Stereo Matching Algorithm Combining Image Segmentation and Improved Belief Propagation

Abstract

In this paper, a stereo matching algorithm combining image segmentation and improved belief propagation is presented to address the problems of poor real-time performance, low matching accuracy of weak texture regions, error matches and disparity holes. The mean shift algorithm is applied to segment the reference image and define the disparity template, which improves the real-time performance of the algorithm and the matching accuracy of the disparity flat regions. The cost value is calculated by the Sum of absolute differences algorithm to obtain the initial disparity map, then the occlusion points are detected by the left-right consistency detection and the invalid values are filled, in the disparity map acquisition. The weighted least squares method is used to calculate the parameters of disparity template, and the template is allocated by the improved belief propagation algorithm to obtain the regional optimal disparity plane. Finally,the dense disparity map is acquired by this algorithm. The result shows that the presented algorithm can greatly reduce the error match rate in weak texture regions and occlusion regions which meets the requirements of stereo matching.