一种基于自适应窗口和图切割的快速立体匹配算法

针对基于图切割的立体匹配算法计算量大的缺点，提出了一种新的快速立体匹配算法。首先根据图像边缘特征自适应变化窗口，并采用灰度差平方和匹配(SSD)作为相似判定准则计算初始视差图，再通过左右一致性校验去除误匹配点，在构造能量函数时，将初始视差作为能量函数的一个参考项，最后采用图切割(graph cuts)算法求取使全局能量最小的视差最优分配。通过标准图像对测试了提出的方法，并与其他方法进行了比较，实验结果表明，该算法不仅能够保留基于图切割的立体匹配算法对大的低纹理区域和遮挡像素较好处理的优点，而且匹配时间短，运行时间比原有算法约缩短了三分之二，能够满足工程实用性的要求。

A Fast Stereo Matching Algorithm Based on Adaptive Window and Graph cuts

According to the disadvantage of the large computation of the stereo matching algorithm based on graph cuts, a new stereo matching algorithm is presented. First, adaptive window is obtained in virtue of image edge features. Second, the SSD matching method as a similarity decision criterion is introduced to determine the initial disparity map, and then left-right conformity verification is adopted to remove mistake matching points. The initial disparity is used as a reference for the function when the energy function is structured. Finally the optimization of disparity distribution is obtained to achieve the minimum global energy by graph cuts algorithm. Experimental results based on standard stereo image rig are compared with other methods and demonstrate that not only the proposed algorithm has better performance of processing the large low texture areas and the occluded pixels, but also running time is short and two-thirds faster than the original algorithm. It can satisfy the demands of engineering practices completely.