一种光照变化条件下的自适用权重局部立体匹配算法

An Adaptive Support-Weight Local Stereo Matching Algorithm Under Variant Illumination

基于Census变换的局部立体匹配算法,对于光照变换具有较高的鲁棒性,但在图像内视差不连续区域存在匹配误差,导致前景膨胀现象。本文提出一种新的在RGB空间基于稀疏Census变换的自适应权重的局部立体匹配算法。通过在RGB 3个彩色通道分别进行稀疏Census变换,根据颜色距离和空间距离,计算在局部窗口中各个像素的权重,从而进行聚合。实验结果表明,该算法能够有效消除前景膨胀现象,提高了对视差不连续区域的匹配准确度,同时提高了光照变换的鲁棒性。

Local correspondence method based on Census transform is robust to illumination change, but exhibit a ＂foreground fattening＂ effect, which is caused by mistakenly matching near depth disconti- nuities areas of the picture. In order to solve this problem, a novel adaptive support-weight local corre- spondence algorithm was proposed within RGB color space based on sparse census transform. The sup- port-weights of each pixel in the given support window were used for correspondence search, which were calculated individually in RGB channels using sparse Census transform according to the color similarity and geometric proximity. The experiments show that the stereo matching algorithm improves the matc- hing accuracy in discontinuities areas to avoid the foreground fattening effect, and it is robust under var- iable illumination.