置信度传播和区域边缘构建的立体匹配算法

Stereo matching using belief propagation and local edge construction-based cost aggregation

讨论了区域匹配代价和全局置信度传播算法之间的相互作用,提出一种基于全局置信度传播和区域边缘构建的立体匹配算法。首先,在每个像素的固定邻域内利用二阶微分边缘算子搜索并构建一条虚拟的封闭边缘,形成相应的区域自适应窗口;然后使用自适应窗口内的支持像素计算中心像素之间的匹配代价;最后利用具备加速消息更新机制的置信度传播算法获取视差。实验结果表明,基于区域边缘构建的匹配代价可以较好地适用于全局置信度传播优化算法,提出的立体匹配算法可以在Middlebury标准下获得良好的匹配结果。

In this paper, the importance of cost aggregation, also called similarity measure,for belief propagation and the interaction of them are discussed. A global stereo matching algorithm is proposed by combining the belief propagation and local edge construction-based cost aggregation. First, a virtual closed edge is formed surrounding each pixel via second derivative operator in order to construct an adaptive window for the centered pixel. Then, the local cost aggregation is calculated on support pixels in an adaptive window. Finally, accelerated belief propagation optimization algorithm is used to obtain the disparity. The experiments based on the Middlebury benchmark indicate that the local edge construction-based cost aggregation can do well with belief propagation optimization and show encouraging results of the proposed stereo matching algorithm.