Learning perceptual organization for straight line segments

Address the problem of structure inference in an image, in the framework of perceptual organization. This paper describes work in progress which builds on a subset of the authors´ previous work on fuzzy perceptual grouping. More precisely, the authors are concerned with obtaining a fuzzy system which can achieve grouping of line segments. The data are obtained from images and consist of the results of edge extraction to which a line segment fitting algorithm has been applied. For each collection of similar and collinear segments used as input, a representative segment is used to summarize this collection. In the training stage both the input collection and the output segment can be either indicated by a human user, or obtained by overlaying the segment and real images