Disparity optimization algorithm of stereo matching based on adaptive weight Markov random field

In this paper, Aiming at the parallax optimization problem in stereo matching, a stereo matching disparity optimization algorithm based on adaptive weight MRF is proposed. Firstly, the local disparity map is generated by using the local algorithm. Then, under the framework of Markov random algorithm, the parallax map is optimized by adaptively adjusting the smoothing weights for the disparity region and the smooth region. The key technology of this paper is the design of smoothing term weights in different regions. The discontinuous region uses the color correlation coefficient, the disparity correlation coefficient and the structural similarity between the disparity map and the same scene color map to construct the smoothing item weight; while the smoothing region directly uses the parallax. The information is used to construct smoothing item weights, and at the same time, the parameters in the weights are adaptively adjusted according to the degree of parallax smoothing in the domain. The experimental results on the Middlebury stereo matching evaluation platform show that the proposed algorithm can obtain better subjective and objective indicators, and the optimized disparity map can maintain a clear disparity map in the edge region and the parallax discontinuous region.