基于形状内容描述子的点特征匹配  

Point Feature Matching Based on Shape Content Descriptor

作　　者：冯晓伟[1] 田裕鹏[1] 

机构地区：[1]南京航空航天大学自动化学院,南京210016

出　　处：《光电工程》2008年第3期108-111,116,共5页Opto-Electronic Engineering

摘　　要：针对两幅图像中特征点的匹配问题,本文提出了一种新的基于形状内容描述子的点特征匹配方法。该方法首先利用基于曲率尺度空间(CSS)的角点检测技术获得两幅图像中的角点及其所在的曲线;然后,计算两幅图像中每个角点的形状内容描述子,运用χ2统计检验函数得到描述子的匹配度,对该匹配度进行评估,如果高于某一个阈值,则认为初始匹配成功;最后对初始匹配成功的点对,利用半局域限制完成点集之间的最终匹配。实验结果表明,本文所提出的匹配算法具有较高的点特征匹配准确率。In order to solve the matching problems of feature points in two images, we introduced a new point feature matching method based on a new shape content descriptor. Firstly, the comers and the curves relative to each comer were obtained using the Curvature Scale Space （CSS） technology. Secondly, the shape content descriptor of each comer was computed, and the matching degree was gained by χ2 distance between the two descriptors. The matching was successful if the degree was bigger than a certain threshold, so the initial set of matches was obtained. Finally, more accurate matches were eventually found by using the semilocal constraints to the initial set of matches. The experimental results show that a very high correction matching rate can be obtained by the algorithm.

关 键 词：CSS角点提取算子 点特征匹配 形状内容描述子 半局域限制 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]