Stereo Matching Based on Adaptive Window and Reliability Constraint ?

Focusing on the problem of stereo matching accuracy not high enough in the image matching with discontinuous or weak texture, this paper is to propose a stereo matching method which takes the advantages of the local matching and global matching. Firstly, color image segmentation is applied for mean shift algorithm. Secondly, in the local matching the multi-window adaptive algorithm is used to split the results on cost function constraints. Thirdly, it is to use pixel classification reliability constraint to differ the parallax results. Finally, the reliability of pixel classification is to guide traditional belief propagation, which can ultimately optimize the result. Compared with the test provided by Middlebury, results show that this algorithm has effective results in image depth experiment of discontinuous areas and weak textured areas.