改进的Yoon自适应加权立体匹配算法

分析了Yoon自适应加权立体匹配算法在权值计算与像素相异性度量上所存在的问题,并对其进行了改进。首先分析了Yoon自适应加权立体匹配算法颜色相似性和空间邻近性对权值(视差)的不同影响,据此提出新的空间临近性权值影响模型;然后将Birchfield相异性度量函数从2-邻域扩展到8-邻域,消除行采样对像素相异性度量的影响,使得匹配代价计算更为准确;最后通过对称加权确定视差。实验表明,新方法改善了Yoon算法在深度不连续区,低纹理区,高纹理区以及重复结构区的匹配性能,可获得更为精密的视差图。

Improved Yoon stereo matching algorithm based on adaptive weight

The disadvantages of Yoon adaptive support-weight stereo matching algorithm about weight determination and the dissimilarity measurement of pixels are analyzed and improved.Firstly,the different influences of color similarity and space proximity in the Yoon adaptive-support stereo matching algorithm on the weight（disparity） are analized according to the disparity assumption based on color segmentation,on the basis of which a new weight determination model of spatial proximity is proposed;Then,the Birchfield dissimilarity measurement function is extended from 2-neighborhood to 8-neighborhood,which eliminates the influence of sampling between rows on the dissimilarity measurement and the matching cost is more accurate;Finally,the disparity is determined by symmetric weight.The experiments show that：the new method improves the matching performance of Yoon＇s in the disparity discontinuity areas,the low textured surfaces,the high textured surfaces and the repeated structure areas and the more accurate disparity map can be obtained.