基于自适应权重马尔科夫随机场的立体匹配视差优化算法

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

针对立体匹配中的视差优化问题,提出一种自适应权重MRF的立体匹配视差优化算法。首先使用局部算法生成初始视差图;然后在马尔科夫随机算法框架下,对视差不连续区域和平滑区域自适应调整平滑项权重来实现视差图的优化.本文的关键技术是不同区域平滑项权重的设计,其中不连续区域利用颜色相关系数、视差相关系数以及视差图与同场景彩色图的结构相似性来构建平滑项权重;而平滑区域则直接利用视差信息来构建平滑项权重,同时,根据领域内视差平滑程度自适应调整权重中参数.通过在立体匹配评估平台Middlebury.上的测试实验表明,所提算法可获得更优的主观效果和客观指标,优化后的视差图在边缘区域和视差不连续区域能够保持清晰的视差图.

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.