Stereo Matching for Larger Disparity Range Using Gradient Information and Adjacent Segments Cooperative Optimization

In this paper, a novel stereo matching algorithm for larger disparity range is proposed that a new local energy function and a new global energy function are presented, which use the gradient information and adjacent segment geometric constraint respectively. First get the reliable pixel disparity, then, infer the disparity of un-reliable pixel from the reliable pixel disparity using the neighbor segments cooperative optimization method. In this paper, penalty terms are used to handle the occlusion, the disparity discontinuity and less-texture with segmentation information. We use image pairs to test, which are still more challenging than the standard stereo benchmarks such as the Middlebury Teddy and Cones images, due to their larger disparity range and higher percentage of un-textured surfaces. Experimental results demonstrate the outstanding performance of the proposed method in accuracy.