A versatile matching algorithm based on dynamic programming with circular order preserving

Abstract

This paper presents an assignment algorithm with circular order preserving constraint. Given a cost affinity matrix and the desired percentage of correspondences, the algorithm implemented using dynamic programming determines the correspondence of type one-to-one of minimum global cost. Here, it was applied to optimize the global matching between two sets of ordered points that represent the contours of objects previously segmented from images. In the tests performed, we considered affinity matrices previously built based on information on curvature and distance to centroid. The results that have been obtained are better than the ones presented in previous studies, for the cases in which partial deformations or occlusions are involved.