Stereo matching is one of the most important topics in computer vision and aims at generating precise depth maps for various smart applications. The major challenge of stereo matching is to suppress inevitable errors occurring in smooth, occluded and discontinuous regions. In this paper, we propose a robust stereo matching system, which is based on segment-based superpixels, to design adaptive matching computation and dual-path refinement. After the selection of matching costs, we suggest the segment-based adaptive support weights for cost aggregation, instead of color similarity and spatial proximity, to achieve precise depth estimation. Then, the proposed dual-path depth refinement, which refers the texture features in a cross-based support region, corrects the inaccurate disparities to successively refine the depth maps with shape reserving. Specially for left-most and right most regions, the segment-based refinement can greatly improve the mismatched disparity holes. The experimental results show that the proposed system achieves higher accurate depth maps than the conventional stereo matching methods.