Stereo matching based on guided filter and segmentation

For improving the accuracy of stereo matching and maintaining discontinuity of object edge and continuity of non-edge area in the matching result, a stereo matching method based on guided filter and mean shift is proposed in this paper. Matching cost function based on Markov random field (MRF) and guided filter are established, and the initial disparity value of each pixel is calculated by minimizing the corresponding matching cost function. Mean shift algorithm is used to segment the stereo images into independent areas, and improve the final disparity map. Results of the proposed method are tested on the international standard Middlebury stereo benchmark and compared with other methods. Comparative results show high accuracy of the proposed algorithm and its superiority to some prevailing algorithms.