基于松弛标签的MR影像特征点匹配研究  

Feature points matching of MR images based on relaxation labeling

作　　者：陈佳 范波 徐源泉 CHEN Jia;FAN Bo;XU Yuanquan(School of Mathematics and Computer Science,Wuhan Textile University,Wuhan 430072,China;Office of Scientific Research and Development,Wuhan University,Wuhan 430072,China;School of Remote Sensing and In{ormation Engineering,Wuhan University,Wuhan 430079,China)

机构地区：[1]武汉纺织大学数学与计算机学院,湖北武汉430072 [2]武汉大学科学技术发展研究院,湖北武汉430072 [3]武汉大学遥感信息工程学院,湖北武汉430079

出　　处：《武汉大学学报（工学版）》2018年第9期837-841,共5页Engineering Journal of Wuhan University

基　　金：国家档案局科技计划项目(编号:No.2014-X-68)

摘　　要：针对非均质柔性生物组织发生非线性形变时很难提取大量分布均匀的特征点并对其进行高精度的匹配问题,提出一种基于局部邻域保持拓扑结构的图像匹配方法 SPRL.该方法将特征点的邻域结构与特征描述向量相结合构造初始匹配概率矩阵,并通过松弛标签法得到最优的匹配概率矩阵,最终达到最优匹配结构.实验结果表明,该算法具有更高的匹配率.The existing medical image registration method is difficult to extract a large number of feature points and make the accurate matching of the nonrigid flexible biological tissues.This paper proposes an image matching method,named SPRL.First,Difference of Guassians（DoG）is used to get many feature points;and then we combine the neighborhood structure of feature points with the feature description vector,and then the optimal matching probability matrix is obtained by relaxation labeling method;finally achieve the optimal matching structure.Experimental results show that the algorithm has a higher matching rate.

关 键 词：核磁共振图像 特征点匹配 非线性形变 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]