一种高效的图像局部特征匹配算法  

An Efficient Algorithm for Local Feature Matching

作　　者：杨恒[1] 王庆[1] 

机构地区：[1]西北工业大学计算机学院,陕西西安710072

出　　处：《西北工业大学学报》2010年第2期291-297,共7页Journal of Northwestern Polytechnical University

基　　金：国家"863"项目(2007AA01Z314);国家自然科学基金(60873085);教育部"新世纪优秀人才"计划(NCET-06-0882)资助

摘　　要：文章提出了一种高效的图像局部特征匹配算法。在特征描述子构建阶段,提出基于梯度的距离和方向直方图(gradient distance and orientation histogram,GDOH)算法,其特征向量维数仅是SIFT和GLOH描述子的一半,然而却具有与SIFT和GLOH相当的性能;在高维特征空间最近邻搜索阶段,提出基于子向量的索引结构(indexing sub-vectors,ISV),ISV算法比BBF(Best Bin First)算法具有更高的搜索精度和更快的搜索速度。实验结果证明文中提出的图像局部特征匹配算法(GDOH+ISV)比目前广泛使用的Lowe的算法[12](SIFT+BBF)更加高效。Aim.In our opinion,we can devise a local feature matching algorithm that is GDOH（gradient distance and orientation histogram） combined with ISV（indexing sub-vectors） and is more efficient than the widely-used algorithm that is SIFT（scale invariant feature transform） combined with BBF（best bin first） and is proposed by D.G.Lowe in Ref.12.Section 1 of the full paper explains the local feature descriptor that is the core of GDOH.Section 2 presents a procedure that consists of three steps： establishment of ISV structure,feature searching and feature matching.Section 3 presents the experimental results and their analysis.Fig.5 in subsection 3.1 compares the matching performance of GDOH descriptor with that of SIFT descriptor.Table 1 in subsubsection 3.2.1 compares the searching precision and searching time of ISV algorithm respectively with those of BBF algorithm.Fig.6 in subsection 3.3 compares the matching performance of GDOH＋ISV algorithm with that of SIFT＋BBF algorithm.The experimental results and their analysis show preliminarily that our GDOH＋ISV algorithm is indeed more efficient than the SIFT＋BBF algorithm proposed by D.G.Lowe.

关 键 词：计算机视觉 局部特征 GDOH ISV 特征搜索 特征匹配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]