Abstract: In order to improve the matching accuracy of the stereo matching algorithm in a low-texture region and its cost volume consistency across multiple scales,an improved Census transform algorithm is proposed，the judge and self-adaptation of weights are conducted based on the mean information of the pixels in a window,and the matching accuracy of the pixels in a discontinuity region is improved by performing a cost calculation.Further- more,the Gaussian Pyramid is introduced in cost aggregation,the Guided image filter algorithm is integrated with the multiple-scale model，and the regularization constraint is also added to strengthen the cost volume consistency with the low-texture region.In the disparity selecting step,a series of optimization methods,such as the outlier detection，the region voting and the sub-pixel enhancement,are used to improve the correct rate of the matching.Experimental results show that the proposed algorithm achieves an average error rate of 5. 91% in the Middlebury benchmark，with a better disparity map in the low-texture and discontinuity regions,and that it is of a strong ro- bustness.

Key words: machine vision, stereo matching, census transform, multiple scale region, guided image filter