基于双向邻域过滤策略的图形匹配类遥感图像配准算法(英文)  

A graph matching algorithm based on filtering strategy of Bi-directional K-Nearest-Neighbors

作　　者：赵明[1,2] 安博文[1,2] 王天真[1] 许媛媛[1] 林长青[2] 孙胜利[2] 

机构地区：[1]上海海事大学,上海200135 [2]中国科学院上海技术物理研究所,上海200083

出　　处：《红外与毫米波学报》2014年第1期78-83,89,共7页Journal of Infrared and Millimeter Waves

基　　金：Supported by National Natural Science Foundation of China(61302132.61171126,61203089);Shanghai Municipal Natural Science Foundation(11ZR1415200,13ZR1418900)

摘　　要：针对遥感图像由于较大仿射变换关系、相似图案和多源性等导致图形匹配时出现伪同构现象,提出了一种基于双向邻域过滤策略的图形匹配方法.本方法采用双向邻域的图形特征描述子来表示特征点的邻域关系.当误配点的双向邻域任意顶点在后期迭代中被视为误配点时,将与匹配点集具有稳定双向邻域结构的点恢复至匹配点集,同时剔除伪同构中残留的误配点.通过与Random Sample Consensus(RANSAC)、Graphing Transformation Matching(GTM)算法以及提出的双向邻域匹配方式比较得出,基于双向邻域过滤策略的匹配方式能够处理空间顺序匹配时存在的伪同构问题,同时获得更高的召回率和匹配率.In this paper, a novel graph matching algorithm, called Filtering Bi-directional K-Nearest-Neighbors Strategy （Filtering BiKNN Strategy） is presented to solve the pseudo isomorphic graph matching for remote sensing images with large affine transformation, similar patterns or from muhisource sensors. BiKNN was proposed to describe the adjacent relation- ships of feature points. Filtering strategy is used to eliminate dubious matches of pseudo isomorphism for restrict con- straints. Any BiKNN vertices of candidate outliers treated as outliers in latter iterations are rechecked with the expanded BiKNN respectively. Candidate outliers with stable graph structures are recovered to the residual sets. Three typical remote sensing images and twenty image pairs were utilized to evaluate the performance. Compared with random sample consensus （RANSAC）, graphing transformation matching （GTM） and the proposed BiKNN matching, Filtering BiKNN Strategy can deal with pseudo isomorphism and obtain the highest recall and precision.

关 键 词：图像配准 图形匹配 遥感图像 伪同构 

分 类 号：TP751.41[自动化与计算机技术—检测技术与自动化装置]