Hierarchical Stereo Matching Via Color Segmentation

This paper presents a new segment-based stereo matching algorithm using a coarse-to-fine multiresolution strategy. In general, the multiresolution scheme was employed in a direct descending optimization to avoid matching in the local minimal. However, it is not effective in homogeneously textured regions and disparity discontinuous boundaries. We implement a novel hierarchical segment-based matching scheme to produce accurate disparity maps, which resolves the above mentioned challenging problems. We utilize the hierarchical color segmentation that is classified in each resolution level to guide the matching process gradually. The principal feature of our scheme is to combine the local and the global optimization in multiresolution stages. To validate the proposed algorithm, we show the experimental results on the new Middlebury database sets