Adaptive epipolar distance transform and adaptive matching cost for stereo matching

This paper presents a new algorithm that robustly performs stereo matching for textureless regions in stereo images. To this end, we design an adaptive matching cost which employs a special term. This term can assign distinguishable values to pixels adaptively according to the texture information. Specifically, first, we improve the epipolar distance transform by utilizing a linear expansion function and obtain an adaptive epipolar distance transform (AEDT); second, we propose an adaptive matching cost utilizing the AEDT to deal with textureless region problems. Experiments on the Middlebury benchmark demonstrate that the proposed method can perform accurate stereo matching on textureless regions. Moreover, the experiments show that the proposed adaptive matching cost can be directly utilized to other methods to improve the disparity results in textureless regions.