Improved belief propagation based on RGB vector measure for stereo matching
Yingnan Geng, Yan Fei Zhao, Hexin Chen · 2011
Many algorithms for stereo matching including local optimization methods and global optimization methods have been proposed. Belief propagation (BP) method as a global method is one of the best methods for stereo matching. But BP method has still problem due to the ambiguous appearances of pixels in the stereo images. In order to make appearances of the pixels more clear, pixels are considered in the RGB vector space and RGB information of the pixels instead of gray information of the pixels is used in this paper. As we known, global algorithms are to minimize a global cost function including the smooth function and the data function. In this work, RGB vector measure is introduced into the smooth function and the data function respectively. Then disparity maps can be obtained based on BP method. Experimental results show that the proposed method is very effective.