Superpixel Smoothing for Disparity Refinement in Stereo Matching

Stereo matching algorithms are mainly divided into two classes, local algorithm and global algorithm. Most local algorithms consist of four stages, initial cost computation, cost aggregation, disparity computation, and refinement. In this paper, we propose a method to enhance the disparity accuracy at regions near depth discontinuities. In the refinement stage, following the traditional Left-Right Consistency check, we classify each pixel by using superpixel segmentation. We then detect and correct outliers in superpixels. The experiments show that the disparity map generated by the proposed superpixel smoothing in the refinement stage is smoother than other existing methods.