Stereo Matching Based on Features of Image Patch

Stereo matching is a branch of 3D vision and has a wide range of applications in 3D reconstruction and autonomous driving. Recently, stereo matching methods leverage the information of two full image to calculate disparity map. However, these methods still have difficulties in texture-less regions and occlusion regions, and post-processing is used to improve accuracy. Therefore, there is a large computational cost in the feature extraction and post-processing. In this paper, we propose a stereo matching method based only on features of image patches and predict the disparity of region without occlusion. And post-processing is performed to modify all kind of mismatching based on the correct disparity. Furthermore, we evaluated our proposed method on the Middlebury dataset. The results show that our method performs well in all areas.