Perceptual grouping based on fuzzy sets

Abstract

The authors propose a new approach based on fuzzy sets for detecting relations among points, line segments, and elliptic arc segments. Appropriate constraints defining the relations are extracted. A suitable fuzzy membership function is assigned to each constraint. Then the constraints are combined by fuzzy set operations to describe meaningful relations. According to these meaningful relations, a perceptual grouping is executed. Grouping methods are described on the basis of collinear and coelliptic relations. A measure of the significance of grouping for high-level vision processing is discussed. A prototype system for perceptual grouping from image data was implemented using a frame-based knowledge representation scheme, and experimental results for real image data are presented.<>