立体匹配, 其目的是在对两幅存在一定视差的图像进行匹配后, 获得两幅图像精确的视差图, 是计算机视觉领域的重点和难点问题。为了快速、高效地获得高精度的稠密视差图, 在充分研究了基于边缘特征的立体匹配和基于局部窗口的立体匹配两类算法的基础上, 提出了一种基于局域边缘特征的自适应立体匹配算法。首先该算法自适应地选择局部匹配窗口大小, 并计算其中相应位置的权值, 实现初次匹配, 然后再以基于边缘特征匹配获得的高精度稀疏视差图作为约束, 修正部分误匹配点, 最终获得准确、稠密的视差图。实验表明, 该算法具良好的实验结果和较大的实用价值。

Stereo matching is to obtain an accurate disparity image from two images whose pixels existing stereo disparity between each other. And it is one of the most important and difficult problems in fields of computer vision. After well researching two sorts of typical stereo matching algorithm called stereo matching based on edge features and stereo matching based on the local window respectively, a new stereo matching method based on edge features and the local adaptive supported-weight window is proposed, which aims at getting dense and high-precision disparity image fast and efficiently. The algorithm completes the initial window-based matching by selecting and computing the matching window’s size and weight adaptively. Then depending on the high-precision sparse disparity image acquired by the method based on edge features, the algorithm can eliminate many mismatching points. After these operations, dense and high-precision disparity image is obtained. The final experiments show the method is effective and of great practical value.