A local stereo matching approach using adaptive weight boundary

A new region-based local stereo matching algorithm with accurate disparity estimation is proposed. For the local stereo matching, finding an appropriate support window is crucial to the performance of disparity estimation. In order to generate an accurate support region, a modified cross-based local approach combined with mean-shift segmentation is performed. We then further improve the reliability of the aggregation cost by adaptively adjusting the weights of the boundary pixels in the support region. With the stable aggregating matching cost, some noises caused by outliers at the object boundaries can be removed, which is greatly beneficial to the quality of the depth estimation. Experimental results show that the proposed method achieves superior performance compared with other state-of-the-art methods.