Stereo matching is to obtain disparity map from two or more images whose pixels existing disparity between each other. Aim at the difficulty of meeting both accuracy and speed when using the adaptive support-weight algorithm, an improved adaptive support-weight sparse region-based algorithm accorded with HVS is proposed. First, improve the traditional support-weight formula, and the support-weight of the points can be calculated according to the improved support-weight formula. Second, the dense disparity map is obtained by sparse region-based matching. Third, left-right consistency check and blocking filling is performed for the obtained disparity map. Finally, median filter is used to remove isolated mismatching points and noise points. Experiment results show that, using the presented method, the matching efficiency is above 90 times faster than that of the adaptive support-weight algorithm proposed by Yoon; and the matching accuracy is 12.34% higher than that of SSD. So the improved algorithm is verified for obtaining accurate disparity map fast and meeting the requirements of system practicability.