Skeleton based shape matching using reweighted random walks

Shape matching is a very important issue and challenging task in computer vision. In this paper, the problem of finding a matching between two shapes is addressed by establishing correspondences between two their skeleton graphs based on random walk framework. We first propose a novel skeleton graph model in which nodes represent end-nodes of skeleton while edges describe relations between two end-nodes. Matching between two skeletons is then formulated as graph matching, which is solved by ranking on an association graph via random walks. By applying the random walks with reweighting jumps on the association skeleton graph, the proposed method can collect potential matches, eliminating the unreliable matches, which are affected by noise and distortion. Comparative experiments on several benchmark data sets show that the proposed method produces more accurate results than the previous works.