中文摘要：

提出一种立体匹配算法．首先采用均值平移算法将参考图像根据彩色信息快速聚类成不同区域；然后通过灰度差平方和匹配计算初始视差图；在构造能量函数时，将分割结果作为视差函数的一个参考项；最后采用图切割算法求取使全局能量最小的视差最优分配．通过标准图像对进行测试，并与其他算法进行了比较．实验结果表明，与原有算法相比，该算法可有效地处理大的低纹理区域，匹配精度高，更有利于估计视差图的边界．

英文摘要：

A stereo matching algorithm is presented. First, regions of homogeneous color are located by applying the meanshift color image segmentation method. Then the sum of squared differences matching method is used to determine the initial disparity. The segmentation result is used as a reference the disparity function. Finally the energy function is globally minimized using graph cuts. Our proposed algorithm is tested using the stereo test bed and compared with other related methods. The experimental results show that compared with the original algorithm, our improved algorithm has a higher matching accuracy than, a better performance in large areas of low texture, and is more adequate for the estimation of disparity map boundaries.