<p>In order to improve the matching results of binocular vision stereo matching in the occluded area, this paper proposes an iterative framework that uses hypothetical optimal values to compensate the disparity space map. First of all, the three-moded Census transform is improved to reduce the excessive dependence on the center pixel of the neighboring pixels in the support window. Then, the disparity map after the complete calculation process is reversely constructed into a disparity space image with disparity information, which is used to compensate for the disparity space image obtained after cost calculation. The multi-path cost aggregation method is employed to complete the aggregation by combining the cost of pixels in multiple directions at different disparity. Finally, the methods including consistency detection, hole filling and sub-pixel refinement are used to further process the disparity map. Experiments show that the matching result of the method proposed in this paper has been improved after several iterations. Not only can it achieve better results in the matching of occluded areas, but also has better performance for uneven illumination.</p>