An Edge and Color Oriented Optical Flow Estimation Using Block Matching

For most tasks in image sequence analysis, the estimation of optical flow is the primary step towards the final solution. Block matching, which is adopted by many current coding standards, is the most popular method among other approaches. Research in this area mainly focuses on three aspects: matching criteria, searching criteria and the determination of block size. In this paper, we propose a block matching approach taking into account edge information and color information to improve matching accuracy. Edge detection is first performed on the frames to be matched. Block matching is then carried out by comparing intensity, edge and color similarity of the blocks. In order to offset the computational cost introduced by the edge and color matching, a fast searching algorithm of four step searching is exploited in the searching stage. Our preliminary results show it performed better than the common intensity based method.