局部不变特征匹配的快速内点选择算法  

Fast Inlier Selection Algorithm for Local Invariant Feature Matching

作　　者：肖春宝[1,2] 冯大政 Xiao Chunbao;Feng Dazheng(School of Computer Science and Technology, Xidian University, Xi’an 710071;Information Engineering College, Henan University of Science and Technology, Luoyang 471023;National Laboratory of Radar Signal Processing, Xidian University, Xi’an 710071)

机构地区：[1]西安电子科技大学计算机学院,西安710071 [2]河南科技大学信息工程学院,洛阳471023 [3]西安电子科技大学雷达信号处理国家重点实验室,西安710071

出　　处：《计算机辅助设计与图形学学报》2017年第10期1891-1897,共7页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61271293)

摘　　要：为了快速、准确地从局部不变特征初始匹配中选取内点,提出一种联合利用特征点在位置、尺度和方向上的分布一致性进行内点选择的算法.首先,采用均值漂移算法按视差对特征点对聚类实现粗选;然后找出特征点在类内的K近邻特征,再利用特征的尺度和方向估算每个特征点对所在的二局部图像区域之间的变换参数,并计算K近邻匹配相似度;最后,依据K近邻匹配相似度从粗选获得的候选匹配集中选出内点.实验结果表明,该算法在准确率、召回率和速度方面优于当前最新的内点选择算法,对图像间大幅度的视点、尺度和旋转变化具有较高的鲁棒性.Aiming to quickly and accurately select inliers from initial matching results of local invariant features,an inlier selection algorithm is proposed by jointly exploiting the distribution consistencies of featuresin location,scale and orientation.Firstly,coarse selection is performed by grouping feature correspondences into clusters according to disparities via the Mean-Shift algorithm.Secondly,K-nearest neighbors of each feature are found in the cluster to which it belongs.Then parameters of transformation between the two local image regions where each pair of features are located are estimated by feature scales and orientations,and a K-nearest neighbor matching similarity is calculated.Finally,according to K-nearest neighbor matching similarities,inliers are selected from the candidate matching set obtained by coarse selection.Experimental results show that the proposed algorithm is superior to the state-of-the-art inlier selection algorithms in terms of precision,recall and speed,and is robust to large changes between images in viewpoint,scale and rotation.

关 键 词：特征匹配 局部不变特征 内点 聚类 K近邻 相似度 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]