Extracting physically nonexistent contours

Abstract

A model of the human vision system is proposed which seems to be related to highly reliable extraction of contours from an original image. In other words, an algorithm is proposed for automatically generating physically nonexistent subjective contours. This algorithm consists of the following five steps: extracting physical contours, determining candidate points through which subjective contours pass, determining a direction of subjective contours at each candidate point, determining pairs of candidate points to be connected by subjective contours, and generating curves connecting each pair of candidate points. This algorithm was applied to various figures containing line segments, and it was found that the same contours as the subjective contours perceived by human beings could be generated automatically. Further, this algorithm was applied to a figure of lines consisting only of external contours at which several convex polygons overlap each other. Then, internal contours could be extracted in the overlapping parts of objects and separate constituent polygons.