Disparity refinement based on segment-tree and fast weighted median filter

The depth of a scene is very important in computer vision. One way to get the depth map is by stereo vision. But because of the noise, textureless area of the scene and the occlusion area, stereo matching becomes rather challenging. Some very good algorithms have been proposed. In this paper, the mainstream semi-global stereo matching algorithm (SGM) is studied, and a disparity refinement algorithm is proposed. By using SGM, an initial disparity map is computed. Then a better disparity map can be obtained by applying the disparity refinement algorithm. The proposed refinement algorithm is based on a segment-tree and a fast weighted median filter (WMF). Some experiments are done based on the well-known Middlebury dataset. The results show that the proposed algorithm can improve the quality of the disparity map effectively in most cases.