Segment-based continuous stereo matching

Stereo matching is a fundamental problem in computer vision, which aims to compute the disparity between stereo image pairs. Depth discontinuous regions are one of the most difficult regions that stereo matching algorithms have to solve, to accurately compute the disparity in those regions, in continuous methods, energy functionals which explicitly modeled the discontinuity characteristic of disparity function are usually proposed, and then those functionals are relaxed into convex. Those energy functionals usually use a simple data term which is not robust to brightness changes, and complex projection algorithms are needed to solve the relaxed convex functional. In this paper, we take the segmentation prior into the data term to make the functional more robust to brightness changes, and we propose a simple projection algorithm to give a approximate solution when solving the relaxed stereo matching energy functional. Experiments showed that our method can give a satisfied solution as other continuous methods while have an advantage that is easy to implement.