A robust matching algorithm for shape recognition based on geometric consistency

Abstract

Though there are many algorithms which devote to extracting the descriptors for the shapes, the correspondences between shapes established only by descriptor distance are not reliably. To address this issue, a new shape matching algorithm is proposed on the basis of the geometric consistency between shapes. Each shape pair is represented as a node in a graph, and the weight of each edge is computed while the descriptor R-histogram which represents the topological relationship between shapes is adopted. Then, the problem of descriptor matching can be formed as finding the principal cluster of the graph, which is solve by the Hungarian algorithm in this paper. Our proposed approach has been implemented and gives encouraging results under rotation, scaling, shearing and noise.