Stereo matching with cross-based region, hierarchical belief propagation and occlusion handling

Abstract

This paper proposes a new global based approach for stereo matching with occlusion handling. We first use a global method based on the energy function to get the initial disparity, in order to get the accurate result, we use cross-based region as data term, and a hierarchical belief propagation strategy is used to smooth the disparity map. To improve performance we introduce the stability factor to determine the smoothness of disparities. For occlusion handling, the main ideal is using segment based strategy to deal with occlusion pixels. Finally, the algorithm is evaluated on the Middlebury data sets, Experimental results shows that the proposed method is competitive with the state of the art.