一种多特征融合的特征匹配算法  

An integrated feature matching algorithm

作　　者：李国祥 王继军[2,3] 马文斌[1,2] LI Guoxiang;WANG Jijun;MA Wenbin(Department of Academic Affairs,Guangxi University of Finance and Economics,Nanning 530003,China;Guangxi Key Lab of Multi-source Information Mining and Security,Guangxi Normal University,Guilin 541004,China;School of Information and Statistics,Guangxi University of Finance and Economics,Nanning 530003,China)

机构地区：[1]广西财经学院教务处,广西南宁530003 [2]广西师范大学广西多源信息挖掘与安全重点实验室,广西桂林541004 [3]广西财经学院信息与统计学院,广西南宁530003

出　　处：《合肥工业大学学报（自然科学版）》2021年第4期484-490,共7页Journal of Hefei University of Technology：Natural Science

基　　金：广西重点研发计划资助项目(2018AB15003);广西多源信息挖掘与安全重点实验室开放基金资助项目(MIMS17-02);广西高校中青年教师基础能力提升资助项目(2021KY0650,2019KY0661);广西跨境电商智能信息处理重点实验室培育基地(广西财经学院)专项资助项目。

摘　　要：特征匹配是目标识别的基础,文章针对特征描述子在多变复杂场景中的自适应问题,从特征向量的低维度、高稳健、易计算3个方面,结合核主成分降维、匹配核的空间映射以及Power-law归一化等特征处理方法,提出一种面向复杂变换环境的集成低维度的特征匹配算法。首先,采用线性内积核将特征向量映射至高维空间提取特征主成分,然后通过Hellinger匹配核完成主成分空间到RootSIFT的映射转换,最后对特征向量进行α中心化和Power-law归一化,形成新的稳健特征向量。实验证明该算法简单易行,特征维度得到了大幅度降低,且在复杂变换的场景中匹配精度优于同类其他算法。The feature matching is the basis of object recognition.In order to improve the adaptability of feature descriptors in complex scenes,this paper proposes a robust feature matching algorithm with low dimension vectors for images matching.Local image descriptors are aggregated into a vector via kernel principal component analysis,matching kernel and Power-law normalization.Firstly,in order to extract feature principal components,the original data is mapped to high-dimensional space by using linear kernel function,then Hellinger matching kernel is used to generate RootSIFT features.Finally,the feature vectors are centralized and normalized to be a new robust vector.Experiments show the superior performance of the proposed aggregation strategy in complex scenes,and the feature dimension is greatly reduced.

关 键 词：特征匹配 核主成分分析 核函数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]