基于动态规划的相位匹配和遮挡检测

Phase Matching and Occlusion Detection Based on Dynamic Programming

本文摒弃相位匹配算法中通常运用的相位差一频率法以及相位相关法,直接利用局部相位结构在对应点位置上必须相等的事实,提出以两匹配点处局域窗内的相位差异来衡量匹配误差的代价函数,并结合动态规划方法搜索满足图像对之间局部结构差异全局最小的视差结果.此外,该算法在最优视差搜索过程中将相位奇点问题以及遮挡检测同时处理,而无需检测和剔除奇点或将遮挡检测后续处理.实验结果的分析表明了该算法具有较高的实用性.

Not using phase-difference methods or phase correlation methods which are popular to gain the disparity map in phase-based stereo, the approach presented in this paper directly uses the fact that the phase structures are equal at the correspondence locations. A cost function is defined to measure the matching error between point x in the left view and the one shifted by the assumed disparity value in the right view as the aggregation of the phase differences between their local windows. To complete a globally optimized search process, dynamic programming technique is employed to find the disparity results which minimize the differences between the local structures in two views. In addition, this approach automatically tackles the singularity problem and the occlusion detection problem in the optimal disparity search process, instead of treating them as secondary processes that are postponed until matching is completed. The experiment results demonstrate that this approach has good applicability.