基于反色均值化的非参数Census立体匹配算法

Non parametric Census stereo matching algorithm based on color reversing equalization

针对传统非参数Census变换过分依赖于变换窗口的中心像素灰度值的问题,提出了一种新颖的非参数Census变换的立体匹配算法。首先,该文研究了非参数Census变换并分析了其局限性;其次,考虑到像素空间位置,对于变换窗口进行反色处理后,求取邻域均值的反色来替代中心像素灰度值进行Census变换,提高了算法的抗干扰能力;最后,采用绝对误差和(SAD)方法和汉明距离作为相似性测度进行立体匹配,减少误匹配率,并通过左右一致性检测和亚像素插值来增大匹配的准确度。实验结果表明,此算法能够提高光照变换下匹配的鲁棒性,尤其对于深度不连续区域,适用于实时的立体匹配。

Aiming at the problem that the traditional nonparametric Census transform is too dependent on the gray value of the center pixel of the transform window, a novel stereo matching algorithm for nonparametric Census transform is proposed. Firstly, the nonparametric Census transform is studied and its limitation is analysed;Secondly, considering the pixel space position, the inverse colour of the transform window is obtained to replace the gray value of the center pixel for Census transform, which improves the anti-interference ability of the algorithm;Finally, the sum of absolute of differences (SAD) method and Hamming distance are used as similarity measures to reduce the mismatch rate, and the accuracy of the matching is enhanced by the left and right consistency detection and sub-pixel interpolation. The experimental results show that this algorithm can improve the robustness of the matching under illumination transform, especially in the depth discontinuous region, and is suitable for real-time stereo matching.