Shading-Based Two-View Matching

This paper presents a region-based stereo matching algorithm, in which the regions are computed from the Gaussian (K) and mean (H) curvature information of the image intensity surface. A region is defined as a connected area of constant K/H sign combination, and thus constitutes the footprint of a local section of the intensity surface, whose shape is a peak, a pit, a positive or a negative saddle. Region adjacency information is explicited by means of a Voronoi graph representation of the region map. Matching between nodes of identical shape types in the two region maps is established by comparing the topological configuration of their immediate neighborhoods. The disparity map is established through a coarseto-fine strategy.