双目立体匹配是计算机视觉中研究的重点，针对立体匹配在深度不连续、低纹理和场景重复区域容易匹配出错的问题，提出了一种基于图像分割和改进的自适应支撑权重的立体匹配算法。该方法首先根据颜色相似性、欧式距离相似性、梯度相似性和自定义颜色内相关相似性定义初始的匹配代价关系，然后利用mean shift算法分割出不同深度区域的匹配点，根据匹配点所在的深度区域进行匹配代价重定义。在代价聚合的过程中，为了消除光照和噪声的影响，对待匹配点进行rank变换后，再进行视差值的计算，从而得到一个更加准确的视差结果。最后在VS2010软件平台上对Middlebury标准图像进行测试，实验结果表明，该方法得到的视差结果要明显优于现有的局部区域立体匹配方法，且具有很强的稳健性和较高的准确匹配率。

Binocular stereo matching is an important issue in computer vision research. In order to solve the problem of stereo matching at the depth discontinuities, low textured regions and repetitive structures incidental matching error, a stereo matching algorithm based on image segmentation and improved adaptive support weight is proposed. the initial matching cost that combines the color similarity, euclidean distance similarity, user-defined inter color correlation similarity and gradient similarity is defined. The mean shift algorithm segment matching pixel is used at different depth regions in order to refine the matching cost. Meanwhile, in the process of cost aggregation, the new cost aggregation is calculated based on compare transform matching pixel by ranking transform in stereo image pairs in order to solve the influence of brightness and noise difference between the stereo image pairs, so a more accurate disparity result can be acquired. Finally, the algorithm is tested by Middlebury stereo benchmark on the VS2010 software platform, and the results show that it has better performance than other local matching methods, and its robustness is very strong and has higher accurate matching rate.