An efficient earth mover's distance algorithm for shape context feature

摘要

Shape context is not rotation invairent as a local visual feature. Using earth mover's distance (EMD) for similarity comparison can make the feature rotation invairent but its computational cost is expensive. In this paper, based on the property which the feature histogram's angel dimansion circle shifts when image rotates, a fast and efficient method is presented to compute EMD between two shape context features for point matching problem. The analysis shows the time complexity is much lower and the experiments show the effective and efficiency of new method.