利用自适应窗口实现不连续保护立体匹配

摘 要：针对计算机视觉中图像对应点的误匹配问题，提出了一种以扫描线优化为能量最小化框架，以最佳自适应窗口为匹配基元的立体匹配算法。该算法首先采用一种融合亮度和梯度信息的非相似度，然后通过平均匹配误差、匹配误差方差及较大窗口的偏移组成的窗口代价函数选取每个视差的最佳窗口作为匹配基元，并利用扫描线优化法及视差变化伴随亮度变化的约束寻找一条最优的匹配路径，最后通过回溯最佳路径得到稠密的视差图。实验结果表明，所提算法不仅能够保留基于自适应窗口的匹配算法能较好地处理大的低纹理区域和视差不连续区域的优点，而且可以加强相邻匹配基元间的视差连续性，从而获得更平滑、准确的视差图，另外所提算法的计算量比原有算法缩短了约3/ 4 。

Discontinuity preserving stereo matching using variable window

For solving the mismatching problem in computer vision,a discontinuity preserving stereo matching algorithm using variable windows in an energy minimization framework was proposed.Firstly,a dissimilarity combing intensity and gradient information was introduced.Then,according to the cost function composed of the average measurement error,error variances and the error biases of larger windows,the best window for each disparity was chosen as the matching cue,and a scanline optimization method was used to sear...