融合梯度特性与置信度的立体匹配算法

针对现有局部立体匹配算法在计算匹配代价时, 不能很好区分强弱纹理区域,及在视差计算过程 中,不能很好的解决视差歧义问题,提出一种融合梯度特性与置信度的立体匹配算法。首先 计算梯度特 征,并根据梯度特征信息选择匹配代价计算的匹配窗口,针对强弱不同纹理区域选择不同尺 寸的匹配窗 口,有效的提高了立体匹配精度,降低了误匹配率；然后在视差计算中引入置信度约束条件 ,解决了视差 计算中视差歧义的问题,提高了立体匹配算法的稳定性与精度；最后使用水平与垂直方向交 叉区域检测进 行奇异值的修正。实验结果表明,该算法在Middlebury数据集中31对 立体图像对的平均误匹配率为7.96%,有效的提高了立体匹配精度。

Stereo matching algorithms fusing gradient characteristic and confidence

As the existing local stereo matching algorithms can not distinguish t he strong and weak different texture regions well,and can not solve the disparity ambiguity p roblem in the step of disparity selection,a stereo matching algorithm based on gradient characteri stics and confidence is proposed in this paper.Firstly,the gradient features are calcula ted,and the support window which is used to calculating the matching cost is selected according to g radient feature information,and different sizes of support windows can be selected according to strong or weak different texture,which effectively improves the accuracy of stereo matching an d effectively reduces the false matching rate when in the process of calculating matching cost .Secondly,the confidence constraint is introduced into disparity calculation to solve disparit y ambiguity in the step of disparity calculation and improve the stability of stereo matching algor ithm.Finally,the singular value which is detected by left and right consistency detection is corr ected by using cross- region detection in horizontal and vertical directions.The experimental results show that the average mismatch rate of 31pairs of stereo images in the Middlebury dataset is 7.96%.Compared with common stereo match algorithms,the result show proposed algorithm is effec tively improves the stereo matching accuracy.