It is crucial to choose a suitable matching cost relationship for area-based stereo matching. Inter color correlation as the inherent characteristics of pixel, can effectively show the small matching differences between pixel. The traditional method of adaptive support weight is improved, and a new matching algorithm that combines the color similarity, Euclidean distance similarity and inter color correlation similarity is proposed to compute the corresponding support weight. Meanwhile, in order to solve the brightness difference between the stereo image pairs, matching pixel are transformed by rank and calculating the matching cost relationship. Various errors in the disparity results are effectively handled in a three-step refinement process. The experimental results based on Matlab show that its performance is the best among local matching methods, and its robustness is very strong, as well as low false match rate.