Segment-Based Multi-View Stereo Matching Using Cooperative Optimization

摘要

In this paper, the disparity map is extracted by the method of segment-based stereo matching using cooperative optimization from multi-view. The Mean-Shift method is employed to segment the reference image in the first step of the algorithm, this is to ensure our method to preciselylocalize the depth boundaries and correctly estimate the large untextured regions. Next, the Self-Adaptation dissimilarity algorithm is utilized to estimate the initial disparity of multi-view. Then the disparity sets of multi-view are merged into one using the disparity plane estimation,which is solved by Singular Value Decomposition (SVD). In order to ensure reliable pixel sets for the segment, the rules are built to filter out outliers. Finally, using the cooperative optimization for a novel energy function is subsequently formulated to refine the disparity map. The experimentalresults are presented to show the effectiveness of our proposal methods.