Abstract: When the point pattern matching problem is resolved with spectral method, one of the challenges which limit the method is that it is very sensitive to the measurement noise of the points’ position. This paper improves the robustness of the spectral method from three aspects. It adopts the proximity matrix which takes the Gaussian weighted Euclidean distances between any two points within the same point pattern as its elements. A new sign correction method is proposed. The matching metric derived from the spectral method is further weighted by the points’ attribute information. Simulation experiments show that when the improved spectral method is applied to the point pattern matching problem on noisy conditions it can achieve a high correct matching ratio.

Key words: point pattern matching, spectral method, representation matrix, sign correction, matching metric

摘要： 利用谱方法进行点模式匹配的主要问题是对点的位置噪声比较敏感。为了提高谱方法对噪声的鲁棒性，该文在表示矩阵的构建过程中采用高斯加权的近邻矩阵对要匹配的点模式进行描述，提出一种新的符号校正方法，利用点的属性信息对根据谱方法得到的匹配度量进行加权。仿真实验表明，在噪声情况下的点模式匹配应用中采用改进的谱方法可以获得较高的正确匹配率。

关键词: 点模式匹配, 谱方法, 表示矩阵, 符号校正, 匹配度量

CLC Number: 

• TP391