Hierarchical Matching Using Combinatorial Pyramid Framework

A string matching approach is proposed to find a region correspondance between two images. Regions and their spatial relationships are represented by two combinatorial pyramids encoding two segmentation hierarchies. Our matching algorithm is decomposed in two steps: We first require that the features of the two matched regions be similar. This threshold on the similarity of the regions to be matched is used as a pruning step. We secondly require that at least one cut may be determined in each hierarchy such that the cyclic sequence of neighbors of the two matched regions have similar features. This distance is based on a cicular string matching algorithm which uses both the orientability of the plane and the hierarchical encoding of the two regions to reduce the computational cost of the matching and enforce its robustness.