A Radiometric Varying Robust Stereo Matching Algorithm

In order to improve the matching rate of radiometric varying stereo images, a novel stereo matching algorithm based on the improved epipolar distance transformation in log-chromaticity space is proposed. In log-chromaticity space, the intensity proportion of stereo image pairs is computed firstly according to raw disparity map; secondly, epipolar distance transformation is performed on left and right images respectively using proportional intensity deviation parameters; at last, the final disparity map is acquired by the belief propagation method. Theoretically, the matching rate of the proposed algorithm is independent of radiometric varying situations including differences in light source's position, spectrum, intensity and the parameters setting of cameras. Experimental results indicate that the matching rate of the proposed algorithm is improved at most 60% comparing with the original epipolar distance transformation algorithm and at most 78% comparing with the state of art algorithms such as ANCC (adaptive normalized cross correlation) when applied to textureless image pairs.