Local Adaptive Support Region Approach for Stereo Correspondence

Abstract

Aiming at the difficulties in choosing the optimal support region for the given point in local stereo matching methods, the depth of pixels in support region should be similar to the above given point. A new local stereo matching algorithm based on region growing is proposed. At first, getting an adaptive support window by means of the method based on region growing. Then, calculating IoU of the support regions of the given point in reference image and the candidate matching point in target image, and the coefficient of shape similarity. Using these criterions to reduce the influence of noise candidate matching points. Whereafter calculating the total matching cost. At last, using WTA (Winner-Takes-All) strategy to get the initial disparity, and computing the frequency of every disparity of the pixels in support region, considering the disparity with highest frequency to be the optimal disparity. Extensive experiments on test images of Middlebury platform demonstrate the excellent performance of proposed method.