基于加权形状上下文的图匹配方法  

Image matching method based on weighted shape context

作　　者：宋琳[1] 程咏梅[1] 宋春华[1] 刘夏雷[1] 

机构地区：[1]西北工业大学自动化学院,陕西西安710072

出　　处：《激光与红外》2013年第7期826-830,共5页Laser & Infrared

基　　金：航空科学基金(No.20100853010)资助课题

摘　　要：针对目前图像发生较大形变时特征点无法精确匹配的问题,提出整数约束下基于加权形状上下文的图匹配方法。利用图理论来描述图像特征间的结构信息,将图像特征匹配转化为图匹配问题。基于相似性度量矩阵构造思想,引入形状上下文特征,以图间所有可能的匹配点对作为分配图的顶点,利用图像点集内和点集间的加权形状上下文统计特征作为相似性度量函数构造图的边的权值,采用整数约束下的迭代求解算法求解图匹配问题。真实图像特征点集的匹配实验验证了本文算法的有效性。A Aiming at the problem of poor accuracy of feature points matching in images with large deformation, an im- age matching method based on weighted shape context under integer constraints is proposed. As graph theory is used to describe the structure information between image characteristics,feature points matching is converted to a problem of graph matching. Shape context is introduced based on the construction of similarity measuring. Possible matching points between graphs are used as nodes of an assignment graph, and the statistical characteristics of weighted shape context within graph point sets and between graph point sets are used as similarity measure function to construct weights of graph edge. At last, the graph matching problem is resolved by iterative solving method under the integer constraints. Experiments on real image feature points matching demonstrate that the proposed method is effective.

关 键 词：图匹配 加权形状上下文 整数约束 迭代求解 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]