Small Baseline Stereo Matching Method based on Maximum Likelihood Estimation

Abstract

A small baseline stereo matching method based on maximum like estimation is proposed. Firstly, the mixed window strategy is used to determine the matching window for each point in the reference image. Secondly, the normalized cross correlation function is employed to compute matching cost for possible matching points in the disparity range, then the “winner-take all” strategy is utilized to compute the disparity. Finally, the sub-pixel stereo matching method based on maximum like estimation is developed to obtain the sub-pixel disparity. The experimental results show that this method efficiently decreases adhesion phenomenon and achieves more accurate sub-pixel disparity