An Adaptive Window Stereo Matching Based on Seed Voting

Stereo matching remains a difficult vision problem for the image noise, textureless regions, depth discontinuities and occlusions. This paper presents a novel adaptive window stereo matching algorithm based on seed voting to solve these ill-posed problems. There are three key contributions in this paper: combining color and gradient as conditional tags to restrain initial window size for further adaptively updating window shape; in terms of disparity optimization, improving the original Left-Right-Difference method to achieve more precise initial seed map; proposing an efficient seed growth based on vote for finial dense disparity map. Experimental results based on Middlebury stereo dataset demonstrate the superior performance of the proposed algorithm. The novel local-based approach stimulates stereo matching research to further development since many advanced stereo algorithms call for accuracy local results as initial disparity map.