Abstract: A novel cross-scale cost aggregation framework with dynamic support windows is introduced to address the fundamental challenge of local stereo matching algorithms, accuracy and computational complexity dilemma. The matching cost volume is aggregated at each scale separately with box filtering whose support windows are dynamic inter-scales. An inter-scale regularizer is introduced into optimization and solving this new cross-scale cost aggregation problem. Weighted median filtering is used for disparity refinement with guided filter weight. Experimental results show that the proposed method achieves theoretically strong results effectively and efficiently, with the two main steps’ complexity independent of the filtering kernel size. The algorithm achieves good performance in terms of both speed and accuracy.

Key words: stereo matching, scale, support window, box filter, guided filter, weighted median filter

摘要： 针对当前局部算法在速度和性能上不能兼顾的问题，提出一种跨尺度变窗口代价聚合的快速立体匹配方法。在图像各内尺度的匹配代价卷用动态支持窗口的盒滤波聚合匹配代价，采取互尺度正则化方法跨尺度聚合匹配代价，利用基于引导滤波权重的加权中值滤波进行视差精化。实验结果表明：该方法匹配精度高，代价聚合与视差精化步骤的时间复杂度都与滤波窗口半径大小无关，在速度和精度上都取得了良好的效果。

关键词: 立体匹配, 尺度, 支持窗口, 盒滤波, 引导滤波, 加权中值滤波