Variable support-weight approach for correspondence search based on modified census transform

The computation of stereo depth is a very important field of computer vision. Aiming at solving the problem of low accuracy of traditional Census-based stereo matching algorithm, a variable support-weight approach for visual correspondence search based on modified Census transform is proposed in this paper. On the basis of analyzing defects of the traditional Census transform, a modified Census transform algorithm using average value of minimum evenness sub-area as a reference instead of the center pixel intensity as a reference is raised which enhances robustness of the algorithm. The matching accuracy is improved by weighting the average value and the standard deviation of Hamming distances in a block. The experiment results indicate that the proposed approach works better than traditional ones. Accurate disparities can be obtained even in the depth discontinuities regions.