer disease was expected to reach 14 million by 2050 (Cowley, 2000). In South Korea, the size of the 65 and older population in Korea was 5 million, 10.3% of the whole population in 2008 (Han, 2008); this number was expected to increase to 10 million in 20 years by 2028, 23.0% of the population. Among this age group, the rate of definite dementia was 8.4% in 2008 and it was estimated to reach 9.6% in 2028 (Seoul National University Hospital, 2008). Art may provide dementia patients with opportunities to express themselves and to take pleasure in using colors, forms, and textures (Stewart, 2006). Also, their drawing outcomes can be compared to determine the course of patient treatment (Betts, 2006). When examined over an extended period of time, their drawings can assist facility staff and families in evaluating the progression of dementia (Stewart, 2004). Couch (1997) stated that drawings can indicate thoughts and emotions that patients may be experiencing but are unable to express due to the disease process. It can provide order and balance to patients whose disordered thoughts, memories, and feelings may be difficult or impossible to access, lost behind the veil of dementia. Therefore, art therapy can invite older adults affected by dementia to freely express their states of mind and emotions through the nonverbal means of drawing (or the art-making process). Practicing art therapy with dementia patients can, however, be a challenging task for art therapists (Couch, 1997). The approach has been applied to three widely used art therapy tools, the PPAT, FSA, and PPAT, the FSA, and the SMC, in estimating the level of dementia. In this SMC study, we formulated a multiple linear regression model to estimate the level of dementia measured by the MMSE-K, based on the evaluations of elements of drawings. The PPAT, developed by Gantt (1990), is a single picture evaluation in which the participant is directed to draw a person picking an apple from a tree. Gantt and Tabone (1998) proposed an approach estimating the level of dementia using the PPAT. The FSA, designed by Betts (2003), is a series of three stimulus pictures. In this research, we only used the first picture, which is a standardized image of a human face for clients to complete. The FSA is known to be targeted for people with multiple disabilities, especially for those with communication disorders. People with dementia tend to exhibit cognitive impediment, which provides an insight that the FSA may be a great fit as an art therapy assessment for this population. The SMC is known to promote concentration and provide relaxation to clients as they color a given geometric pattern (Curry & Kasser, 2005).

236

PART TWO ART INTERPRETATION

Figure 15.1 PPAT, FSA, and SMC samples completed by a participant with an MMSE-K score of 21.

Couch (1997) discovered that dementia patients obtain a state of balance as they choose and arrange a balanced combination of colors in the SMC. In Chapter 14, we introduced a computer system, the c_SMC, and thus automatically estimates the level of dementia (Kim, Betts, et al., 2009), which can be substituted for art therapists in evaluating and interpreting the SMC. In Chapter 11, we introduced an expert system for interpreting the SMC (Kim, Kim, & Kim, 2008). For these reasons, we selected the PPAT, the FSA, and the SMC as suitable art therapy tools for assessing the level of dementia. We collected samples from 58 elderly patients with suspected dementia in a psychiatric unit. The types of dementia include Alzheimer disease, vascular dementia, senile dementia, and Parkinson disease; 28.6% of the participants are males and the average age is 81.6 years with a standard deviation of 7.77 years. The subjects are tested by the MMSE-K and the scores of dementia are obtained. The MMSE-K is composed of 7 categories and its total score is 30. A person who scores from 24 to 30 is classified as “definitely normal”, one who scores from 0 to 10, as “definitely having dementia”, and one who scores from 10 to 23, as “having potential dementia”. The participants are also directed to draw the PPAT, the FSA, and the SMC under the following instructions: PPAT - markers, B4 drawing paper; FSA - markers, A4 drawing paper printed pattern of eyes, nose, lips, ears, neck, and face; and SMC - crayons, A4 drawing paper printed pattern of mandala. Figure 15.1 shows the samples completed by a participant with an MMSE-K score of 21, who is classified as “potential dementia” but close to “normal.”

Chapter 15 Effectiveness Comparison of Several Art Therapy Tools

________________________________________________________________________ Elements

Rater1 : Rater2 _____________

Rater1 : Rater3 _____________

Table 15.1 Inter-rater reliabilities Rater2 : Rater3 2 _____________ (QWK values, κ ) about the PPAT and FSA PPAT FSA

PPAT FSA PPAT FSA ________________________________________________________________________

Color fit Implied energy Integration Logic Realism Problem solving Developmental level Line quality Person Rotation Perseveration

0.908 0.860 0.917 0.878 0.921 0.951 0.973 0.781 0.918 0.752 0.846

0.960 0.887 - 0.905 0.830 - 0.943 0.948 - - 0.952

0.947 0.879 0.931 0.886 0.874 0.952 0.969 0.944 0.973 0.785 0.905

0.934 0.931 - 0.894 0.971 - 0.954 0.896 - - 0.963

237

0.899 0.854 0.949 0.946 0.890 0.928 0.961 0.778 0.909 0.885 0.933

0.908 0.911 0.877 0.828 0.946 0.873 0.962

________________________________________________________________________ -: Elements not used in the corresponding art therapy tool.

15.3.2 Independent variables Except for 26 elements, such as animals, word inclusions, and landscapes from the DDS and the DAPA, which are not suitable for the PPAT, the FSA, and the SMC, we evaluated all the elements of the FEATS and the C_CREATES. Overall, 38 elements were considered to estimate the MMSE-K score: number of colors (X1), main color (X2), number of clusters (X3), area colored (X4), length of edge (X5), ratio of primary colors (X6), ratio of secondary colors (X7), ratio of warm colors (X8), ratio of cool colors (X9), complementary colors (X10), area of colored convex hull (X11), number of colored grids (X12), completeness (X13), accuracy (X14), concentration (X15), prominence of color (X16), color fit (X17), implied energy (X18), space (X19), integration (X20), logic (X21), realism (X22), problem-solving (X23), developmental level (X24), details of objects and environment (X25), line quality (X26), person (X27), rotation (X28), perseveration (X29), prominence of color in the FSA (X30), color fit in FSA (X31), implied energy in FSA (X32), logic in FSA (X33), realism in FSA (X34), developmental level in FSA (X35), details of object and environment in FSA (X36), line quality in FSA (X37), and perseveration in FSA (X38). Twenty-six elements, X1 - X12, X16 - X29, were considered for the PPAT; Nineteen, X1 - X10, X30 - X38, for the FSA, and thirteen X1 - X10, X13 - X15, for the SMC. Three human art therapists (Rater-1, Rater-2, and Rater-3) evaluate X16 - X38, 23 elements, by grading on a scale of 1 to 5. The remaining elements were automatically evaluated by the C_CREATES. Table 15.1 shows the inter-rater reliabilities of the grades given by the raters. All inter-rater reliabilities were satisfactory, as demonstrated by the Quadratic Weighted Kappa (QWK) values between the pairs of raters. All the SMC elements were automatically evaluated by the C_CREATES.

238

PART TWO ART INTERPRETATION

15.3.3 Results Regression models For the PPAT, two elements, the realism (X22) and the line quality (X26), and CODs were chosen as important variables to predict the dependent variable. Croof three tools nin and Werblowsky (1979) found that people with organic brain syndrome tended to draw short or repeating lines during the art-making process. For the PPAT, realism is a meaningful variable, since it is based on freestyle drawings, unlike the FSA and the SMC, which have given patterns. The regression function was obtained as: MMSE-K (PPAT) = 3.002 + 2.367 (0.440) x X22: Realism [0.001] + 1.835 (0.555) x X26: Line quality [0.002]. The median of three ratings for each independent variable was used. The numbers in parentheses are standardized coefficients. The numbers in brackets are significance probabilities. The COD was R2 = 0.591. As an example, the client who drew the three drawings in Figure 15.1 with the MMSE-K score of 21 obtained grade 4 for reality and 4 for line quality in the PPAT. Thus, the MMSE-K score is estimated to be: MMSE-K (PPAT) = 3.002 + 2.367 x 4 + 1.835 x 4 = 19.8. For the FSA, line quality (X26) and number of colors (X1) were chosen. The regression function was: MMSE-K (FSA) = 0.742 + 2.931 (0.517) x X26: Line quality [0.000] + 0.991 (0.294) x X1: Number of colors [0.008]. The COD was R2 = 0.390. For the SMC, accuracy (X14) was chosen. For the SMC, the accuracy in coloring within the patterns is an important variable. The regression function was MMSE-K (SMC) = 7.341 + 0.157 (0.703) x X14: Accuracy [0.000]. The COD was R2 = 0.494. Each of the individual variables, i.e. realism, line quality, number of colors, and accuracy, could not predict alone the MMSE-K. Rather, all 38 elements were considered to be inter-correlated in this approach. Table 15.2 shows means and standard deviations of the selected variables. For the FSA, the same client earned 4 for line quality and for number of used colors. Thus, the estimated MMSE-K score was calculated to be 16.4.

Chapter 15 Effectiveness Comparison of Several Art Therapy Tools

239

Figure 15.2 Sample by participants with different MMSE-K scores.

240

PART TWO ART INTERPRETATION

Figure 15.2 (continued).

For the SMC, accuracy was 68.9%, and thus the estimated MMSE-K score was 18.2. Overall, the PPAT estimated the MMSE-K score most closely to the actual score, followed consecutively by the SMC and the FSA. Figure 15.2 shows another set of seven samples. These samples are placed in the order of the MMSE-K scores from high to low with the first two being definitely normal (28 and 24) the next four with potential dementia (20, 19, 16, and 11), and the last with definitely dementia (2). We can clearly see the differences in the level of dementia depending on the MMSE-K scores. Figure 15.2A is completed by an outpatient with no special symptoms of dementia. Figure 15.2F is done by a patient who could not recognize her relatives, and is in a constant state of bewilderment. Figure 15.2G is completed by a patient who could not recognize his relatives, had no association with others, and was unable to understand the PPAT. Analysis results of drawings in Figure 15.2 are displayed in Table 15.3. For Figure 15.2A, the estimated MMSE-K score was 22 for the PPAT, 17 for the FSA, and 19 for Table 15.2 Means and standard deviations of selected variables

________________________________________________________________________ MMSE-K

PPAT _________________

FSA ________________

SMC _________

Realism

Line quality

Line quality

Num. of colors

Accuracy

14.5

2.5

3.0

3.2

4.5

45.5

5.9

1.0

1.3

1.0

1.7

26.5

________________________________________________________________________ Mean Standard deviation

________________________________________________________________________

Chapter 15 Effectiveness Comparison of Several Art Therapy Tools

241

________________________________________________________________________

Table 15.3 Analysis results of Num. of MMSE-K Line _______________ drawings in Figure 15.2 Figure Accuracy Realism colors quality Score Estimated score ________________________________________________________________________ 15.2A PPAT 6 - 5 4 22 FSA 5 - 5 4 28 17 SMC 7 76.3 - - 19 ________________________________________________________________________ 15.2B PPAT 3 - 4 4 20 FSA 5 - 3 4 24 17 SMC 7 74.2 - - 19 ________________________________________________________________________ 15.2C PPAT 6 - 4 4 20 FSA 3 - 5 4 20 15 SMC 7 54.6 - - 16 ________________________________________________________________________ 15.2D PPAT 5 - 3 4 17 FSA 5 - 4 4 19 17 SMC 5 73.7 - - 19 ________________________________________________________________________ 15.2E PPAT 5 - 3 3 16 FSA 3 - 4 4 16 15 SMC 6 42.1 - - 14 ________________________________________________________________________ 15.2F PPAT 7 - 3 2 14 FSA 6 - 2 3 11 15 SMC 3 27.9 - - 12 ________________________________________________________________________ 15.2G PPAT FSA SMC

2 5 6

- - 3.5

1 0 -

1 1 2 -

7 9 8

________________________________________________________________________ -: Elements not used in the corresponding art therapy tool.

the SMC. The actual MMSE-K score was 28. Thus, the PPAT achieved the closest estimate. For Figure 15.2D, the estimated MMSE-K score was 17 for the PPAT, 17 for the FSA, and 19 for the SMC. The actual MMSE-K score was 19. Thus, the SMC achieved the closest estimate. The ranking of the estimated MMSE-K scores for PPAT was exactly equivalent to the ranking of the actual MMSE-K scores. The rankings for the FSA and the SMC were almost the same as the actual ranking, except for Figure 15.2C. Now, we examine the effectiveness of the three tools based on the overall 58 Comparing samples. Figure 15.3a shows the plots of the actual MMSE-K scores versus effectiveness of three the estimated MMSE-K scores of the PPAT. If all of the estimated MMSE-K regression models are equal to the actual MMSE-K, then all points are on the line with a 45 degree slope. Figure 15.3b and Figure 15.3c show the plots of the FSA and the

242

PART TWO ART INTERPRETATION

Figure 15.3 Scatter plots of the actual MMSE-K scores versus the estimated MMSE-K scores.

SMC. The points in the left side indicate higher estimation of the MMSE-K, while the points in the right side indicate lower estimation. This may imply that nonlinear regression functions would be more appropriate. We measured the effectiveness of tools in the COD. The COD was R2 = 0.591, 0.390, and 0.494, for the PPAT, FSA, and SMC, respectively. All of these values were not satisfactory (R2 over 0.7 is desirable). The PPAT was marginally satisfactory, while the FSA and the SMC were not satisfactory, but they revealed valuable information about the level of dementia and thus were applicable. The nonlinear regression models improved R2 slightly, but it is not presented here. The results showed that the PPAT was the most effective, followed by the SMC. This may be due to the fact that the PPAT does not have patterns like the FSA and the SMC. Therefore, the clients can draw more freely, which may improve scores by line quality and realism compared to the other two. In conclusion, even though the R2 of PPAT, FSA, and SMC were not satisfactory, the ranking of the estimated MMSE-K and the actual MMSE-K were comparatively correlated. The Spearman’s Rank Correlation Coefficient (RCC) was R2 = 0.786, 0.576, and 0.639, for the PPAT, FSA, and SMC, respectively. Finally, the approach developed in this study can be used as an art therapy tool in estimating the MMSE-K scores. This is a similar approach to Wang, Ericsson, Winblad, and Fratiglioni (1998) in that the Human Figure Drawing (HFD) (Machover, 1949) can be a useful tool in estimating the MMSE scores. While the HFD cannot completely replace the MMSE in the case of patients with slight cognitive impediments, the HFD can, nonetheless, be effectively used in diagnosing dementia.

Chapter 15 Effectiveness Comparison of Several Art Therapy Tools

15.4 Discussion and conclusion We have developed a generalized approach using a regression model to compare any kind of art therapy tools such as the DDS, KFD, DAP, and HTP to estimate the levels of any kind of psychological disorders such as ADHD, sexual abuse, and depression. This study applied the approach to compare the effectiveness of the PPAT, FSA, and SMC in estimating the level of dementia by comparing the actual and the estimated MMSE-K scores. The scatter plots of correlations between actual and estimated MMSE-K scores show the relative effectiveness of the regression model for the three tools. The CODs of the regression models for the PPAT, FSA, and SMC were R2 = 0.61, 0.45, and 0.59, respectively, which indicate that all three tools were marginally satisfactory or applicable and that the PPAT was the most effective of the three, and the SMC the second most. The ranking of the estimated MMSE-K and the actual MMSE-K were partly correlated. The Spearman’s RCC was Rs = 0.786 for the PPAT, 0.576 for the FSA, and 0.639 for the SMC. The level of dementia for 58 patients with suspected dementia in a psychiatric unit was measured by the MMSE-K. There are other tests such as the Geriatric Deterioration Scale (Yesavage, Brink, Rose, Lum, Huang, Adey, & Leirer, 1983) and the Clinical Dementia Rating (Morris, 1993) that measure the level of dementia. The size of our sample may not be large enough. Also, the sample was obtained from one single psychiatric unit. Thus, more tests and a larger sample size from a broader range of units across different regions might yield a different conclusion. We might obtain improved results when the study is targeted at people with slight dementia (Wang, Ericsson, Winblad, & Fratiglioni, 1998). Also, we can obtain improved R2 through the nonlinear regression models. Many art therapists may struggle to convince other mental health professionals on the effectiveness of art therapy tools. This approach of choosing an effective tool and analyzing the relationship between the tool’s estimated values and the actual test scores may assist art therapists who are seeking to find evidence to support their observations and evaluations for official documents. It may not necessarily suggest how to diagnose the clients, but it can assist the judgment process. As this approach is generalized, we can extend the application of this approach to any art therapy tool and to any psychological state in comparing the effectiveness as long as we have reliable psychological test scores and suitable drawing elements to evaluate. This approach can also support the results of psychological tests for clients with limited ability in speaking, reading, and comprehension. Furthermore, it can be applied to statistically identify (or predict the probability of existence of) a psychological disorder, which could aid the art therapist’s judgment process of their clients’ conditions. We note that the probability of having a certain psychological dis-

243

244

PART TWO ART INTERPRETATION

order can be predicted using the regression model described in Chapter 14 (Kim, Betts, et al., 2009). The only difference in this model is that when the dependent variable Y scores below a certain threshold value, it indicates the existence of a psychological state. The approach developed in this chapter is expected to promote the use of appropriate art therapy tool and to pave the way for the development of new methodologies of art therapy analysis and application, and to advance the theoretical understanding of art therapy.

Probabilistic Art Interpretation Using Bayesian Network

Chapter 16 Abstract and keynote Probabilistic art interpretation using the Bayesian Network (BN) is devised for resolving the issues of diversity, uncertainty, and even contradiction in traditional deterministic art interpretation. The BN represents cause-and-effect relationships between factors such as psychological states, disorders, backgrounds of drawers, and elements of drawings via a directed acyclic graph. We illustrate the principle, usefulness, and mechanism of the BN with a simple example based on real data. Through a case study with numerous nodes and its simulation experiments, we demonstrate how an “objective” interpretation of art can be obtained by the BN. The BN can contribute to opening new horizons in the practice and theory of art therapy.

16.1 Probabilistic interpretation vs. deterministic interpretation People feel awkward or have trouble with exposing their negative feelings, such as grief, fear, frustration, and worry. They especially have difficulties in representing these feelings in words. A drawing contains many elements that are believed to reveal a drawer’s emotional state. For most people, a drawing is known as a more comfortable medium than speech in revealing one’s feelings. Since expressions and symbols in a drawing represent conscious and unconscious states of a person’s inner world (Jung, 1958), many art therapists interpret drawings to effectively evaluate and treat client’s psychological states or symptoms. However, the main difficulties in the interpretation of art lie in the fact that Difficulties a client’s personal and cultural background influences his or her choice of shapes, colors, and styles in the drawings. This implies that even seemingly identical drawings may be subject to diverse, inconsistent, and sometimes 245

246

PART TWO ART INTERPRETATION

even contradictory interpretations. This phenomenon seems to be natural because the domain of art therapy is classified into a so called Ill-Structured Paradigm (ISP) (Giarratano & Riley, 2005) in the artificial intelligence domain. An ISP mirrors problems where data are conflicting or inclusive. For example, one study found that persons in a depressed state were likely to draw human figures smaller than those who were not in such a state (Lewinsohn, 1964), whereas another study reported no relationship between depression and the size of human figure drawings (Salzman & Harway, 1967). Omission of the mouth in Human Figure Drawings (HFD) meant that a child tends to be shy and lack leadership (Lewinsohn, 1964), whereas Koppitz (1968) reported that omission of the mouth reveals fear, anxiety, and depression. King, Gullon, and Ollendick (1992) found that girls were more likely to feel fear than boys, whereas Carroll and Ryan-Wenger (1999) reported that there was no difference between boys and girls in the degree of fear experienced. A solution Judgment of mental states and treatments for psychological disorders based on traditional art therapy methodologies proves problematic in achieving objective and consistent conclusions and serving as a basis for administration of correct cures for clients, since they depend on the subjective knowledge and individual experience of each art therapist. As a solution for diversity and uncertainty, this study adopts the concept of probability in the interpretation of art. The traditional deterministic interpretation of art is, for example, that children suffering from ADHD tend to draw a full picture. However, the probabilistic interpretation that children suffering from ADHD draw a full picture with a probability of 0.75, and normal children do so with a probability of 0.1, will elevate the logical interpretation to a higher level. It may open new horizons in art interpretation. This study applies the Bayesian Network (BN), which visually represents cause-and-effect relationships among a set of random variables (nodes) and their conditional independencies, via a directed acyclic graph (Charniak, 1991). Specifically, the BN is a useful tool widely applied to research where uncertainty is present. In particular, by providing a system in which the advantages of humans and computers can be integrated, the BN has been widely applied in areas where uncertainty prevails (Burnside, 2005). For example, Cowell, Dawid, Hutchinson, and Spiegelhalter (1991) expressed the causal relationship between adverse drug reactions and pseudomembranous enterocolitis through a BN based expert system. Weber, Medina-Oliva, Simon, and Jung (2010) used BNs to analyze credibility as well as to manage risks, and proved that the BN is better for modeling complicated systems, forecasting, and diagnosing results when compared with the traditional studies of Markov chains, fault tree analysis, and petri nets. The BN has not, until now, been applied to art therapy, but has proven its effectiveness in the field of medical diagnosis. As an example, Cruz-Ramirez, Acos-

Chapter 16 Probabilistic Art Interpretation Using Bayesian Network

ta-Mesa, Carrillo-Calvet, Nava-Fernandez, and Barrientos-Martinez (2007) used seven BNs to accurately diagnose breast cancer through fine-needle aspiration of the breast. They also used the BN to present feelings of subjects with probability values. Due to the ISP nature of the art therapy field, it is inevitable for causeand-effect relationships and their conditional probabilities to be subjectively determined. We provide an example of the BN whose conditional probabilities are given objectively. As a case study, we propose a BN comprised of different types of nodes: Mental disease nodes, drawer’s environment nodes, psychological symptom nodes, and drawing element nodes. Schizophrenia, ADHD, and depression are chosen as the subjects of the case study, since these conditions affect a rapidly growing number of people worldwide. These mental diseases become significant social problems, since they not only make victims feel a sense of isolation and loss, but also occasionally lead to death. The cause-and-effect relationships between the nodes are established via a directed acyclic graph based on the literature. The conditional probability of a node, given its parent node, is assigned by the experience and knowledge of experts. The effects of these conditional probabilities on the parent nodes are experimented through simulation. Simulation is a method by which we may overcome the limit of subjective conditional probabilities.

16.2 Methods The BN is a very effective approach in the areas of complication and uncertainty, because it only considers variable nodes that have higher causal relationships to present a directed acyclic graph. It expresses probability distribution by defining conditional probabilities for variables that have direct causal relationships, and assuming mutual independence for other variables (Korb & Nicholson, 2004). The BN explains how or why it reaches a specific conclusion by not only presenting the correlation of variables visually, but also providing the conditional probabilities of variables (Cruz-Ramirez et al., 2007). Knowledge in the field of art interpretation is subjective, empirical, and uncertain in nature. Knowledge that does not come in the form of writing is very difficult to interpret through heuristics or algorithms. The BN can be a useful method to systematize and materialize knowledge in the fields where knowledge itself is unorganized. The BN enables art therapists to comprehend the cause-and-effect relationships between the variables of mental diseases, psychological symptoms, drawers’ environments, and drawing elements, and to integrate individual knowledge. Also, the BN can aid art therapists to make a more objective and logical decision by reasoning the interactions of variables based on probability values.

247

248

PART TWO ART INTERPRETATION

Figure 16.1 A Bayesian network example.

The principle and value of BN can be illustrated with a simple example. Figure 16.1 is a BN composed of seven nodes of dementia, family_history, female, memory_impairment, disorientation_to_place, use_of_purple, and use_of_brown. Henceforth, these nodes will be mentioned in italics. Causeand-effect relationships between parent node(s) and child node have been determined based on various literature and research. Letenneur, Gilleron, Commenges, Helmer, Orgogozo, and Dartigues (1999) reported that 4.86% of 2,881 elderly people (over the age of 65) showed dementia, and that women had a higher prevalence rate of dementia (5.8%) than men (3.5%). Devi, Ottman, Tang, Marder, Stern, and Mayeux (2000) found evidence for familial aggregation of dementia (family_history) by studying 1,577 persons who were relatives of patient with dementia (7.2% of subjects suffered from dementia) and 3,952 persons who were relatives of people without dementia (4.6% of subjects suffered from dementia). Kim, Betts, et al. (2009) showed through a statistical approach that severe dementia affected memory_impairment and disorientation_to_place, people with memory_impairment tended to use purple (use_of_purple) more than other colors in their drawings, and people having disorientation_to_place tended to use brown (use_of_brown) more than other colors in their drawings. Let us denote the conditional probability of a given b is true as P(a | b) and the conditional probability of a given b is not true as P(a | bc). In Figure 16.1, based on the above reports, the probability that elderly people suffer from dementia is given as 0.05, P(family_history | dementia) = 0.07, P(family_history | dementiac) = 0.05, P(female | dementia) = 0.56, P(female | dementiac) = 0.50, P(memory_impairment | dementia) = 0.38, P(memory_impairment | dementiac) = 0.12, P(disorientation_to_place | dementia) = 0.80, P(disorientation_to_place | dementiac) = 0.03, P(use_of_purple | memory_impairment) = 0.07, P(use_of_purple | memory_impairmentc) = 0, P(use_of_brown | disorientation_to_place) = 0.07, and P(use_of_brown | disorientation_to_placec) = 0.01.

Chapter 16 Probabilistic Art Interpretation Using Bayesian Network

249

Figure 16.2 Computation of the probability value.

Now, when the evidence of a node(s) is revealed, this information will be used to revise the probability value of other nodes. Open software such as that provided by Microsoft (2011) automatically computes the probability. For example, when the evidence that one of the relatives of an elderly person is suffering from dementia is revealed (family_history), the probability that the old person might have dementia increases from the prior probability of 0.05 to 0.07, as illustrated in Figure 16.2a. If the old person shows memory_impairment and it is found that brown is colored the most in his or her drawing (use_of_brown), the probability that the old person might have dementia rises to 0.25, as illustrated in Figure 16.2b.

250

PART TWO ART INTERPRETATION

16.3 A Bayesian network-based art interpretation This chapter addresses people under 30 years old with suspected depression, ADHD, and schizophrenia. The World Health Organization (2012) reported that every year almost one million people commit suicide, and in the last 45 years suicide rates have increased by 60% worldwide. It appears that these people killed themselves because of disease and economic distress including unemployment, but many experts argue that depression might have a great effect on suicide. Unlike in the past, the age group that shows symptoms of depression due to environmental factors is becoming younger, and when depression is caused by childhood psychological diseases such as ADHD, there is a high risk of it leading to behavioral disorders such as aggressive or anti-social behaviors. According to Lim (2010), 1,212 (11.4%) of 89,629 elementary students participating in a survey as part of a mental health diagnosis project were classified as an ADHD suspect group. ADHD symptoms are specifically related to one’s environment, such as financial instability of the family, poor educational background of the parents, and a deteriorating relationship with the parents. In addition, ADHD children were reported to exhibit behavior such as impulsivity, attention deficit disorder, and a number of psychological behavioral problems including being a loner or learning disorders. It was reported that schizophrenia, one of the most representative mental diseases, occurred with a probability of about 1% within a lifetime, and shows symptoms of hallucinations and delusions (Fletcher & Frith, 2009). Example Based on the above mentioned facts the causal relationships between several different nodes were represented via a BN, as illustrated in Figure 16.3. It includes 3 mental disease nodes of depression, ADHD, and schizophrenia. It has 5 environmental nodes of parents’_divorce, family_abuse, economic_hardship, unemployment, and social_isolation. It includes 10 psychological symptom nodes composed of prominent_hallucinations, loosening_ of_associations, psychomotor_retardation, attention_deficit, impulsivity, insomnia, frustration, feeling_of_suicidal, poor_academic_performance, and learning_disorder. The network also includes 8 drawing element nodes including lack_of_color_fit, low_implied_energy, less_space, lack_of_integration, illogicalness, lack_of_realism, lack_of_problem_solving, and missing_details. An arrow directly connects two nodes with a high causal relationship. The cause-and-effect relationship of the directly connected nodes was based on various literatures, including the Diagnosis and Statistics of Mental Disorders, Vol. 4 (DSM-IV) (American Psychiatric Association, 1994), and the FEATS. For example, Gantt and Tabone (1998) reported that patients who were suffering from depression were more likely to use less colors or darker colors, compared to patients without depression; and depressed people

Chapter 16 Probabilistic Art Interpretation Using Bayesian Network

251

Figure 16.3 A diagram of the Bayesian network.

showed not only psychomotor retardation and diminished abilities to concentrate, but also the characteristics of lack of color fit, and lack of space usage in the drawing. These cause-and-effect relationships were reflected in a directed acyclic graph as two nodes connected by an arrow in Figure 16.3 such that the depression node is directly connected to the psychomotor_retardation node, attention_deficit node, lack_of_color_fit node, and less_space node. Dawson (1984) mentioned that patients with depression tended to use less_space than those without depression. According to DSMIV, ADHD children exhibited badly damaged social (social_isolation) and academic functions (poor_academic_performance), and showed a number of symptoms including impulsivity. A person with schizophrenia showed symptoms such as prominent_hallucinations and loosening_of_association. These symptoms were reflected as lack_of_integration, illogicalness, and lack_of_realism in their drawings. Cronin and Werblowsky (1979) stated that people with mental disorders, by temperament, skipped a great majority of detailed descriptions (missing_details). The probabilities of schizophrenia and ADHD in Figure 16.3 are giv-

252

PART TWO ART INTERPRETATION

Figure 16.4 Sensitivity analysis of P(ADHD).

en as 0.01 and 0.11, which are based on Fletcher and Frith (2009), and Lim (2010), respectively. Unfortunately, statistical data and studies of the conditional probability between the psychological symptoms, drawer’s environment, and drawing elements are nonexistent so far. For this reason, subjective probability values inferred from the professional knowledge and experience of art therapists, psychiatrists, and psychologists have been used for the conditional probabilities of child nodes. Experts in various fields need to collaborate on more objective data and statistics. In this study, the conditional probabilities between nodes were determined based on the knowledge and opinions collected from three art therapists.

16.4 System verification The validity of these conditional probabilities has been verified through a simulation experiment. Figure 16.4 shows the simulation results. It is a sensitivity analysis of P(ADHD) with varying conditional probabilities of several nodes. We see P(ADHD) was very sensitive to the changes of the conditional probability of its directly connected child nodes. Varying P(ADHD | depression) and P(ADHD | impulsivity) resulted in great change of P(ADHD). P(ADHD) was not that sensitive to the nodes that were not directly connected to the ADHD node. For example, varying each of P(ADHD | learning_disorder), P(ADHD | unemployment), P(ADHD | psychomotor_retardation), and P(ADHD | missing_details) did not change P(ADHD) much, compared to varying P(ADHD | depression) and P(ADHD | impulsivity). Experts including art therapists, psychiatrists, and psychologists participated in the process of sensitivity analysis, and shared views on the given conditional probabilities are reasonable.

Chapter 16 Probabilistic Art Interpretation Using Bayesian Network

When the evidence of a node(s) becomes known, the posterior probability values of nodes are calculated with the evidence(s) reflected. This calculation is extremely complicated and time-consuming, but it can be easily obtained by available open software (Microsoft, 2010). For example, if it is known that a child lives in a needy family (economic_hardship), the probability that the child suffers from ADHD rises from an existing prior probability of 0.1 to a post probability of 0.24. Furthermore, if it is found that a child tends to be impulsive (impulsivity), the probability of ADHD increases from 0.24 to 0.52. In addition to that, if it is known that the child is suffering from a learning_disorder, and also uses inappropriate colors in a drawing (lack_of_color_fit), the probability that the child has ADHD increases from 0.52 to 0.68. Another example shows that the probability of depression rises to 0.18 from a prior probability of 0.11 if it is known that the child suffers from insomnia. Additionally, if the child uses very little space (less_space) and skips a majority of details (missing_details) in his or her drawing, the probability of having depression rises from 0.18 to 0.36.

16.5 Discussion and conclusion There is still heated debate among art therapists about whether or not art interpretation can be objectively measured, and whether personal judgments by non-art therapists can be allowed in art interpretation (Gantt, 1998). Although art therapy as a professional discipline has been in evidence since the mid-1950s, it has achieved only limited acceptance in the larger mental health field. The major reasons for this failure of art therapists to be taken seriously as mental health professionals is because art therapy, as a discipline, has failed to generate empirical data that would support its claims to clinical effectiveness (Tibbetts, 1995). Some art therapists stressed that integrating art and science in art therapy can be a solution to resolve the implicit ambiguity and uncertainty of art therapy (Kaplan, 2000). Kaplan (1998) reported that subjective impressions need to be carefully evaluated in the light of existing empirical evidence, and although science does not yet, and may never, supply all the answers to what the mind-body means, rapid progress is being made in that direction and much is already known. In practice, artificial intelligence techniques of expert systems, BN, computer algorithms, and statistical models are emerging as a solution to overcome the limitations of the art therapy domain. With the BN, we expanded the application of art interpretation by enabling probabilistic interpretation instead of the traditional deterministic interpretation. The BN enables art therapy experts to more easily communicate their theories and results, by showing probabilistic values along with cause-and-effect relationships in a directed acyclic graph (Uusitalo, 2007). The individual interpretations of each art therapist can be accumulated into

253

254

PART TWO ART INTERPRETATION

a shared knowledge based on the BN, and more logical interpretations can be formulated, based on that knowledge. It seems promising that a convergence of the BN and the experiences and knowledge of art therapists can provide both quantitative and qualitative aspects in the art therapy domain. The nature of art therapy is more likely to be qualitative than quantitative, thus it is not restricted to an unambiguously demarcated data set (Babor, Stenius, Savva, & O’reilly, 2009). Art therapists can construct reasonable cause-and-effect relationships among psychological disorders, symptoms, and elements in a drawing through their experiences and interventions. This knowledge can be used as a basis for developing the directed acyclic graph in the BN. An example with seven nodes based on real and statistical data was presented to illustrate the principles and mechanisms of the BN. It was shown how objective cause-and-effect relationships and conditional probabilities were obtained. A case study established the cause-and-effect relationships among 26 nodes based on the literature. However, the conditional probabilities were determined subjectively, based on professional experience and knowledge. There is nothing wrong with gathering probabilities subjectively from experts (Charniak, 1991). For example, doctors’ subjective conditional probabilities for a BN representing congenital heart disease were compared to probabilities that are subsequently collected, and found to be pretty close (Spiegelhalter, Franklin, & Bull 1989). For the ideal case, these probability values should be revised to be more credible, based on objective and statistical facts. Furthermore, experts in art therapy, psychiatry, and psychology have to work together to select nodes for art interpretation, set the cause-and-effect relationships, and provide conditional probabilities. Simulation experiments helped us understand the causal relationships between nodes and the effects of nodes on other nodes. The BN methodology with the expert systems in Chapters 9, 10, 11, and 12 (Kim, Ryu, et al., 2006; Kim, Kim, et al.; Kim, You, et al., 2007; Kim, Kim, & Kim, 2008), the statistical models in Chapters 1, 2, 6, 7, 8, 13, 14, and 15 (Kim, Betts, et al., 2009), and the computer algorithms in Chapters 3, 4, and 5 (Kim, Bae, & Lee, 2007; Kim, Kang, & Kim, 2008; Kim, 2008a; Kim, Kang, & Kim, 2009; Kim & Hameed, 2009), are expected to contribute to widening the horizon of the practice and theory of art therapy.

Searching for Advancement of Art Therapy

Epilogue In this book, we have continuously emphasized an interdisciplinary ap- Interdisciplinary proach for art therapy and presented various computer technologies which approach can be used to advance its practice and theory. In today’s world, we often hear terminologies common to the data mining field such as big data, computer algorithm, machine learning, and artificial intelligence. Data mining is a hybrid disciplinary (Zhou, 2003) that integrates technologies of databases, statistics, machine learning, signal processing, and high performance computing. A number of data mining applications and prototypes have been developed for a variety of domains (Liao, 2003; Mitra, Pal, & Mitra, 2002), including marketing, banking, finance, manufacturing, and health care. When data mining deals with the data of the unstructured type, such as Image mining drawings, photographs, or X-rays, as opposed to the structured type such as alphanumeric characters, we call it image mining. Art therapy has been developed as a separate field of image mining. It is interesting that art therapy fits the exact definition of image mining, which is as the following (in the parenthesis are the corresponding word examples used in the case of art therapy). Image mining can be classified into two kinds (Kannan, Mohan, & Anbazhagan, 2010). One is a domain specific application where the focus is on the process of extracting the most relevant image features (colors used, accuracy, concentration) into a suitable form (number of used colors, percentage of accuracy, degree of concentration). The other is a general application where the focus is on the process of discovering image patterns (level of dementia as a function of accuracy, knowledge on the relationship between the number of used colors and the level of anxiety) that may be helpful in the understanding (interpretation) of the interaction between low-level image features (accuracy, colors used) and high-level human perception (level of dementia, anxiety) based on a single image (a 255

256 Epilogue Searching for Advancement of Art Therapy drawing). This image mining technology provides different methodologies for decision-making, problem solving, analysis, planning, diagnosis, interpretation, detection, integration, prevention, learning, and innovation. All of these approaches, technology, systems, and methods of image mining can be effectively used and open up new horizons for art therapy. Role of AATA The reason why projective techniques and art interpretations are often viewed with skepticism lies in their lack of reliability in the evaluation of elements and also the lack of validity in interpretation, due to the subjectivity associated with the personal or intuitive decisions of an art therapist. To remedy the lack of reliability in evaluating elements and extracting new useful elements, we have developed the computer system, C_CREATES, designed to provide accurate and objective information on 19 elements in the form of quantitative data derived from analysis and evaluation. Through quantitative results, the system can evaluate these elements on a more detailed scale than traditional methods. For example, in the evaluation of prominence of color, assigning a ranking in the sample size provides more precise and comprehensive information than the kind provided by the Likert type. Experts in art therapy can compose a standard sample of enough size which appropriately represents the whole range of variations for each of the elements and establish a system based on this sample. From this, certain official institutions, such as the AATA, can be expected to recognize the system as a standard method and criteria for the evaluation of the elements. Computer evaluation has another obvious advantage. In cases where a large number of drawings need to be analyzed, evaluated, and interpreted, for example, 471 drawings (Couch, 1997) and about 4,000 drawings (Cox & Cohen, 2000), the reduction of fatigue, time, and effort of human raters can be one of the greatest merits of a computer system. To remedy the lack of validity in interpretation resulting from the illstructured paradigm nature of art therapy, this book has presented several computer applications. As has been pointed out in the preceding pages, art interpretation by therapists is marked by frequent discordances and contradictions. An expert system can serve as a solution for the problem of interpretative inconsistency by establishing a knowledge base. The diverse and even contradictory knowledge can be classified and systemized. An expert system can display the occurrences of contradictory interpretations; it can assign a confidence level to each knowledge; it can also display the knowledge with a probability value. The relevant data then will be placed at the disposal of the human expert for determination. Which knowledge is selected, how it is classified and systemized, and which value of confidence level is assigned will be the decisions to be made by an authoritative organization of art therapists such as the AATA. The usefulness of statistical methods in art interpretation cannot be exaggerated. Despite the failures and warnings in using only projective meth-

Epilogue Searching for Advancement of Art Therapy

257

ods to identify, differentiate, and estimate psychological states, we have developed several statistical models. The reason for its success may be ascribed to the appropriate tool, which is SMC and the suitable samples from patients in hospitals of large size. In view of the current incipient stage of computer technology in understanding arbitrary drawings, we recommend not only the SMC, the p_KFD, and the Computerized Face Stimulus Assessment (c_FSA) (Kim, Kim, & Hong, 2003), but also the DAP, the HTP, and the Kinetic School Drawing (KSD) with given patterns. Once drawings are composed of given patterns and are colored, the computer can process comprehensively all the elements of freely-drawn drawings. Thus, a series of the SMC, the p_KFD, and the patterned KSD with patterns is conceivable: one asks children to fill out a questionnaire on their color preference, color a SMC drawing, and then complete a p_KFD and a patterned KSD. These tools can easily be used, enabling art therapists to obtain information on the children’s status in their primary living space such as home and school. In both art evaluation and art interpretation, therapists can use various tech- Joint work niques including the built-in functions of a computer, digital image processing, computer algorithms, statistics, expert systems, database management, and Bayesian analysis. Art therapists likely do not know all of these techniques and they in fact do not need to know them all. For example, in applying statistical methods, there are various statistical techniques and statistical software packages available, such as the SAS and the SPSS, which can perform the necessary tasks of calculating, analyzing, and inferring. A statistical analysis of a problem or a situation may require expert statistical advice. The issue is then to know how to bridge the gap between the ‘person on the spot (art therapists)’ who knows the factors involved and the statistician who knows the right techniques. Clearly, this joint work process requires an individual art therapist to know the basic concepts of statistics such as correlation coefficient, significance level, Type I and Type II errors, coefficient of determination, etc. in order to communicate effectively with a statistician. First of all, art therapists should collect suitable data to use the statistical methods to good effect and should not forget that simple methods are often the most powerful - do not make them more complicated or inhibiting if not necessary! We believe it would be ideal if one semester course on each application of computer methods and applied statistical methods could be included in undergraduate art therapy programs. The purpose of this book has been to describe the creation of computer CAT applications which will provide critical tools for the field of art therapy. We hope these applications will pave the way for the development of new technology for the analysis and application of art therapy, as well as its theoretical understanding. A large amount of further study remains to be

258 Epilogue Searching for Advancement of Art Therapy done. We suggest that this interdisciplinary study, incorporating art therapy, psychology, psychiatry, art, computer science, and applied statistics, be undertaken under the umbrella of a new field called the Computational Art Therapy (CAT). The active participation of professionals in these fields is essential for the development of this type of system. We fully acknowledge that the system requires much time, effort, manpower, and cost before its practical use can be envisaged. Finally, the issue of whether human experts can be substituted by such a computer system should be briefly mentioned here. The proposed system can offer various types of information rapidly and accurately, and help the art therapists’ decision-making process. The final decision remains in the hands of art therapists. However, there is no reason for an individual art therapist to refuse the information which he or she cannot possibly extract or for which he or she has to spend much time and effort. A computer system can provide much service for them.

Appendix: Companion S/W

All the computer systems for art evaluation in Part One are visualized as a software package. This software provides the analysis process and evaluation results of 19 elements in the C_CREATES (Computer Color-Related Elements Art Therapy Evaluation System) (Kim, Bae, & Lee, 2007; Kim, 2010). However, the software does not include the computer systems for art interpretation in Part Two, which is beyond the scope of a book software and readers are advised to use MATLab (MATrix Laboratory) (MathWorks, 2016) for expert systems and Statistical Package for the Social Sciences (SPSS) (IBM Corporation, 2015) or Statistical Analysis System) (SAS) (SAS Institute, 2014) for regression methods.

1. Software installation, five main menus, and Q&A Load the CD included in the back of this book. Or to purchase a download of the software visit www.ccthomas.com/cat. On the CD, click on the C_ CREATES file and follow the installation instructions. When the software is will appear on your desktop. Click it, and successfully installed, an icon you will see Screen-1 having five menus in top line indicated by red circle that will perform main functions and twelve windows that will show analysis process and evaluation results in graphics. The five menus are: [Load], [Graphics], [Save], [Parameters], and [Quit] which, respectively, loads a drawing to be evaluated, provides the process and results in graphics, save the outcomes in your PC, sets system parameters, and quit the program. This is version 1.0 firstly being delivered to readers. Click [Version 1.0] located top-left indicated by white circle, and the system provides the manual of this software. Before we proceed further, we ask, if you have any questions, suggestions for improvements, and reports on program errors, please do not hesitate to contact us by clicking [Email Technical Support] located bottom-right indicated by yellow circle.

259

260 Appendix: Companion S/W

Screen-1 Menus and graphic information.

2. Steps Step-1. Loading a drawing to be evaluated Click [Load], and then you will see six sub-menus asking what type of drawing you want to evaluate, as in Screen-2: {Free}, {SMC}, {FSA}, {PPAT}, {KFD}, or {Etc.}. Choose one, and you will have two options asking methods to load a drawing from: {PC} or {Sample}. When you choose and click {PC} and , follow the instruction to select a drawing that is stored in your PC. Alternatively, when you choose and click {Samples} and , there appear sample drawings that were used in this book, as in Screen-3. Click the one you have selected among drawings. Then the drawing you have selected will appear in top-left window (Window-1), as in Screen-4. If you like to change the drawing to be evaluated, repeat Step-1. Step-2. Graphics showing analysis process and evaluation results Cases of Free, PPAT, KFD, Etc. Click [Graphics] as shown in Screen-5, and eleven windows (Window-2 to Window-12) showing the analysis process and evaluation results will appear, as in Screen-6. We call this Graphics-panel. As you see, each window has its name, main for expansion. information in numbers or words if it has, and a button

Appendix: Companion S/W

261

Screen-2 Menu Load.

Screen-3 Samples.

262 Appendix: Companion S/W

Screen-4 Original drawing to be evaluated.

If you have selected {Free}, {PPAT}, {KFD}, or {Etc.}, the following windows will appear showing the analysis process or evaluation results in graphics. Window-1 (Original): Original drawing you want to evaluate. Window-2 (Blurring): Image after blurring. Window-3 (Clustering): Image after clustering and number of clusters. Window-4 (Color classification): Image after color classification into 15 colors and the number of used colors. Window-5 (Edges): Edges detected and their total length of edges. Window-6 (Placement): Placement and its category among C (Center), T (Top), B (Bottom), L (Left), R (Right), TL (Top Left), TR (Top Right), BL (Bottom Left), and BR (Bottom Right). Window-7 (Grids): Grids and number of colored grids.

Screen-5 Menu [Graphics] selected.

Appendix: Companion S/W

263

Screen-6 Graphics-panel: Analysis process and evaluation results of a free drawing.

Window-8 (Convex hull): Convex hull colored and area of colored convex hull in percentage. Window-9 (Main color): Main color judged and its color name. Window-10 (Color type): Primary (in yellow) / Secondary (in green) colors, Warm (in red) / Cool (in blue) colors, or Complementary colors. Window-11 (Area of colors): Histogram or Pie chart showing area of each color. Window-12 (Others): Variety, Prominence, Detail, and Space usage in a grade of 6-point Likert type and in a rank of 100 drawings. (Rank only for variety.)

Screen-7 Window examples.

264 Appendix: Companion S/W

Screen-8 A Graphics-panel of SMC.

Screen-7a is an example of Window-9 (Main color), showing window , and the name “Main color,” main color “Yellow,” magnification button when you want to magnify a drawing analysis results in graphics. Click in the window. Note that Window-10 (Color type) has 3 options: {P/S}, {W/C} and {C/C} which, respectively, evaluate the primary / secondary colors, the warm / cool colors, and the complementary colors. Click {P/S}, and the primary colors appear in yellow and the secondary colors appear in green. Click {W/C}, and the warm colors appear in red and the cool colors appear in blue. Click {C/C}, and one set of the complementary colors appear. Screen-7b is an example showing the warm / cool colors. Also, note that Window-11 (Area of each color) has 2 options: {H} and {P} which presents the results in histogram and pie chart, respectively. Screen-7c is an example of Window-11 showing the results in a pie chart. Case of SMC The windows in the case of {SMC} are shown in Screen-8. Note that Window-6 “Placement” in the case of {Free} has been replaced with “Pattern” showing the given geometric lines, and Window-7 “Grid” with “Accuracy” showing the accordance in coloring with the given pattern and its evaluation in percentage, and Window-8 “Convex hull” with “Completeness” showing how much the coloring is done and its evaluation in percentage.

Appendix: Companion S/W

265

Screen-9 A Graphics-panel of FSA.

Case of FSA The windows in the case of {FSA} are shown in Screen-9. Note that Window-6 “Placement” in the case of {Free} has been replaced with “Pattern” showing the facial contours, and Window-8 “Convex hull” with “Area division” showing the area divided for the evaluation, and Window-12 adds more 4 elements. Step-3. Statistics of evaluation results Now the menu [Graphics] has been changed to [Statistics] as you shown in Screen-10. Click [Statistics], and you can see the evaluation results of 19 elements in quantitative statistics, as in Screen-11. We call this Statistics-panel. It consists of evaluation results of 19 el-

Screen-10 Menu [Statistics] selected.

266 Appendix: Companion S/W

Screen-11 A Statistics-panel.

ements in C_CREATES. The elements not relevant to the type of drawing selected are indicated by “N/A” such as the accuracy, completeness, and concentration in {Free}, {PPAT}, and {Etc.}, number of colored grid, placement, and area of colored convex hull in a {SMC}, and placement and area of colored convex hull in {FSA}. Note that in {FSA} five elements are added in “20. Others,” and “18. Space usage (R/G)” is changed to “N/A.” Basic elements 1. Number of used colors. The colors are limited to ones that occupy more than given percentage (Parameter-1), e.g., 5%, of whole colored area. 2. List of used colors. Three colors are given that occupy most colored area. 3. Number of clusters. 4. Length of edges. Total length of edges is given in number of pixels. 5. Area of each color. The area colored by each of 15 colors given in percentage of whole colored area. 6. Number of colored grids. The grids are limited to ones whose colored area is more than given percentage (Parameter-2), e.g., 30%. 7. Area of colored convex hull (%). 8. Completeness (%). 9. Accuracy (%). 10. Primary / secondary colors. They are presented in percentages of whole

Appendix: Companion S/W

267

Screen-12 Menu [Graphics] selected.

colored area. 11. Warm / cool colors. They are presented in percentages of whole colored area. 12. Complementary colors. Color lists are presented. Applied elements 13. Variety of colors. It is presented in a rank of 100 drawings. 14. Placement. The placement is classified into one of 9 categories: C (Center), T (Top), B (Bottom), L (Left), R (Right), TL (Top Left), TR (Top Right), BL (Bottom Left), and BR (Bottom Right). 15. Main color. Colors are currently classified into 15 colors, but we will allow more colors in the future. 16. Prominence of color. It is presented in a grade of 6-point Likert type and in a rank of 100 drawings. 17. Details. Same as above. 18. Space usage. Same as above. 19. Concentration. Same as above. Now the menu [Statistics] has been changed to [Graphics] as shown in Screen-12. Click [Graphics], and you return to Graphics-panel. In this way, you can across back-and-forth between the two panels. Step-4. Saving outcomes Click [Save] as shown in Screen-13, and then the results in both Graphic-panel and Statistics-panel, even when you have not yet seen the Statistics-panel, are saved in directory and under file name you have designated.

Screen-13 Menu [Save] selected.

268 Appendix: Companion S/W

Screen-14 Setting system parameters.

While Graphic-panel is saved in an image file, Statistics-panel in a text file. Step-5. Repeat or Quit Click [Load] again, and you can start evaluating another drawing. When or [Quit]. you want to close the window, click

3. System Parameters We note that there are 7 system parameters affecting analysis process and evaluation results related to extent of blurring and clustering, calibration for color classification, criteria for a color to be considered being used, a grid to be colored, and placement to be considered. You can control the system by setting these parameter values, as in Screen-14. The default parameter values are shown in the screen. Blurring has four options for area: {none}, {3x3}, {5x5}, and {7x7}. When you choose {none}, no blurring occurs. The effect of blurring is greatest when you choose {7x7}. Clustering has three options for area (radius): {3x3}, {5x5}, and {7x7}. The effect of clustering is greatest when you choose {7x7}. It also has three options for minimum value to control the degree of clustering: {40%}, {50%}, and {60%}. The effect of clustering is greatest when you choose {40%}.

Appendix: Companion S/W

We note that color can be classified incorrectly. There are various reasons for this, such as the coloring material of crayon or marker, scanners or monitors being used, or the environment of surrounding lights, etc. To solve this problem you can calibrate the color classification by setting the relative weights of colors. All initial values are set as 100. Suppose pink is often incorrectly classified into red. Then set the weights of pink as a lower value, e.g., 95. The appropriate values are determined by trial and error. Also, the system has some criteria, such as the minimum percentage of area colored by a color in the whole area colored for the color to be considered as one used. It is the same for the grid colored and the placement.

269

Glossary

Acronyms AATA : American Art Therapy Association ADHD : Attention Deficit and Hyper-Activity Disorder ANOVA : ANalysis Of VAriance BN : Bayesian Network CAT : Computational Art Therapy C_CREATES : Computerized Color-Related Elements Art Therapy Evaluation System (Kim, Bae, & Lee, 2007; Kim, 2010) c_FSA : Computerized Face Stimulus Assessment (Kim, Kim, & Hong, 2013) COD : Coefficient Of Determination c_SMC : Computerized Structured Mandala Coloring (Kim, Betts, Kim, & Kang 2009) DAP : Draw A Person (Goodenough, 1926) DDS : Diagnostic Drawing Series (Cohen, 1986/1994) FB : Fact Base FEATS : Formal Elements Art Therapy Scale (Gantt & Tabone, 2003) FSA : Face Stimulus Assessment (Betts, 2003) HFD : Human Figure Drawing (Machover, 1949) HTP : House-Tree-Person (Buck, 1949) ISP : Ill-Structured Paradigm KB : Knowledge Base KIS : Korea Industry Standards KFD : Kinetic Family Drawing (Burns & Kaufman, 1972) KSD : Kinetic School Drawing MATLab : MATrix Laboratory (MathWorks, 2016) PCC : Pearson Correlation Coefficient 270

Glossary



p_KFD : Patterned Kinetic Family Drawing (Kim, Han, Kim, & Oh, 2011) PPAT : Person Picking an Apple from a Tree (Gantt, 1990) QWK : Quadratic Weighted Kappa RCC : Rank (Spearman) Correlation Coefficient SMC : Structured Mandala Coloring (Curry & Kasser, 2005) SAS : Statistical Analysis System (SAS Institute, 2014) SPSS : Statistical Package for the Social Science (IBM Corporation, 2015)



Nomenclature

κC : Cohen’s kappa value κF : Fleiss’ kappa value κ2 : QWK (Quadratic Weighted Kappa) value R2 : COD (Coefficient Of Determination) Rp : PCC (Pearson Correlation Coefficient) Rs : RCC (Rank (Spearman) Correlation Coefficient)

271

Copyright Permissions

We appreciate the following copyright permissions from Taylor & Francis, Elsevier, and Wiley.

Articles Kim, S. I., Ryu, H. J., Hwang, J. O., & Kim, M. S. H. (2006). An expert system approach to art psychotherapy. The Arts in Psychotherapy, 33(1), 59-75. Kim, S. I., Kim, K. E., Lee, Y., Lee, S. K., & Yoo, S. (2006). How to make a machine think in art psychotherapy: An expert system’s reasoning process. The Arts in Psychotherapy, 33(5), 383-394. Kim, S. I., Bae, J., & Lee, Y. (2007). A computer system to rate the color-related formal elements in art therapy assessments. The Arts in Psychotherapy, 34(3), 223-237. Kim, S. I., Kang, H. S., & Kim, K. E. (2008). Computer determination of placement in a drawing in art therapy assessment. The Arts in Psychotherapy, 35(1), 49-59. Kim, S. I. (2008). Computer judgment of main color in a drawing for art psychotherapy assessment. The Arts in Psychotherapy, 35(2), 140-150. Kim, S. I., Kim, Y. H., & Kim, E. J. (2008). An expert system for interpretation of structured mandala. The Arts in Psychotherapy, 35(5), 320-328. Kim, S. I., Kang, H. S., & Kim, Y. H. (2009). A computer system for art therapy assessment of elements in structured mandala. The Arts in Psychotherapy, 36(1), 19-28. Kim, S. I., & Hameed, A. I. (2009). A computer system to rate the variety of color in drawings. Art Therapy: Journal of the American Art Therapy Association, 26(2), 73-79. Kim, S. I., Betts, D. J., Kim, H. M., & Kang, H. S. (2009). Statistical models to estimate level of psychological disorder based on a computer rating 272

Copyright Permissions

system: An application to dementia using structured mandala drawings. The Arts in Psychotherapy, 36(4), 214-221. Kim, S. I. (2010). A computer system for the analysis of color-related elements in art therapy assessment: Computer_Color-Related Elements Art Therapy Evaluation System (C_CREATES). The Arts in Psychotherapy, 37(5), 378-386. Kim, S. I., Han, J., Kim, Y. H., & Oh, Y. J. (2011). A computer art therapy system for the Kinetic Family Drawing (CATS_KFD). The Arts in Psychotherapy, 38(1), 17-28. Kim, S. I., Han, J., Oh, Y. J., & Kim, Y. H. (2012). A computer art therapy system for evaluation of space usage in drawings with application to the analysis of its relationship to level of dementia. New Ideas in Psychology, 30(3), 300-307. Kim, S. I., Kang, H. S., Chung, S., & Hong, E. J. (2012). A statistical approach to comparing the effectiveness of several art therapy tools in estimating the level of a psychological state. The Arts in Psychotherapy, 39(5), 397-403. Kim, S. I., Kim, J. H., & Hong, E. J. (2013). A computer system for the face stimulus assessment with application to the analysis of dementia. The Arts in Psychotherapy, 40(2), 245-249. Kim, S. l, Ghil, J. H., Choi, E. Y., Kwon, O. S., & Kong, M. (2014). A computer system using a structured mandala to differentiate and identify psychological disorders. The Arts in Psychotherapy, 41(2), 181-186.

Book chapter Kim, S. I. (2016). Assessments and computer technology. In D. Gussak & M. Rosal (Eds.), The Wiley handbook of art therapy (pp. 587-599). Hoboken, NJ: Wiley-Blackwell.

273

Bibliography

Abraham, A. (1989). Hgalui vehasamui bziurai dmut enush (The exposed and secret of A human figure drawings). Tel Avivi, Israel: Reshafim, Adler, P. T. (1970). Evaluation of the figure drawing technique: Reliability, factorial structure, and diagnostic usefulness. Journal of Consulting and Clinical Psychology, 35(1), 52-57. Adler, P. T. (1971). Ethnic and socioeconomic status differences in human figure drawings. Journal of Consulting and Clinical Psychology, 36(3), 344-354. Adobe Creative Team. (2004). Adobe Photoshop 7.0 classroom in a book. San Diego, CA: Adobe Press. Alschuler, R. H., & Hattwick, L. W. (1947). Painting and personality: A study of young children. Chicago, IL: University of Chicago Press. Altman, D. G. (1991). Practical statistics for medical research. Boca Raton, FL: Chapman & Hall. American Psychiatric Association. (2002). Diagnostic and statistical manual of mental disorders (4th ed.). Washington, D.C.: American Psychiatric Association. Asawa, P. (2009). Art therapists’ emotional reactions to the demands of technology. Art therapy: Journal of the American Art Therapy Association, 26(2), 58-65. Austin, B. D. (2009). Renewing the debate: Digital technology in art therapy and the creative process. Art therapy: Journal of the American Art Therapy Association, 26(2), 83-85. Babor, T. F., Stenius, K., Savva, S., & O’reilly, J. (2009). Publishing addiction science: A B guide for the perplexed (2nd ed.). Brentwood, United Kingdom: Multi-Science Publishing Company Ltd. Baraev, V. V. (1991). Decembrists and Kandinsky family. Moscow, Russia: Politizdat. Bartkowiak, M., & Domanski, M. (1995). Vector median filters for processing of color images in various color spaces. Proceedings of IEEE Conference on Image Processing and Its Applications, 833-836. Beck, A. T., Ward, C. H., Mendelson, M., Mock, J., & Erbaugh, J. (1961). An inventory for measuring depression. Archives of General Psychiatry, 4(6), 561-571. Belkofer, C. M. (2011). Understanding social media culture and its ethical challengers for art therapists. Art therapy: Journal of the American Art Therapy Association, 28(4), 159-164. Betts, D. J. (2003). Developing projective drawing test: Experience with the Face Stimulus Assessment (FSA). Art Therapy: Journal of the American Art Therapy Association, 20(2), 77-82.

275

276 Bibliography Betts, D. J. (2005). A systematic analysis of art therapy assessment and rating instrument literature. Dissertation. Florida State University, Tallahassee, FL. Published online at http://www.art-therapy.us/images/Donna Betts.pdf. Betts, D. J. (2006). Art therapy assessments and rating instruments: Do they measure up? The Arts in Psychotherapy, 33(5), 422-434. Billington, J. (2016). IBM’s Watson cracks medical mystery with life-saving diagnosis for patient who baffled doctors. Retrieved February 17, 2017, from http://www.ibtimes. co.uk/ ... /Innovation. Boley, S., Ammen, S., O’Conner, K., & Miller, L. (1996). The use of the Color-Your-Life technique with pediatric cancer patients and their siblings. International Journal of Play Therapy, 5(2), 57-78. Bonny, H. L., & Kellogg, J. (1977). Mandalas as a measure of change in psychotherapy: Mandalas used in conjunction with music therapy. American Journal of Art Therapy, 16(4), 126-128. Breaem, H. (1986). Die macht der Farben. München, Germany: Wirtschaftsverlagen Langen Müller. Brooke, S. (2004). Tools of the trade: A therapist’s guide to art therapy assessments (2nd ed.). Springfield, IL: Charles C Thomas. Buck, J. (1964). The H-T-P manual supplement. Beverly Hills, CA: Western Psychological Services. Buck, J. (1966). The House-Tree-Person technique: Revised manual. Los Angeles, CA: Western Psychological Services. Buck, J. N. (1948). The H-T-P test. Journal of Clinical Psychology, 4(2), 151-159. Buck, J. N. (1949). The H-T-P test. Journal of Clinical Psychology, 5(1), 37-74. Buck, J. N. (1987). The House-Tree-Person technique: Revised manual. Los Angeles, CA: Western Psychological Services. Burkitt, E., Barrett, M., & Davis, A. (2009). Effects of different emotion terms on the size and colour of children’s drawings. International Journal of Art Therapy, 14(2), 74-84. Burns, A., Purandare, N., & Craig, S. (2002). Mental health in older people. London, United Kingdom: Royal Society of Medicine Press. Burns, R. C. (1987). Self-growth in families: Kinetic Family Drawings (K-F-D) research and application. New York: Brunner/Mazel. Burns, R. C. (1982). Self-growth in families. New York: Brunner/Mazel. Burns, R. C., & Kaufman, S. H. (1972). Actions, styles, and symbols in Kinetic Family Drawings (K-F-D): An interpretive manual. New York: Brunner/Mazel. Burnside, E. S. (2005). Bayesian networks: Computer-assisted diagnosis support in radiology. Academic Radiology, 12(4), 422-430. C Carroll, M. K., & Ryan-Wenger, N. A. (1999). School-age children fears, anxiety and human figure drawings. Journal of Pediatric Health Care, 13(1), 24-31. Chang, P.-C., Wang, Y.-W., & Liu, C.-H. (2007). The development of a weighted evolving fuzzy neural network for PCB sales forecasting. Expert Systems with Applications, 32(1), 86-96. Chantler, L., Pelco, I., & Mertin, P. (1993). The psychological evaluation of child sexual abuse using the Louisville behavior checklist and human figures drawing. Child Abuse & Neglect, 17(2), 271-179. Chapman, L. J., & Chapman, J. P. (1967). Genesis of popular but erroneous psychodiagnostic observations. Journal of Abnormal Psychology, 72(3), 193-204. Charniak, E. (1991). Bayesian networks without tears: Making Bayesian networks more accessible to the probabilistically unsophisticated. AI Magazine, 12(4), 50-63. Chilton, G., L., Gerity, L., LaVorgna-Smith, M., & MacMichael, H. N. (2009). An online

Bibliography

277

art exchange group: 14 secrets for a happy artist’s life. Art Therapy: Journal of the American Art Therapy Association, 26(2), 66-72. Cicchetti, D. V., & Sparrow, S. S. (1982). Developing criteria for establishing the inter-rater reliability of specific items in a given inventory. American Journal of Mental Deficiency, 86(2), 127-137. Clercq, P. A., Hasman, A., Blom, J. A., & Korsten, H. H. M. (2001). Design and implementation of a framework to support the development of clinical guidelines. International Journal of Medical Information, 64(2), 285-318. Cohen-Liebman, M. S. (1994). The art therapist as expert witness in child sexual abuse litigation. Art Therapy: Journal of the American Art Therapy Association, 11(4), 260-265. Cohen, B. M. (Ed.) (1986/1994). The diagnostic drawing series rating guide. Unpublished guidebook. Cohen, B. M., & Mills, A. (1994). Writing up a diagnostic drawing series. Unpublished guidebook. Cohen, B. M., Hammer, J. S., & Singer, S. (1988). The diagnostic drawing series: A systematic approach to art therapy evaluation and research. The Arts in Psychotherapy, 15(1), 11-21. Cohen, F. W., & Phelps, R. E. (1985). Incest markers in children’s artwork. The Arts in Psychotherapy, 12(4), 265-283. Corcho, O., Fernandez-Lopez, M., & Gomez-Perez, A. (2003). Methodologies, tools and languages for building ontologies. Where is their meeting point? Data and Knowledge Engineering, 46(1), 41-64. Cornell, J. (1994). Mandala: Luminous symbols for healing. Wheaton, IL: Quest Books Wheaton. Coseo, A. (1997). Developing cultural awareness for creative arts therapists. The Arts in Psychotherapy, 24(2), 145-157. Couch, J. B. (1997). Behind the veil: Mandala drawings by dementia patients. Art Therapy: Journal of the American Art Therapy Association, 14(3), 187-193. Cowell, R. G., Dawid, A. P., Hutchinson, T., & Spiegelhalter, D. J. (1991). A Bayesian expert system for the analysis of an adverse drug reaction. Artificial Intelligence in Medicine, 3(5), 257-270. Cox, C. T., & Cohen, B. M. (2000). Mandala artwork by clients with DID: Clinical observations based on two theoretical models. Art Therapy: Journal of the American Art Therapy Association, 17(3), 195-201. Cronin, S. M., & Werblowsky, J. H. (1979). Early signs of organicity in art work. Art Psychotherapy, 6(2), 103-108. Cruz-Ramirez, N., Acosta-Mesa, H. G., Carrillo-Calvet, H., Nava-Fernandez, L. A., & Barrientos-Martinez, R. E. (2007). Diagnosis of breast cancer using Bayesian networks: A case study. Computers in Biology and Medicine, 37(11), 1553-1564. Cummings, J. A. (1980). An evaluation of an objective scoring system for the KFDs. Dissertation Abstracts, 41(6-A), 2313. Curry, N. A., & Kasser, T. (2005). Can coloring mandalas reduce anxiety? Art Therapy: Journal of the American Art Therapy Association, 22(2), 81-85. Dawson, C. F. S. (1984). A study of selected style and content variables in the drawings D of depressed and nondepressed adults. Unpublished dissertation, University of North Dakota, Grand Forks, ND. de Clercq, P. A., Hasman, A., Blom, J. A., & Korsten, H. H. M. (2001). Design and implementation of a framework to support the development of clinical guidelines. International Journal of Medical Information, 64(2-3), 285-318. de Jong, E. (1993). Flying colors. http://www.trademakia.com/flying- colors-74425988. html.

278 Bibliography Deaver, S. P. (2002). What constitutes art therapy research? Art therapy: Journal of the American Art Therapy Association, 19(1), 23-27. Derogatis, L. R. (1977). SCL-90-R: Administration, scoring and procedures manual-I for the revised version. Baltimore, MD: Johns Hopkins School of Medicine. Derogatis, L. R. (1992). SCL-90-R: Administration, scoring of procedures manual-II for the revised version and other instruments of the psychopathology rating scale series. Towson, MD: Clinical Psychometric Research. Devi, G., Ottman, R., Tang, M., Marder, K., Stern, Y., & Mayeux, R. (2000). Familial aggregation of Alzheimer disease among Whites, African Amer- icans, and Caribbean Hispanics in northern Manhattan. Archives of Neurology, 57(1), 72-77. DiLeo, F., & Kellogg, J. (1977). Mandalas as a measure of change in Psychotherapy: Mandalas used in conjunction with psychedelic drugs. American Journal of Art Therapy, 16, 128-130. DiLeo, J. (1970). Young children and their drawings. New York: Brunner/Mazel. Draper, B., Peisah, C., Snowdon, J., & Brodaty, H. (2010). Early dementia diagnosis and the risk of suicide and euthanasia. Alzheimer’s & Dementia, 6(1), 75-82. E Elin, N., & Nucho, A. O. (1979). The use of Kinetic Family Drawing as a diagnostic tool in assessing the child’s self-concept. The Arts in Psychotherapy, 6(4), 241-247. F Fincher, S. F. (1991). Creating mandalas for insight, healing, and self-expression. Boston, MA: Shambhala Publications, Inc. Finger, D. R. (1997). Child case study Alan, before and during therapy. In E. H. Hammer (Ed), Advances in projective drawing interpretation (pp. 263-267). Springfield, IL: Charles C Thomas. Fleiss, J. L. (1971). Measuring nominal scale agreement among many raters. Psychological Bulletin, 76(5), 378-381. Fletcher, P. C., & Frith, C. D. (2009). Perceiving is believing: a Bayesian approach to explaining the positive symptoms of schizophrenia. Nature Reviews Neuroscience, 10(1), 48-58. Folstein, M. F., Folstein, S. E., & McHugh, P. R. (1975). Mini-mental state: A practical method for grading the cognitive state of patients for the clinician. Journal of Psychiatric Research, 12(3), 189-198. Forrest, M., & Thomas, G. V. (1991). An exploratory study of drawings by bereaved children, British Journal of Clinical Psychology, 30(4), 373-374. Fowler, J. P., & Ardon, A. M. (2002). Diagnostic Drawing Series and disassociate disorders: A Dutch study. The Arts in Psychotherapy, 29(4), 221-230. Francis, D., Kaiser, D., & Deaver, S. P. (2003). Representation of attachment security in the Bird’s Nest Drawings of clients with substance abuse disorders. Art Therapy: Journal of the American Art Therapy Association, 20(3), 125-137. Fryrear, J. L., & Corbit, I. E. (1992). Photo art therapy. A Jungian perspective. Springfield, IL: Charles C Thomas. Furth, G. M. (1988). The secret world of drawings: Healing through art. Boston, MA: Sigo Press. G Gantt, L. M. (1990). A validity study of the Formal Elements Art Therapy Scale (FEATS) for diagnostic information in patients’ drawings. Unpublished dissertation, University of Pittsburgh, Pittsburgh, PA. Gantt, L. M. (1998). A discussion of art therapy as a science. Art Therapy: Journal of the American Art Therapy Association, 15(1), 3-12. Gantt, L. M. (2004). The case for formal art therapy assessments. Art Therapy: Journal of the American Art Therapy Association, 21(1), 18-29.

Bibliography

279

Gantt, L. M., & Tabone, C. (1998). The formal elements art therapy Scale: The rating manual. Morgantown, WV: Gargoyle Press. Gantt, L. M., & Tabone, C. (2001). Measuring clinical changes using art. Paper presented at the meeting of the American Art Therapy Association, Albuquerque, NM. Gantt, L. M., & Tabone, C. (2003). The formal elements art therapy scale and “Draw a Person Picking an Apple from a Tree”. In C. A. Malchiodi (Ed.), Handbook of art therapy (pp. 420-427). New York: The Guilford Press. Garai, J. E. (1976). New vistas in the exploration of inner and outer space through art therapy. The Arts in Psychotherapy, 3(3-4), 157-167. Gennari, J. H., Musen, M. A., Fergerson, R. W., Grosso, W. E., Crubezy, M., Eriksson, H., Noy, N. F., & Tu, S. W. (2003). The evolution of Protégé: an environment for knowledge-based systems development. International Journal of Human-Computer Studies, 58(1), 89-123. Ghaffurian, M. A. (1995). Visual arts and healing. Geelong, Victoria, Australia: Deakin University. Ghil, J. H., & Kim, S. I. (2010). Probabilistic interpretation of arts using the Bayesian network of artificial intelligence technique. Paper to be presented at the 41st Annual American Art Therapy Association Conference, Sacramento, California Giarratano, J., & Riley, G. (2005). Expert systems, principles and programming (5th ed.). Boston, MA: Course Technology. Golomb, C. (1999). Art and the young: The many faces of representation. Visual Arts Research, 25(1), 27-50. Gonzalez, R. C., & Woods, R. E. (2002). Digital image processing (2nd ed.). Englewood Cliff, NJ: Prentice Hall. Goodenough, F. L. (1926). Measurement of intelligence by drawing. New York: Yonkers on Hudson, World Book. Goodreads, Inc. (2010, November 10). Quotes, Lord Kelvin. Retrieved November 15, 2010, from. http://www.goodreads.com/quotes/show/166961. Gordon, J., & Shortliffe, E. H. (1984). The Dempster-Shafer theory of evidence. In B. G. Buchanan & E. H. Shortliffe (Eds.), Rule-based expert systems. Reading, MA: Addison-Wesley Publishing Company. Gregorian, V. S., Azarian, A., DeMaria, M. B., & McDonald, L. D. (1996), Colors of disaster: The Psychology of the “black sun.” The Arts in Psychotherapy, 23(1), 1-14. Groth-Marnat, G. (1990). Handbook of psychological assessment (2nd ed.). New York: Wiley. Gulbro-Leavit, C., & Schimmel, B. (1991). Assess depression in children and adolescents using the Diagnostic Drawing series modified for children (DDS_C). The Arts in Psychotherapy, 18(4), 353-356. Gunter, M. (2000). Art therapy as an intervention to stabilize the defenses of children undergoing bone marrow transplantation. The Art in Psychotherapy, 27(1), 3-14. Gussak, D. E., & Nyce, J. M. (1999). To bridge art therapy and computer technology: The visual toolbox. Art Therapy: Journal of the American Art Therapy Association, 16(4), 194-196. Hackbarth, S. G., Murphy, H. D., & McQuary, J. P. (1991). Identifying sexually abused H children by using Kinetic Family Drawings. Elementary School Guidance & Counseling, 25(4), 225-260. Hacking, S. (1999). The psychopathology of everyday art: A quantitative study. Dissertation. University of Keele, Sheffield, United Kingdom. Published online at http://www. masictherapyworld.de/modules/archive/stuff/papers/Hacking.pdf.

280 Bibliography Hacking, S., Foreman, D., & Belcher, J. (1996). The descriptive assessment for psychiatric art: A new way of quantifying paintings by psychiatric patients. Journal of Nervous and Mental Disease, 184(7), 425-430. Hamilton, M. K. (2008). Developing a standardized rating system for the Face Stimulus Assessment (FSA) using nine scales adapted from the Formal Elements Art Therapy Scale (FEATS). Unpublished master’s thesis, Avila University, Kansas City, MO. Hammer, E. F. (1953). The role of the HTP in the prognostic battery. Journal of Clinical Psychology, 9(4), 371-374. Hammer, E. F. (1969). Hierarchical organization of personality and the HTP achromatic and chromatic. In J. N. Buck & E. F. Hammer (Eds.), Advances in the House-Tree-Person technique: Variations and applications. Los Angeles, CA: Western Psychological Services. Hammer, E. F., & Piotrowski, Z. A. (1977). Advances in projective drawing. Springfield, IL: Charles C Thomas. Han, S. K. (2008). Municipal yearbook of Korea. Seoul, Korea: Ministry of Public Administration and Security. Hanes, M. J. (1997). Utilizing the circus phenomenon as a drawing theme in art therapy. The Arts in Psychotherapy, 24(4), 375-384. Hartwich, P., & Brandecker, R. (1997). Computer-based art therapy with inpatients: Acute and chronic schizophrenics and borderline cases. The Arts in Psychotherapy, 24(4), 367-375. Haykin, S. (2008). Neural networks and learning machines (3rd ed.). Upper Saddle River, NJ: Prentice Hall. Hibbard, R. A., Roghmann, K., & Hockelman, R. A. (1987). Genitals in children’s drawings: An association with sexual abuse. Pediatrics, 79(1), 129-137. Hollins, S., Horrocks, C., & Sinason, V. (1998). I can get through it. London, United Kingdom: Gaskell/St. George Hospital Medical School. Holt, E. S., & Kaiser, D. H. (2001). Indicators of familial alcoholism in children’s Kinetic Family Drawings. Art Therapy: Journal of the American Art Therapy Association, 18(2), 89-95. Hoshino, J., Silbert, R., Knapp, N., & Weaver, K. (1998). A comparative analysis of pre-and post-electroconvulsive therapy drawings. The Arts in Psychotherapy, 25(3), 189-194. I Im, Y. H., Oh, S. G., Yu, J. H., Lee, H. S., Chang, J. K., & Park, D. H. (2010). A KFD web database system with an object-based image retrieval for family art therapy assessments. The Arts in Psychotherapy, 37(3), 163-171. Ireland, M. S., & Brekke, J. (1980). The mandala in group psychotherapy: Personal identity and intimacy. The Arts in Psychotherapy, 7(3), 217-231. Itten, J. (1961). Kunst der Farbe. Ravensburg, Germany: Otto Maier Verlag. J Jacobi, J. (1979). The psychology of C. G. Jung: An introduction with illustrations. New Haven, CT: Yale University Press. Jay, J. (1999). Art therapy lecture notes. London, United Kingdom: University of Western Ontario. Joiner, T. E., Schmidt, K .L., & Barnett, J. (1996). Size, detail and line heaviness in children’s drawings as correlates of emotional distress: (more) negative evidence. Journal of Personality Assessment, 67(1), 127-141. Jones, F., Warren, A., & McElroy, S. (2006). Home-based art therapy for older adults with mental health needs: Views of clients and caregivers. Art Therapy: Journal of American Art Therapy Association, 23(2), 52-58. Jung, C. G. (1958). Psyche and symbol. Violet de Laszlo (Ed.), New York: Doubleday.

Bibliography

281

Jung, C. G. (1973). Mandala symbolism (3rd printing) (F. F. C. Hull, trans.) Bollingen Series. Princeton, NJ: Princeton University Press. Jung, C. G. (1989). Memories, dreams, and reflections (A. Jaffe, Ed.; R. Winston & C. F. Baynes, Trans.). New York: Vintage Books. Kahill, S. (1984). Human figure drawing in adults: An update of the empirical evidence, K 1967-1982. Canadian Psychology, 25(4), 269-292. Kaiser, D. H., & Deaver, S. (2009). Assessing attachment with the Bird’s Nest Drawing: A review of the research. Art Therapy: Journal of the American Art Therapy Association, 26(1), 26-33. Kaiser, D. H., St. John, P., & Ball, B. (2006). Teaching art therapy research: A brief report. Art Therapy: Journal of the American Art Therapy Association, 23(4), 186-190. Kannan, A., Mohan, V., & Anbazhagan, N. (2010). An effective method of image retrieval using image mining techniques. The International Journal of Multimedia & Its Application (IJMA), 2(4), 17-26. Kapitan, L. (2007). Will art therapy cross the digital culture divide? Art Therapy: Journal of the American Art Therapy Association, 24(2), 50-51. Kapitan, L. (2009). Introduction to the special issue on art therapy’s response to techno-digital culture. Art therapy: Journal of the American Art Therapy Association, 26(2), 50-51. Kapitan, L. (2010). Introduction to art therapy research. New York: Routledge. Kaplan, F. F. (1998). Scientific art therapy: An integrative and research-based approach. Art Therapy: Journal of the American Art Therapy Association, 15(2), 93-98. Kaplan, F. F. (2000). Art, science and art therapy: Repainting the picture. London, United Kingdom: Jessica Kingsley. Kaplan, F. F. (2003). Art-based assessments. In C. A. Malchiodi (Ed.), Handbook of art therapy (pp. 25-35). New York: The Guilford Press. Karp, A., Paillard-Borg, S. P., Wang, H. X., Silverstein, M., Winblad, B., & Fratiglioni, L. (2006). Mental, physical, and social components in leisure activities equally contribute to decrease dementia risk. Dementia and geriatric cognitive disorders, 21(2), 65-73. Kaya, N., & Epps, H. H. (2004). Relationship between color and emotion: a study of college students. College Student Journal, 38(3), 396-405. Kellogg, J. (1977). The meaning of color and shape in mandalas. American Journal of Art Therapy, 16(4), 123-126. Kent, K. E. G. (1999). Relationship between the Draw-A-Person questionnaire and the Rorchach in the measurement of psychopathology. Dissertation Abstracts International: Section B: The Science and Engineering, 60(1-B), 368.PD. Kim, J. E. (1988). Psychological diagnosis of children by drawings (in Korean). Seoul, Korea: Science Education Co. Kim, J. E., & Lim, H. J. (1972). Relationships between color and personality in children’s drawing (in Korean). Journal of Korean Culture, 20, 285-297. Kim, S. I. (2008a). Computer judgment of main color in a drawing for art psychotherapy assessment. The Arts in Psychotherapy, 35(2), 140-150. Kim, S. I. (2008b). Commentaries [To the editor]. Art Therapy: Journal of the American Art Therapy Association, 25(1), 41. Kim, S. I. (2010). A computer system for the analysis of color-related elements in art therapy assessment: Computer_Color-Related Elements Art Therapy Evaluation System (C_CREATES). The Arts in Psychotherapy, 37(5), 378-386. Kim, S. I. (2016). Assessments and computer technology. In D. Gussak & M. Rosal (Eds.), The Wiley handbook of art therapy (pp. 587-599). Hoboken, NJ: Wiley-Blackwell. Kim, S. I., Bae, J., & Lee, Y. (2007). A computer system to rate the color-related formal elements in art therapy assessments. The Arts in Psychotherapy, 34(3), 223-237.

282 Bibliography Kim, S. I., Betts, D. J., Kim, H. M., & Kang, H. S. (2009). Statistical models to estimate level of psychological disorder based on a computer rating system: An application to dementia using structured mandala drawings. The Arts in Psychotherapy, 36(4), 214221. Kim, S. I., Ghil, J. H., Choi, E. Y., Kwon, O. S., & Kong, M. (2014). A computer system using a structured mandala to differentiate and identify psychological disorders. The Arts in Psychotherapy, 41(2), 181-186. Kim, S. I., & Hameed, I. A. (2009). A computer system to rate the variety of color in drawing. Art Therapy: Journal of the American Art Therapy Association, 26(2), 73-79. Kim, S. I., Han, J., Kim, Y. H., & Oh, Y. J. (2011). A Computer Art Therapy System for Kinetic Family Drawing (CATS_KFD). The Arts in Psychotherapy, 38(1), 17-28. Kim, S. I., Han. J., & Oh, Y. J. (2012). A computer art assessment system for the evaluation of space usage in drawings with application to the analysis of its relationship to level of dementia. New Ideas in Psychology, 30(5). 300-307. Kim, S. I., Kang, H. S., Chung, S., & Hong, E. J. (2012). A statistical approach to comparing the effectiveness of several art therapy tools in estimating the level of a psychological state. The Arts in Psychotherapy, 39(5), 397-403. Kim, S. I., Kang, H. S., & Kim, K. E. (2008). Computer determination of placement in a drawing in art therapy assessment. The Arts in Psychotherapy, 35(1), 49-59. Kim, S. I., Kang, H. S., & Kim, Y. H. (2009). A computer system for art therapy assessment of elements in structured mandala. The Arts in Psychotherapy, 36(1), 19-28. Kim, S. I., Kim, J. H., & Hong, E. J. (2013). A computer system for the face stimulus assessment with application to the analysis of dementia. The Arts in Psychotherapy, 40(2), 245-249. Kim, S. I., Kim, Y. H., & Kim, E. J. (2008). An expert system for interpretation of structured mandala. The Arts in Psychotherapy, 35(5), 320-328. Kim, S. I., Kim, Y. H., Lee, C. W., Kim, S. K., & Baik, D. K. (1992). An expert system to facilitate the uniform administration of justice in criminal cases. Expert Systems with Applications, 5(1), 103-110. Kim, S. I., Kim, K. E., Lee, Y., Lee, S. K., & Yoo, S. (2006). How to make a machine think in art psychotherapy: An expert system’s reasoning process. The Arts in Psychotherapy, 33(5), 383-394. Kim, S. I., Ryu, H. J., Hwang, J. O., & Kim, M. S. H. (2006). An expert system approach to art psychotherapy. The Arts in Psychotherapy, 33(1), 59-75. Kim, S. I., Yoo, S., Kim, K. E., & Lee, Y. (2007). A framework for expert system knowledge base in art psychotherapy. In A. R. Tyler (Ed.), Expert systems research trends (pp. 181-206). Haupauge, NY: Nova Science Publishers. King, N., Gullon, E., & Ollendick, T. H. (1992). Manifest anxiety and fearfulness in children and adolescents. Journal of Genetic Psychology, 153 (1), 63-73. Klepsch, M., & Logie, L. (1982). Children draw and tell: An introduction to the projective uses of children’s human figure drawings. New, York: Brunner/Mazel. Klopfer, W. G., & Taulbee, E. S. (1976). Projective tests. Annual Review of Psychology, 27(54), 3-567. Knapp, N. M. (1994). Research with Diagnostic Drawings for normal and Alzheimer’s subjects. Art Therapy: Journal of the American Art Therapy Association, 11(2), 131138. Koppitz, E. M. (1968). Psychological evaluation of children's human drawings. New York: Crune and Stratton. Kopytin A. (2002). The Silver drawing test of cognition and emotion: Standardization in Russia. American Journal of Art Therapy, 40(4), 223-237. Korb, K. B., & Nicholson, A. E. (2004). Bayesian artificial intelligence. London, United Kingdom: Chapman & Hall/CRC.

Bibliography

283

Korean Art Therapy Association. (2013). Code of ethics. Retrieved from http://www.korean-arttherapy.or.kr/ Kreitler, H., & Kreitler, S. (1980). Psychology of the arts. Tel Aviv, Israel: Poalim (Hebrew). Kutner, M. H., Nachtsheim, C. J., Neter, N., & Li, W. (2005). Applied linear statistical models (5th ed.). New York: McGraw-Hill. Kwon, Y. C., & Park, J. H. (1989). Korean version of mini-mental state examination (in Korean). Journal of Korean Neuropsychiatry Association, 28(1), 125-135. Leavitt, C. G., & Schimmel, B. (1991). Assessing depression in children and adolescents L using the diagnostic drawing Series modified for children (DDS-C). The Arts in Psychotherapy, 18(1), 353-356. Lehmann, H., & Risquez, F. (1953). The use of finger paintings in the clinical evaluation of psychotic conditions: A quantitative and qualitative approach. Journal of Mental Science, 99(417), 763-777. Letenneur, L., Gilleron, V., Commenges, D., Helmer, C., Orgogozo, J. M., & Dartigues, J. F. (1999). Are sex and educational level independent predictors of dementia and Alzheimer’s disease? Incidence data from the PAQUID project. Journal of Neurology, Neurosurgery & Psychiatry, 66(2), 177-183. Lev-Wiesel, R., & Daphna-Tekoha, S. (2000). The self-revelation through color technique: Understanding clients’ relations with significant others, silent language, and defense mechanism through the use of color. American Journal of Art Therapy, 39(2), 35-41. Lev-Wiesel, R., & Drori, D. (2000).The effects of social status upon the self-concept of elderly widows and wives assessed by human figure drawings. The Arts in Psychotherapy, 27(4), 263-267. Lev-Wiesel, R., and Shvero, T. (2003). An exploratory study of self-figure drawings of individuals diagnosed with schizophrenia. The Arts in Psychotherapy, 30(1), 13-16. Levick, M. F. (2001). The Levick Emotional and Cognitive Art Therapy Assessment (LECATA) (rev. ed.). Boca Raton, FL: South Florida Art Psychotherapy Institute. Lewinsohn, P. M. (1964). Relationship between height of figure drawings and depression in psychotic patients. Journal of Consulting Psychology, 28(4), 380-381. Lim, M. J. (2010, August 25). 10,212 first year students in Gyeonggi-do are doubt to have ADHD. JoongAng Daily, pp. 18-19. Lopata, H. Z. (1976). Current widowhood, myths and realities. Thousand Oaks, CA: Sage. Lopata, H. Z. (1996). Widows as a minority group. In D. Bell (Ed.), Contemporary social gerontology (pp. 348-355). Springfield, IL: Charles C Thomas. Lorge, I., Tuckman, I., & Dunn, M. B. (1958). Human figure drawings by younger and older adults. Journal of Clinical Psychology, 14(1), 54-56. Louw, A. E., & Ramkisson, S. (2002). The suitability of the Roberts Apperception Test for Children (RATC), The House-Tree-Person (H-T-P) and Draw-a-Person (D-A-P) scales in the identification of child sexual abuse in the Indian community: An exploratory study. South African Journal of Child and Adolescent Mental Health, 14(2), 96-106. Lowenfeld, V., & Brittain, W. L. (1982). Creative and mental growth (7th ed.). New York: Macmillan. Luscher, M. (1971). The Luscher color test. Hamburg, Germany: Rowohlt. Machover, K. A. (1949). Personality projection in the drawing of the human figure. Spring- M field, IL: Charles C Thomas. Malchiodi, C. A. (1998). Understanding children’s drawings. New York: Guilford Press. Malchiodi, C. A. (1998). The art therapy sourcebook. Lincolnwood, IL: Lowell House. Malchiodi, C. A. (2012). Art therapy: There’s an App for that. Psychology Today.

284 Bibliography Malchiodi, C. A. (1990). Breaking the silence: Art therapy with children from violent homes. New York: Brunner/Mazel. Malchiodi, C. A. (1998). The art therapy sourcebook. Lincolnwood, IL: Lowell House. Malchiodi, C. A. (1998). Understanding children’s drawings. New York: Guilford Press. Malchiodi, C. (1999). Computer art therapy: A virtual studio of possibilities. London, United Kingdom: Jessica Kingsley. Malchiodi, C. A. (2000). Art therapy & computer technology: A virtual studio of possibilities. London, United Kingdom: Jessica Kingsley. Manning, T. M. (1987). Aggression depicted in abused children’s drawings. The Arts in Psychotherapy, 14(1), 15-24. Martin, J., & Oxman, S. (1988). Building expert systems. Englewood Cliffs, NJ: Prentice-Hall International, Inc. Mattson, D. C. (2009). Accessible image analysis for art assessment. The Arts in Psychotherapy, 36(4), 208-213. Mattson, D. C. (2011). Standardizing the Formal Elements Art Therapy Scale (FEATS) rotation scale with computerized technology: A pilot study. The Arts in Psychotherapy, 38(2), 120-124. Mattson, D. C. (2012a). Constructing the computer-rated Face Stimulus Assessment-Revised (FSA-R) to assess formal elements of Major Depressive Disorder (MDD). The Arts in Psychotherapy, 39(1), 31-37. Mattson, D. C. (2012b). An introduction to the computerized assessment of art-based instruments. Art Therapy: Journal of the American Art Therapy Association, 29(1), 2732. McDermott, J., & Bachant, J. (1984). R1 revisited: Four years in the trenches. AI Magazine, 5(3), 21-32. McGraw, K. L., & Seale, M. R. (1988). Knowledge elicitation with multiple experts: Considerations and techniques. Artificial Intelligence Review, 2(1), 31-44. McLeod, C. (1999). Empowering Creative with Computer-assisted Art Therapy: An Introduction to Available Programs and Techniques. Art therapy: Journal of the American Art Therapy Association, 16(4), 201-205. McNiff, S. (1998). Art-based research. Philadelphia, PA: Jessica Kingsley. McNiff, S. (1999). The virtual art therapy studio. Art Therapy: Journal of the American Art Therapy Association, 16(4), 197-200. McNiff, S. (1998). Enlarging the vision of art therapy research. Art Therapy: Journal of the American art therapy Association, 15(2), 86-92. McPhee, J. P., & Wegner, K. W. (1976). Kinetic Family Drawing styles and emotionally disturbed childhood behavior. Journal of Personality Assessment, 40(5), 487-491. Microsoft. (2011). Microsoft Bayesian network editor. Retrieved April 10, from http://research.microsoft.com/en-us/um/redmond/groups/adapt/ms bnx/. Mills, A., Cohen, B. M., & Meneses, J. Z. (1993). Reliability and validity tests of the diagnostic drawing series. The Arts in Psychotherapy, 20(1), 83-88. Milne, L. C., & Greenway, P. (1999). Color in children’s drawings: The influence of age and gender. The Arts in Psychotherapy, 26(4), 261-263. Mitchell, T. M., Shinkareva, S. V., Carlson, A., Chang, K.-M., Malave, V. L., Mason, R. A., & Just, M. A. (2008). Predicting human brain activity associated with the meanings of nouns. Science, 320(5880), 1191-1195. Monahan, M. (1986). Situation influences on children’s Kinetic Family Drawings. Dissertation Abstracts International, 46(12), 4444-B. Morris, J. C. (1993). The Clinical Dementia Rating (CDR): Current version and scoring rules. Neurology, 43(11), 2412-2414. Mostkoff, D. L., & Lazarus, P. J. (1983). The Kinetic Family Drawing: The reliability of an objective scoring system. Psychology in the Schools, 20(1), 16-20.

Bibliography

285

Munley, M. (2002). Comparing the PPAT drawings of boys with AD/HD and age matched controls using the Formal Elements Art Therapy Scale. Art Therapy: Journal of the American Art Therapy Association, 19(2), 66-76. Murphy, E., & Meisgeier, C. (1987). Murphy-Meisgeier type indicator for children. Palo Alto, CA: Consulting Psychologists Press. Myers, D. V. (1978). Toward an objective evaluation procedure of the Kinetic Family Drawings (KFD). Journal of Personality Assessment, 42(4), 358-365. Myers, I. B., & McCaulley, M. (1985). Manual: A guide to the development and use of the Myers-Brigg Type Indicator. Palo Alto, CA: Consulting Psychologists Press. Naglieri, J. A. (1988). Draw a Person: A quantitative scoring system. New York: Psycho- N logical Corporation. Nakanishi, Y. (2002). Children’s mind in their drawings (in Korean). Seoul, Korea: Four Seasons Publishing Co. Neglieri, J. A., McNeish, T. J., & Bardos, A. N. (1991). Draw a Person: Screening procedure for emotional disturbance, Austin, TX: Pro-Ed. Nwana, H. S., Paton, R., Bench-Capon, T. J. M., & Shave, M. J. R. (1991). Facilitating the development of knowledge based systems: A critical review of acquisition tools and techniques. Artificiall Intelligence Communications, 4(2-3), 60-73. Orr, P. P. (2006a). A documentary film project with first-year art therapy students. The Arts O in Psychotherapy, 33(4), 281-287. Orr, P. P. (2006b). Reviews. Art therapy & computer technology: A visual studio of possibilities by Malchiodi, C.A. Art Therapy: Journal of the American Art Therapy Association, 23(3), 145-148. Oster, G. D. & Gould, P. (2004). Using drawings in assessment and therapy: A guide for mental health professionals (2nd ed.). New York: Brunner-Routledge. Parker-Bell, B. (1999). Embracing a future with computers and art therapy. Art therapy: P Journal of the American Art Therapy Association, 16(4), 180-185. Peterson, B. C., Stovall, K., Elkins, D. E., & Parker-Bell, B. (2005). Art therapists and computer technology. Art therapy: Journal of the American Art Therapy Association, 22(3), 139-149. Piaget, J. (1959). Judgment and reasoning in the child. Patterson, NJ: Littlefield, Adams. Pianetti, C., Palacios, M., & Elliott, L. (1964). Significance of color in the drawings of chronic schizophrenics. American Journal of Occupational Therapy, 18(2), 137-140. Pitak-Davis, S. (1992). Book review, Creating mandalas: For insight, healing, and self-expression by S. F. Fincher, 1991. Boston, MA: Shambhala. The Arts in Psychotherapy, 19(3), 223. Precker, J. A. (1950). Paintings and drawing in personality assessment. Journal of Projective Techniques, 14(3), 262-286. Raggad, B. G. (1996). Expert system quality control. Information Processing & Manage- R ment, 32(2), 171-183. Rankin, K. P., Liu, A. A., Howard, S., Slama, H., Hou, C. E., Shuster, K., & Miller, B. L. (2007). A case-controlled study of altered visual art production in Alzheimer’s and FTLD. Cognitive Behavioral Neurology, 20(1), 48-61. Reddy, K. S., Bhadramani, G., & Samiullah, S. (2002). Placement of family members by normal and neglected boys: A study of family drawings. Social Science International, 18(1), 72-82. Reitman, F. (1947). Dynamics of creative activity. Journal of Mental Science, 92, 314-320.

286 Bibliography Reitman, F. (1951). Psychotic art. New York: International Universities Press. Reynolds, C. R. (1978). A quick-scoring guide to the interpretation of children’s Kinetic Family Drawings (KFD). Psychology in the School, 15(4), 489-492. Richards, M., & Ross, H. (1967). Developmental changes in children’s drawings. British Journal of Educational Psychology, 37(1), 73-80. Richter, H. G. (1987). Die Kinderzeichung. Düsseldorf, Germany: Schwann. Roback, H. B. (1968). Human figure drawings: Their utility in the clinical psychologist’s armamentarium for personality assessment. Psychological Bulletin, 70(1), 1-19. Rodgers, J. L., & Nicewander, W. A. (1988). Thirteen ways to look at the correlation coefficient. The American Statistician, 42(1), 59-66. Rorschach, H. (1951). Psychodiagnostics (5th ed.). Bern, Switzerland: Hans Huber. Rorschach, H. (1951). Psychodiagnostik: Psychodiagnostics. New York: Grune and Stratton. Rowe, N.C. (1988). Artificial intelligence through Prolog. Englewood Cliffs, NJ: Prentice-Hall International, Inc. Rubin, J. (1986). From psychopathology to psychotherapy through art expression: A focus on Hans Prinzhorn and others. Art therapy, 3(1), 27-33. Russell-Lacy, S., Robinson, V., Benson, J., & Cranage, J. (1979). An experimental study of pictures produced by acute schizophrenic subjects. British Journal of Psychiatry, 134(2), 195-200. S Salzman, L., & Harway, N. (1967). Size of figure drawings of psychotically depressed patients. Journal of Abnormal Psychology, 72(3), 205-207. Sanders, R. E. (1998). ODBC 3.5 developer’s guide. New York: McGraw- Hill. SAS Institute, Inc. (1999). SAS/STAT user’s guide (version 8). Cary, NC: SAS Publishing. Schornstein, H. M., & Derr, J. (1978). The many applications of Kinetic Family Drawings in child abuse. British Journal of Projective Psychology and Personality Study, 23(1), 33-35. Seoul National University Hospital. (2008). Nationwide study on the prevalence of dementia in Korean elders. Seoul, Korea: Ministry of Health and Welfare. Shin, M. S., Kim, S. K., Kim, J. Y., Park, H. K., Lee, H. R., Jun, S. Y., & Han, S. J. (2002). Diagnosis and understanding of children through their drawings (in Korean). Seoul, Korea: Hak Ji Sa. Shortliffe, E. H. (1972). Computer-based medical consultations: MYCIN. New York: Elsevier. Silver, R. (2002). Three art assessments: the Silver drawing test of cognition and emotion; Draw a story: Screening for depression; and Stimulus drawing techniques. New York: Brunner-Routledge. Slegelis, M. H. (1987). A study of Jung’s mandala and its relationship to art psychotherapy. The Arts in Psychotherapy, 14(4), 301-311. Small, G. W., Moody, T. D., Siddarth, P., & Bookheimer, S. Y. (2009). Your brain on Google: Patterns of cerebral activation during Internet searching. American Journal of Geriatric Psychiatry, 17(2), 116-126. Smith, S. (2015). 5 ways the IBM Watson is changing health care. Retrieved February 17, 2017, from http://www.medicaldaily.com/5-ways-ibm-wea tson. Spiegelhalter, D., Franklin, R., & Bull, K. (1989). Assessment, criticism and improvement of imprecise subjective probabilities for a medical expert system. In proceedings of the Fifth Workshop on Uncertainty in Artificial Intelligence (pp. 335-342). Mountain View, CA: Association for Uncertainty in Artificial Intelligence. Stawar, T. L., & Stawar, D. E. (1987). Family Kinetic Drawings as a screening instrument. Perceptual and Motor Skills, 65(3), 810.

Bibliography

287

Steinhardt, L. (1977). Beyond blue: The implications of blue as the color of the inner surface of the sandtray in sandplay. The Arts in Psychotherapy, 24(5), 455-469. Stewart, E. G. (2006). Kaleidoscope: Color and form illuminate darkness. Chicago, IL: Magnolia Street Publishers. Stewart, E. G. (2004). Art therapy and neuroscience blend: Working with patients who have dementia. Art Therapy: Journal of the American Art Therapy Association, 21(3), 148-155. Suinn, R. M., & Oskamp, S. (1969). The predictive validity of projective measures: A fifteen year evaluative review of research. Springfield, IL: Charles C Thomas. Swensen, C. H. (1957). Empirical evaluations of human figure drawings. Psychological Bulletin, 54(6), 431-466. Swensen, C. H. (1968). Empirical evaluations of human figure drawings: 1957-1966. Psychological Bulletin, 70(1), 20-44. Taurau, P. (1999). Kernel Prolog: A lightweight Prolog-in-Java interpreter with fluent T based built-ins user guide. Denton, TX: BinNet Corp. Taylor, S. A., Kymissis, P., & Pressman, M. (1998). Prospective kinetic family drawing and adolescent mentally ill chemical abusers. The Arts in Psychotherapy, 25(2), 115-124. Thomas, G. V., & Silk, A. M. J. (1990). An introduction to the psychology of children’s drawing. New York: New York University Press. Thomson, W. (1889). Popular lectures and addresses. New York: Macmillan And Co. Thong, S. A. (2007). Redefining the Tools of Art Therapy. Art therapy: Journal of the American Art Therapy Association, 24(2), 52-58. Tibbetts, T. J. (1995). Art therapy at the crossroads: Art and science. Art Therapy: Journal of the American Art Therapy Association, 12(4), 257-260. Tripp, D. (2016). Georgia: A novel of Georgia O’Keeffe. New York: Penguin Random House. Trowbridge, M. M. (1995). Graphic indicators of sexual abuse in children’s drawing: A review of the literature. The Arts in Psychotherapy, 22(5), 485-493. Uusitalo, L. (2007). Advantages and challenges of Bayesian networks in environmental U modeling. Ecological Modelling, 203(3-4), 312-318. Veltman, M. W. M., & Browne, K. D. (2001). Identifying childhood abuse through favorite V kind of day and kinetic family drawings. The Arts in Psychotherapy, 28(4), 251-259. Veltman, M. W. M., & Browne, K.D. (2000a). An evaluation of the favorite kind of day drawings from abused and non-abused children. Child Abuse & Neglect, 24(10), 12491255. Veltman, M. W. M., & Browne, K.D. (2000b). Pictures in the classroom: Can teachers identify abused children’s drawings? Child Abuse Review, 9(5), 328-336. Veltman, M. W. M., & Browne, K. D. (2003). Trained raters’ evaluation of Kinetic Family Drawings of physically abused children. The Arts in Psychotherapy, 30(1), 3-12. Vennet, R. V. D., & Serice, S. (2012). Can coloring mandalas reduce anxiety? A replication study. Art Therapy: Journal of the American Art Therapy Association, 29(2), 87-92. Vick, R. M. (1999). Utilizing prestructured art elements in brief group art therapy with adolescents. Art Therapy: Journal of the American Art Therapy Association, 16(2), 68-77. Vlacour, V. G., Masaki, K. H., Curb, J. D., & Blanchette, P. L. (2000). The detection of dementia in the primary care setting. Archives of Internal Medicine, 160(19), 2964-2968. W Wadeson, H. (1980). Art psychotherapy. New York: John Wiley & Sons. Wadeson, H., & Carpenter, W. (1976). A comparative study of art expression of schizo-

288 Bibliography phrenic, unipolar depressive, and bipolar manic-depressive patients. Journal of Nervous and Mental Disease, 162(5), 334-344. Walpole, R. E., & Myers, R. H. (2006). Probability and statistics for engineers and scientists (8th ed.). New York: Macmillan. Wan, X., & Kuo, C. C. (1998). A new approach to image retrieval with hierarchical color clustering. IEEE Transactions on Circuits, Systems and Video Technologies, 8(5), 628643. Wanderer, Z. W. (1997). Validity of clinical judgments based on human figure drawings. In E. F. Hammer (Ed.), Advances in projective drawing. Springfield, IL: Charles C Thomas. Wang, H., Ericsson, K., Winblad, B., & Fratiglioni, L. (1988). The human figure drawing test as a screen for dementia in the elderly: A community-based study. Archives of Gerontology and Geriatrics, 27(1), 25-34. Weber, P., Medina-Oliva, G., Simon, C., & Jung, B. (2012). Overview on Bayesian networks applications for dependability, risk analysis and maintenance areas. Engineering Applications of Artificial Intelligence, 25(4), 671-682. Wegmann, P., & Vusenbrink, V. B. (2000). Kinetic family drawing scoring method for cross-cultural studies. The Arts in Psychotherapy, 27(3), 179-190. Welbank, M. (1990). An overview of knowledge acquisition methods. Interacting with Computer, 2(1), 83-91. White, C. R., Wallace, J., & Huffman, L. C. (2004). Use of drawings to identify impairment among students with emotional and behavioral disorders: An exploratory study. Art Therapy: Journal of the American Art Therapy Association, 21(4), 210-218. Wilson, R. S., Leon, C. F. M., Barnes, L. L., Schneider, J. A., Bienias, J. L., Evans, D. A., & Bennett, D. A. (2002). Participation in cognitively stimulating activities and risk of incident Alzheimer disease. The Journal of the American Medical Association, 287(6), 742-748. Winnicott. D. W. (1971). Therapeutic consultations in child psychiatry, London, United Kingdom: Hogarth. Witten, I. H., Frank, E., & Hall, M. A. (2011). Data mining: Practical machine learning tools and techniques (3rd ed.). Burlington, MA: Morgan Kaufmann. Wohl, A., & Kaufman, B. (1992). Casualties of childhood: A developmental perspective on sexual abuse using projective drawings. New York: Brunner/Mazel. World Health Organization. (n.d.). Suicide prevention. Retrieved January 30, 2012, from http://www.who.int/mental_health/prevention/suicide/su icideprevent/en/index.html. Y Ye, Q., Gao, W., & Zeng, W. (2003). Color image segmentation using density-based clustering. IEEE International Conference on Acoustics, Speech & Signal Processing, 2, 401-404. Yesavage, J. A., Brink, T. L., Rose, T. L., Lum, O., Huang, V., Adey, M., & Leirer, V. O. (1983). Development and validation of a geriatric depression screening scale: A preliminary report. Journal of Psychiatric Research, 17(1), 37-49.

Index

A AATA, See American Art Therapy Association Accuracy algorithm, 56 concentration, 120 definition, 50, 218 example, 55 ADHD, See Attention Deficit and Hyperactivity Disorder Algorithm accuracy, 56 area colored, 53 area of colored convex hull, 53 definition, 6 example, 6 flow chart, 54 number of colored grids, 53 American Art Therapy Association (AATA) standardization of method, 107, 256 ANalysis Of VAriance (ANOVA) definition, 131 dementia, 124, 132 ANOVA, See ANalysis Of VAriance Applied elements C_CREATES, 12 taxonomy, 13 Area colored example, 53 prominence (rank), 102 space usage, 111 Area of colored convex hull details (grade), 102 details (rank), 103 example, 53 prominence (grade), 101 Area of each color

C_CREATES, 38 Art evaluation difficulties, 18 introduction, 19 Artificial intelligence art therapy, 4 interdisciplinary approach, xiii purpose, xiii techniques, 253 Art interpretation C_CREATES, 16 DDS, 15 difficulties, 17 introduction, 20 PPAT, 15 Artistic skill eight levels, 157 Art therapy artificial intelligence, 4 computer technology, 23 debates, 24 nature, xiv, 3, 169 scientific approach, 23 Attention Deficit and Hyperactivity Disorder (ADHD) example, 150, 167, 201, 247 seriousness, 250 B Basic elements case study I, 40 case study II, 43 C_CREATES, 11 evaluation, 38, 40 example, 38 procedure, 38 taxonomy, 13 Bayesian Network (BN)

289

advantages, 247, 253 causal relationships, 247 definition, 8, 247 example, 9, 248 introduction, 246 Blurring definition, 34 example, 36 BN, See Bayesian Network Book brief organization, xv interdisciplinary study, xiv organization, 18, 22 Part One, 19 Part Two, 20 Built-in functions computer, 4 example, 4 C CAT, See Computational Art Therapy Causal relationships BN, 247 example, 250 Caution interpretation, 138 C_CREATES, See  Computerized ColorRelated Elements Art Therapy Evaluation System c_FSA, See Computerized Face Stimulus Assessment c_SMC, See Computerized Structured Mandala Coloring Changes a series of the SMC drawings, 170 two KFD drawings, 193 Cluster definition, 11

290 Index example, 38 space usage, 115 Clustering definition, 36 example, 36 method, 36 COD, See Coefficient Of Determination Coefficient Of Determination (COD) concentration, 120 definition, 8 details, 102 effectiveness of tools, 242 FSA, 133 level of dementia, 221 model appropriateness, 95 model effectiveness comparison, 238 probability of severe dementia, 225 prominence, 102 space usage (grade), 115 space usage (rank), 115 Color 15 standard, 32 47 standard, 33 art therapy, 29, 30 distance, 32 emotional state, 70 emotions, 30 Color calibration appendix, 46 Color classification definition, 32 example, 33 method, 32 Colored convex hull definition, 52 space, 112 Colored grids definition, 52 space usage, 111, 115 Colored pixels space usage, 111 Color-related elements importance, 59 Color spaces, 30 Complementary color definition, 52 example, 52 Completeness definition, 50, 55, 218 Components addition definition, 128 Computational Art Therapy (CAT) definition, 4 interdisciplinary approach, 25 merit in art evalution, 15 merit in art interpretation, 18

purpose, 257 Computer built-in functions, 4 Computer algorithm, See   Algorithm Computerization c_FSA, 124 c_SMC, 205 p_KFD, 184 Computerized Color-Related Elements Art Therapy Evaluation System (C_CREATES) applied elements, 11, 12 art interpretation, 16 basic elements, 11 case study II, 43 DAPA, 43 evaluation elements, 11 inter-rater reliability, 14 taxonomy, 13 Computerized Face Stimulus Assessment (c_FSA) definition, 123 elements, 126 Computerized Structured Mandala Coloring (c_SMC) elements, 205 Computer system main color, 71 solution, 217 Computer technology skeptical, 139 Concentration 58 SMC samples, 117 COD, 120 definition, 110 estimation, 218 interpretation, 120 pattern, 110 RCC, 121 three SMC samples, 117 validity, 119 Conditional probability, 248 Consistency art interpretation, 138 example, 159 maintaining, 159 Convex hull definition, 11 space usage, 115 Cool color definition, 52 example, 52 D DAPA, See Descriptive Assessment of

Psychiatric Artwork Data mining, 255 DDS, See Diagnostic Drawing Series Degree of concentration, See Concentration Dementia prevalence, 235 Descriptive Assessment of Psychiatric Artwork (DAPA) 6 rating scales, 43 case study II, 43 evaluation elements, 10 Hacking, 43 Details, See Details of drawing Details of drawing COD, 102 definition, 93 samples, 95 stepwise regression (grade), 102 system validity (grade), 104 system validity (rank), 105 Details of objects and environment, See Details of drawing Diagnosis process example, 154 model, 145 Diagnostic Drawing Series (DDS) art interpretation, 15 evaluation elements, 10 inter-rater reliability, 14 system validity, 15 Differentiation of groups Reddy et al., 206 statistically significant elements in SMC, 209 Differention definition, 205 Digital image processing definition, 5 E Edge detection definition, 36 example, 34, 36 Effectiveness comparison regression function, 238 samples, 236 subject of art therapy tools, 235 subject of psychological disorder, 235 Effectiveness of tools COD, 242 RCC, 242 regression model, 233 Elements, See Evaluation elements

Index invaluableness of art, xiii, 1 Evaluation elements Grid C_CREATES, 11 definition, 52 color definition, 48, 50 c_SMC, 208 H DAPA, 10 DDS, 10 HFD, See Human Figure Drawing FEATS, 10 Human Figure Drawing (HFD) FSA, 10 inconsistent interpretation, 17 KFD, 185 interpretation based on size, 109 model effectiveness comparison, 237 system validity, 17 pattern coloring, 55 PPAT, 10 I space colored, 52 subjectivity, 60 Identification Expert system definition, 205 a solution, 138 Identification of group advantages, 146, 167 example, 214 architecture, 7 regression function, 210 artificial intelligence, 153 Rubin, 206 case study, 149, 163 system validity, 212 definition, 6, 138, 153 unsuccessful, 206 development environment, 147 Veltman and Browne, 206 frame, 139 IF - THEN format reasoning process, 157 example, 171 SMC, 177, 183 knowledge expression, 171 software menu, 148 Ill-Structured Paradigm (ISP) solution to ISP problems, 203 art therapy, 3, 246 definition, xiii, 169 F expert system, 169 nature of art therapy, 139 Face recognition Image mining, 255 definition, 128 Individual diagnosis, 145 Face Stimulus Assessment (FSA) Information, 80 COD, 133, 238 Information and statistics evaluation elements, 10, 125 example, 177 Fact Base (FB) Interactive relation action-related FB, 192 definition, 152 changes-related FB, 193 example, 155 color-related FB, 192 Interdisciplinary approach, 255 form-related FB, 192 CAT, 25 performance-related FB, 193 Interdisciplinary study space-related FB, 193 Gantt, 168 Factor analysis Kapitan, 168 statistically significant elements, 208 Kaplan, 168 FB, See Fact Base perspective, xvi FEATS, See Formal Elements Art Interpretation Therapy Scale a series of the SMC drawings, 170 Feedback process, 146 one KFD drawing, 200 Formal Elements Art Therapy Scale p_KFD, 200 (FEATS) two KFD drawings, 193, 201 evaluation elements, 10 Interpretation of coefficients samples, 96 level of dementia, 221 FSA, See Face Stimulus Assessment Interpretation of color brown, 223 G green, 223 Georgia O’Keefe

291

yellow-green, 223 Interpretation (space usage, grade) stepwise regression, 115 Inter-rater reliability C_CREATES, 14 c_FSA, 129 DDS, 14 details (grade), 96 details (rank), 99 placement, 91 PPAT, 14 prominence, 106 prominence (grade), 96 prominence (rank), 99 RCC, 119 space usage, 114 variety of color, 64 ISP, See Ill-Structured Paradigm J Jung mandala, 216 Jungian method critique, 217 K KB, See Knowledge Base KFD, See Kinetic Family Drawing KFD with pattern, See Patterned Kinetic Family Drawing Kinetic Family Drawing (KFD) evaluation elements, 185 questions about validity, 185 validity, 185 Knowledge 10 categories, 161 questionnaire, 175 structure of the art therapy system, 140 Knowledge Base (KB) action-related KB, 197 changes-related KB, 199 color-related KB, 198 form-related KB, 197 performance-related KB, 199 space-related KB, 198 structure, 172 Knowledge expression changes in drawings, 176 colors, 173 example, 171 Knowledge structure example, 141

292 Index L Learning mechanism, 160 Length of edges C_CREATES, 38 placement, 86 variety of colors, 60 Level of dementia classification, 218 COD, 221 estimation, 220 interpretation of coefficients, 221 regression function, 221 score, 218 severe or not, 218 standardized regression, 225 List of colors C_CREATES, 38 Logistic regression definition, 94, 208 Lord Kelvin quantification, 3 scientific thought, xiii M Main color computer algorithm, 74 definition, 70 inter-rater reliability, 76 other useful information, 80 subject and inconsistent, 71 two samples, 73 Mandala dementia patients, 218 effect, 216 Jung, 216 two types, 204 Mini-Mental State Examination-Korean (MMSE-K) estimation, 228 score, 219 Mini-Mental State Examination (MMSE) twelve questions, 219 MMSE, See Mini-Mental State Examination MMSE-K, See Mini-Mental State Examination-Korean Motor skill definition, 126 N Natural color definition, 127

p_KFD, See Patterned Kinetic Family Drawing Placement categories, 85 emotional states, 82 interpretation, 82 judgment, 82 other useful information, 91 sample examples, 87 subjective judgment, 82 two samples, 83 unusual, 81 usual, 81 PPAT, See Person Picking an Apple from a Tree Primary color definition, 52 example, 52 Principal color definition, 50 example, 51 Probabilistic interpretation example, 246 Probability of severe dementia P COD, 225 estimation, 228 Paradigm, xiii regression function, 225 Part One standardized regression, 226 summary, 19 Progress detection Part Two between two drawings, 56 summary, 20 Projective techniques (tools) Pattern coloring challenges, 24 art therapy, 49 contribution, 203 definition, 49 lack of reliability, 203 evaluation elements, 55 questionable validity, 203 Patterned KFD, See Patterned Kinetic usefulness, 203 Family Drawing Prominence, See Prominence of color Patterned Kinetic Family Drawing (p_ Prominence of color KFD) COD, 102 advantages, 202 definition, 93 four stages, 186 samples, 95 information, 187 system validity (grade), 104 subject of computerizing, 185 system validity (rank), 105 PCC, See Pearson Correlation Coefficient Psychological diagnosis, 146 Pearson Correlation Coefficient (PCC) c_FSA and dementia, 132 Q definition, 7 space usage and dementia, 131 Quadratic Weighted Kappa (QWK) Person Picking an Apple from a Tree c_FSA (between human and system), (PPAT) 129 art interpretation, 15 c_FSA (between raters), 129 COD, 238 definition, 8 evaluation elements, 10 dementia, 226, 227 inter-rater reliability, 14 details (between human and system, system validity, 15 grade), 104 Pixel details (between raters, grade), 96 definition, 5 Noise definition, 34 example, 34 Noise removing example, 34 Number of clusters C_CREATES, 38 concentration, 120 details (grade), 102 details (rank), 103 prominence (grade), 101 Number of colored grids example, 53 Number of colors concentration, 120 Number of used colors C_CREATES, 38 details (grade), 102 details (rank), 103 prominence (grade), 101 prominence (rank), 102 variety of colors, 60

Index prominence (between human and system, grade), 104 prominence (between raters, grade), 96 space usage (between human and system, grade), 116 space usage (between raters, grade), 114 Quantification Lord Kelvin, 3 scientific methods, 3 Questionnaire example, 177 QWK, See Quadratic Weighted Kappa R Rank (Spearman) Correlation Coefficient (RCC) concentration, 119, 121 definition, 61 details (rank), 99 effectiveness of tools, 242 prominence (rank), 99 space usage, 114 RCC, See Rank (Spearman) Correlation Coefficient Reasoning process, 142 definition, 155 example, 156 general diagnosis, 156 individual diagnosis, 156 model, 155 order of steps, 156 Regression analysis dementia, 132 Regression function probability of severe dementia, 225 Regression model definition, 94 details, 100 prominence, 100 Reliability seven levels, 159 S Sample c_FSA, 129 SAS, See Statistical Analysis System Secondary color definition, 52 example, 52 Sensitivity analysis BN, 252 Severe dementia

probability, 220 Severity of dementia definition, 131 SMC, See Structured Mandala Coloring Software appendix, 259 Space HFD drawing, 109 Space usage algorithm in c_FSA, 128 c_FSA, 128 COD, 115 DAPA, 109 DDS, 109 definition, 108 degree of dementia, 131 FEATS, 109 information, 108 inter-rater reliabilities, 114 Method-C (convex hull), 112 Method-G (grids), 111 Method-P (pixels), 111 Model-G (grade), 111 Model-R (rank), 111 QWK, 114 RCC, 114 sample, 113 severity of dementia, 131 stepwise regression, 115 subjective, 109 three examples, 112 SPSS, See Statistical Package for the Social Science Standard diagnosis, 144 reasoning process, 155 Standardized regression definition, 94 level of dementia, 225 probability of severe dementia, 226 Statistical Analysis System (SAS), 8, 93, 221 Statistical method ANOVA, 131 c_FSA, 132 example, 7 factor analysis, 132 SAS, 8 SPSS, 8 Statistical Package for the Social Science (SPSS), 8, 93, 221 Stepwise regression concentration, 120 definition, 94 grade of prominence, 101 interpretation, 120 interpretation (space usage, grade),

293

115 prominence, 102 space usage (grade), 114, 115 space usage (rank), 115 Stepwise regression (grade) details, 102 Stepwise regression (rank) details, 103 Structured Mandala Coloring (SMC) C_CREATES, 170 COD, 238 definition, 49, 204 effect, 205 effectiveness, 217 elements, 208 example, 49 expert system, 177 sample, 56, 181, 207, 219, 228 subject of computer application, 170 Subsidiary color definition, 50 System reliability details, 106 System validity 5-fold cross validation, 209 BN, 252 c_FSA, 129 concentration, 119 DDS, 15 dementia, 221 details (grade), 104 details (rank), 105 HFD, 17 identification, 209 identification of group, 212 main color, 79 p_KFD, 203 placement, 91 PPAT, 15 prominence (grade), 104 prominence (rank), 105 QWK (space usage), 116 simulation experiment, 252 variety of colors, 66 T Taxonomy applied elements, 13 basic elements, 13 C_CREATES, 13 Type I error definition, 206 identification of group, 212 KFD and FKD, 206 placement, 86

294 Index Type II error definition, 206 identification of group, 212 KFD and FKD, 206 placement, 86 U Unstructured mandala definition, 204 V Validity, See System validity Variety of colors definition, 62 inter-rater reliability, 64 principle of the transitivity, 62 sample, 61 simple rule of evaluation, 60 system validity, 66 W Warm color definition, 52 example, 52