A hierarchical binocular stereo matching algorithm combined gabor features and gray information

This paper proposed a new stereo matching method by combining gabor features and gray similarity of pixels. The main idea is hierarchical matching. The pixels with different depths are believed to locate at different layers. Our method using four steps to find the most possible pixels in each depth layer. First is coarse matching using an initial matching threshold. Second is threshold updating by analyzing statistical characteristics of these matching points. Then reliable matching points are obtained using the updated threshold and the correlation coefficient coming from gabor filters with different directions. At last the smooth disparity map is produced by gaussian filter after morphological filtering which can remove some false matching points efficiently. Experimental results in tsukuba gallery show that the proposed algorithm is more accurate and the disparity map is clear obtained hierarchically.