Accurate Belief Propagation with parametric and non-parametric measure for stereo matching

Stereo matching is one of the most active research areas in computer vision. Recently, Belief Propagation algorithm based on global optimization has great advances. However, traditional data term of Belief Propagation algorithm mainly lies on pixels-based intensity measure, and its effect is not very well. In this paper, a novel stereo matching is proposed that utilizes Census measure and pixels-based intensity measure into data term of Belief Propagation algorithm. But only simply adding Census measure is not enough to improve the accuracy of Belief Propagation, therefore the post procession for the algorithm is very essential. We combine intensity measure with Census algorithm into data term of Belief Propagation algorithm and acquire more accurate results by using the post procession. This proposed method may be more exacter than traditional Belief Propagation algorithm. The experimental results demonstrate the superior performance of our proposed method.