基于最大特征点对互信息的图像配准  

Image registration based on maximization of feature point pair mutual information

作　　者：廖勇军[1] 杨丹[1] 张小洪[1] 

机构地区：[1]重庆大学软件学院,重庆400044

出　　处：《计算机应用研究》2008年第3期939-941,944,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(60604007);重庆市自然科学基金资助项目(CSTC2005BA2002)

摘　　要：定义了局部特征点对互信息量,并在两个合理的假设下提出了基于最大特征点对互信息量的一种新的图像配准方法。该方法包含了两个搜索过程:第一个过程是基于特征点对的全局搜索过程,该过程得到初始匹配点对;第二个过程是在初始匹配点对的邻域内利用互信息基本性质的局部搜索,该过程可以进一步提高匹配点对的精度。全局搜索过程能够提高搜索效率,而局部搜索又可以克服依赖于准确提取特征点的不足。最后通过理论分析和实验证明了所提出方法的有效性。This paper defined the concept of feature point pair mutual information, and presented a novel image registration approach based on maximization of feature point pair mutual information under two rational assumptions. The presented approach was composed of two procedures. One was the global search procedure according to feature point pairs, where the initial match points were obtained. The other was the local search procedure in small neighborhood of original match points, in which the locations of the initial match points were refined. The global procedure could improve the efficiency of search and the local one could overcome the drawback of depending on accurately extracting feature points. Finally, the effectiveness of proposed approach was demonstrated by the theoretical analysis and the experiments.

关 键 词：互信息 特征点对 图像配准 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]