Stereo matching is a process of estimating the disparity for every pixel in a pair of stereo images. Among the various stereo-matching algorithms, the adaptive supportweight (ASW) approach is the most representative local method that determines the disparity by using statistical correlation within a local support window. In the ASW approach, the degree of correspondence between two pixels within the support window is jointly measured using color similarity and geometric proximity. Although ASW is simple and typically effective, it does not provide satisfactory matching results for regions with ambiguous patterns or no identifiable image features. This study improved ASW by incorporating numerous local image features. First, an alternative adaptive cross-based support window, instead of the fixed square window, was used to increase the matching accuracy if a sparse-textured region surrounds the anchor pixel. Second, Canny edges in dense-textured regions were used for correspondence matching. Third, a smoothness constraint on disparity continuity was imposed to shrink the errors on object boundaries. Fourth, a disparity-refinement process was performed on the inconsistent pixels from the left-right consistency check. Experimental results on the Middlebury test bed confirmed the superiority of the proposed method over the ASW. Compared with other state-ofthe- art ASW-based algorithms, the proposed method is among the most effective and achieves stable and efficient performance with only a slight increase in processing time.