Adaptive Epipolar Distance Transform-Based Stereo Matching

Considering the epipolar distance transform-based stereo matching is sensitive to noise and ambiguous in discontinuous regions, an adaptive transform-based stereo matching algorithm is proposed. By employing image structure features, an iterative object’s scale algorithm and a block discontinuous map are proposed to choose the parameters adaptively. This transform converts image intensity values to a relative location inside a planar segment along the epipolar line, so that the pixels in the low-texture regions become distinguishable, and the edges are preserved. Besides, the mean value in the local window is adopted to calculate the length of segmentation line, improving the robustness to noise of pixels in low texture area. Experimental results on several real images demonstrate the effectiveness of the proposed transform in low-texture regions and discontinuities regions. When the stereo matching algorithms use the proposed transformed images to obtain disparity maps, the matching accuracy is improved and the matching errors of discontinuous regions, such as edges and noises, are reduced.