Matching difficulty and the occurrence of large errors in the weak and repeated texture areas of an image are the problems associated with the stereo matching algorithm. To solve these problems, this paper proposes a stereo matching algorithm based on improved cost calculation and a disparity candidate strategy. First, the improved Census transform and adaptive weighted bidirectional gradient information are combined to estimate the initial matching cost, improving the reliability of cost calculation. Here, inner circle coding is added to the traditional Census transform for improving the utilization of neighborhood information while reducing the impact of noise. The adaptive weight function is used to combine the horizontal and vertical gradient costs for reducing the mismatching rate of the object edge areas. Second, after cost aggregation with an adaptive cross-window, the initial disparity can be obtained by establishing candidate disparity sets and introducing neighborhood disparity information. Finally, the disparity is optimized via two-round interpolation. Experimental results demonstrate that the proposed algorithm can improve the stereo matching of the weak and repeated texture areas and that the average mismatching rate on four standard stereo image pairs in Middlebury is 5.33%.