基于隔离子采样的鲁棒图模型点匹配算法

Graphical model for point pattern matching based on sampling of separator

为解决概率图模型匹配算法对模板点集中的外点敏感的问题,对隔离子进行了采样,并推导出采样次数和模板点集中的外点比例关系,以保证隔离子中的点能最大概率地为内点,使得推理算法中每个点的信息能得到传递。另外,算法中的互匹配解决了传统图模型匹配算法中多对一问题。实验证明,算法相对于原先的JT算法具有很高的鲁棒性和正确率。

The graphical model for point pattern matching in JT algorithm is proved sensitive to the outliers in template pointset recently. To efficiently solve the problem, this paper sampled on the point in the separator, and derived the relationship between the times of the sampling and the ratio of the outliers. So that the message on each point could be delivered in the inference, and achieved finally the global MAP. Experimental results show that the approach improves the accuracy and is more robust than current methods.