Gestalt Theory in Image Processing: A discussion paper Mario K¨oppen, Kaori Yoshida Kyushu Institute of Technology, 680-4, Kawazu, Iizuka, Fukuoka 820-8502 Japan mkoeppen@ieee.org,yoshida@ai.kyutech.ac.jp Pablo A. Valle Technical University Berlin Strasse des 17. Juni 135, 10623 Berlin, Germany valepnbb@mailbox.tu-berlin.de 1 Introduction The apparent phenomenon of Gestalt laws in visual perception has stimulated a steady flow of research contributions in the image processing and pattern recognition research community. In this paper, we want to provide a critical review of a representative number of such contributions and point out some is- sues that have not been reflected well in all these past works. A rough estimate gives a few hundert research con- tributions on the topic of Gestalt (theory, psychol- ogy, laws) in computer science 1 . The recent num- ber of publications is about twice as large as 20-30 years ago, which is conform with the general increase of the number of research publications. Thus, there seems to be a small, but rather constant share of pa- pers devoted to the general theme Gestalt. It is also notable that mostly these contributions are isolated ones, with few amount of referencing between them. So far, we could not witness any generally accepted means of introducing Gestalt in image processing. The comparable low amount of research devoted to the direct linking between Gestalt theory and im- age processing does not entail a low practical im- pact. There are many potential industrial applica- tions waiting for any more thorough approach to the Gestalt theme. The most obvious one is gaining a deeper insight into the functionality of human (or higher animal) vision, which has been the pre-cursor of any image processing related development so far. But also, it has a strong relation to any design study. 1 For example, the SCOPUS lists about 200 relevant publi- cations for the past 30 years. There are iconic signs in use all over the world (issu- ing warnings, guidance, information), which have to be designed in a manner to be easily comprehensible in their meaning by a human observer (and often from a larger distance, within a short time, and while be- ing only partially visible). These signs often directly employ Gestalt concepts, but this is usually the re- sult of the considerations of an experienced graphic designer, that means, a human. At present, there is no test known for the suitability of a, for example, warning sign that does not involve a human observer. Same for specific, content-related designs, like a web- page: the guidance of a webpage visitor’s attention is clearly following Gestalt laws, but after publishing a webpage, any measure for its content delivery will in- volve a human. In a similar manner, product design, advertisement banner, human-machine computer in- terfaces and the more are all referring to Gestalt con- cepts in their design today (and there are much more Gestalt laws appearing in an urbanized environment than in the free nature), but the effectiveness of ful- filling their purpose can only be valued by the human, and not a machine. In the following, we will discuss these issues more deeply, especially focussing on the fact that there is barely a visual scene in the real world, where only one Gestalt law is active. We will present some ex- amples in section 2, and then consider the relevance of Gestalt theory related publications to these ex- amples in section 3. Then, we will conclude with a short comment about the suitability of soft com- puting methodologies for Gestalt-related processing, which has not been considered much in the past. 1 SMCia/07 2007 Three-Rivers Workshop on Soft Computing in Industrial Applications University of Passau, Germany, August 1 - 3, 2007