Extracting and matching perceptual groups for hierarchical stereo vision

Hierarchical systems have demonstrated to provide a robust an efficient framework to perform stereo vision. In this scheme, regions are utilized as high level primitives to guide the correspondence of lower-level primitives such as segments and edges. We propose to use perceptual groups instead of regions on top of such a hierarchy. A perceptual group is a set of segments generated by the application of purely Gestalt rules as well as convexity. In this paper we sustain our approach and describe the perceptual group extraction and matching process