A Novel Measurement and Statistics of Similarity Cue for Edges in Contour Grouping

Abstract

Similarity is one of the important Gestalt laws for contour grouping. Edges are the basic elements of contour grouping. We define a novel measurement of edge similarity on the case of that an edge representation includes an estimate of the image intensity on either side of each edge. We discuss how to choose a proper feature to encode edge regions and which shape and size region is proper to edge similarity measurement. We also propose a new statistic of similarity cue which considers similarity on the basis of proximity, and which enhances the power of similarity cue in contour grouping effectively.