Guest Editors' Introduction: Perceptual Organization in Computer Vision: Status, Challenges, and Potential

1 In a classic experiment, Smith [1] presented a perceptu random stimulus to subjects who were asked to reproduc The reproduced set was then used as a stimulus for a seco of subjects who were also asked to reproduce what they w shown. After about 12 cascades the final reproduction h definite structure to it. The subjects had (gradually) impo organization on an entirely random stimulus! The experim indicates the importance of organization in human percep The human visual system values organization to such a de that it attempts to find (or impose) it even when none actu exists. Clearly this heuristic behavior has evolved in a wo which, generally, is very much structured. By perceptual or nization we refer to this ability of a vision system to organ detected features, or primitives, in images based on for insta Gestaltic criteria. To put it another way, perceptual organiza can be defined as the ability to impose structural regularity sensory data, so as to group sensory primitives arising fro common underlying cause. This sort of organization then mits the formation of object hypotheses with minimal dom knowledge and, therefore, minimal restrictions. The importance of finding organization in sensory data long been recognized by researchers in human vision, espe the Gestalt psychologists. However, until relatively recently, roles of structure and organization have been minimal in c puter vision systems. Nevertheless, perceptual organizatio been identified as one of the insufficiently emphasized area computer vision, lying as it does in the “middle ground” betwe low-level and high-level processing. Early work in perceptual organization in computer visi dates back to Marr [2] (curvilinear groupings), Witkin a Tenenbaum [3], and Lowe [4]. It was then that the well-kno principle of nonaccidentalness, also known as the principl common cause or the coincidence explanation, was postu for perceptual organization. This principle states that it is hig unlikely for organized arrangements of image features to a by chance and hence, the occurrence of such is significant. S the work of Lowe, who showed that even simple organizatio such as parallel lines and rectangles, can drastically prun recognition search tree, there have been a number of cont tions that demonstrate the importance of perceptual organiz for various vision tasks, e.g., object recognition [5, 6], stereo ally e it. d set ere d a ed ent on. gree lly rld aze nce, ion on m a erin