基于改进Census变换和网状聚合的立体匹配方法

Stereo Matching Method Based on Improved Census Transform and Mesh Aggregation

为了获得连续稠密的视差图像,提出了一种基于改进Census变换和网状代价聚合的立体匹配方法。在初始匹配代价计算中,用灰度排序剔除极值的方法获取参考像素,改善Census变换对于亮度差异的抑制效果。在代价聚合计算中,采用SLIC超像素分割方法对图像进行区域划分,在区域内按照网状搜索去寻找邻域像素,并根据邻域像素距离待匹配像素的远近配置权重,完成待匹配像素的代价聚合。实验结果表明,提出算法计算出的视差图像,坏像素比例低、视差连续而稠密,其性能与Middlebury平台上排名第一的LocalExp算法接近。

In order to obtain continuous dense disparity images,a stereo matching method based on improved Census transform and mesh cost aggregation is proposed.In the initial matching cost calculation,we use gray sorting to eliminate the extreme value to get reference pixels,and improve the suppression effect of Census transform on luminance difference.In the cost aggregation calculation,the SLIC super pixel segmentation method is used to divide the image region.In the area,the neighborhood pixels are searched according to the pixel network,and the distance and near configuration weight of the pixels to be matched according to the neighborhood pixel distance is used to achieve the cost aggregation of the pixels to be matched.The experimental results show that disparity are continuous and dense in disparity image calculated by proposed algorithm,and its performance is close to the LocalExp algorithm ranking first on the Middlebury platform.