For color image stereo matching based on the traditional census transform easily affected by illumination, and matching accuracy was not high. An color image stereo matching based on improved census transform was proposed. Firstly, it taken the color distance standard deviation of census mask instead of the centre pixel, comparing with the color distance difference of per neighborhood pixels and the color distance mean intensity of the census mask to build the census transform. Secondly, Hamming distance was taken as cost function, and then, Winner-take-all strategy and left and right check were implemented to obtain disparity dense map. The experiment results show that the proposed method had high computational efficiency, and strong robustness for the interference caused by illumination change.