基于像元集的置信传递立体匹配

Pixel-set Based Stereo Matching by Using Belief Propagation

为了提高立体匹配效率和克服处理区域的视差跳跃,提出了一种基于像元集的置信传递立体匹配方法。该方法首先以像素为基元,利用层次置信传递算法得到较为准确的初始视差;然后依次根据颜色和初始视差对参考图像进行分割,再利用分裂合并策略对分割后的像元集进行平面拟合,以消除颜色分割错误对匹配造成的影响;最后在拟合后的像元集空间,利用标准置信传递优化算法得到最终解。采用国际标准图像进行测试的实验结果表明,该方法的匹配效率和精度优于同类方法。

In order to improve the efficiency of stereo matching and resolve the problem caused by discontinuity of the disparities in one region, new pixel-set based stereo matching algorithm using belief propagation is proposed in this paper. Firstly, the initial disparity estimate is evaluated by hierarchical belief propagation in pixel domain. Secondly, the reference image is segmented in turn according to color and disparity information. Thirdly, the segmented pixel-sets are fitted to a set of planes based on the strategy of splitting and merging, which can eliminate the mistakes due to color segmentation. Finally, the ultimate disparity estimate is achieved in the pixel-set domain by employing standard belief propagation. Experiments on the international benchmark demonstrate that the performance of our algorithm is comparable to the state-of-the-art stereo algorithms on various data sets.