一种改进的重启动随机游走立体匹配算法

An improved random walking with restart stereo matching algorithm

通过分析概率模型下重启动随机游走立体匹配算法的边缘权重仅由颜色相似性确定以及邻接矩阵不均衡的问题,对其提出了一种改进算法。首先对边缘权重进行改进,采用颜色相似性和空间邻近度共同确定,使算法在边缘区域和弱纹理区域的匹配更加精确;然后对邻接矩阵做均衡化处理,使算法对几何畸变干扰和异常值都具有很好的鲁棒性。测试结果显示,改进的算法在深度不连续区域和弱纹理区能获得更为精确的视差图。

After analysis of the problems of the random walking with restart stereo matching algorithm based on probability model that the edge weights were only decided by color similarity and the adjacency matrix was not balanced,an improved algorithm of stereo matching is proposed.First,the edge weights are changed to improve the accuracy of edge area and low textured surface marching,by introducing both of color similarity and space proximity.Then,a balance method for adjacency matrix is presented,which is robust to outliers and geometric deformation.The experiments show that the new method improves the matching performance in the disparity discontinuity areas and the low textured surfaces,and the more accurate disparity map can be obtained.