基于大窗口的自适应立体匹配

Adaptive stereo matching based on large window

针对局部匹配难以获得精确的支持区域以及在遮挡区域、纹理单一区域易引起误匹配的问题,利用窗口内图像的边缘信息,提出了一种基于大窗口的自适应支持区域选择,自适应匹配代价选择,自适应错误点处理的立体匹配方法。定义一个较大的支持窗口,通过图像的边缘信息在此窗口内自适应寻找匹配点的支持区域;根据支持区域中最小匹配块的数目自适应选择匹配代价;对初始视差图通过左右一致性检验得到的错误点进行分类处理。实验结果表明,该方法能够取得理想的效果,并且在遮挡区域和纹理单一区域具有较好的精度。

Local stereo matching is difficult to obtain the accurate support area, and error matching is easy to be caused in occluded regions and textureless regions. Aiming at this problem, a new stereo matching is presented, which uses adaptive support region selection based on large window, adaptive matching cost choice and error point processing. Firstly, a larger support window is defined, in which support area of matching points is searched adaptively through the image edge information. Then, the matching cost is selected adaptively according to the number of minimum matching blocks in the support area. Finally, the error points from the left-right consistency checking of initial parallax image is classified. Experiments show that this method can achieve a good effect, as well as for occluded regions and textureless regions.