不同分辨率图像的角点匹配方法  

Method of Matching Images with Different Resolutions

作　　者：周志强[1] 汪渤[1] 吕冀[1] 

机构地区：[1]北京理工大学信息科学技术学院自动控制系,北京100081

出　　处：《北京理工大学学报》2008年第7期598-601,共4页Transactions of Beijing Institute of Technology

基　　金：国家部委预研项目(51405030104BQ0171)

摘　　要：针对不同分辨率图像的匹配问题,提出了一种利用高斯金字塔的一对多角点匹配方法.采用一种变形的特征描述算子(SIFT)对角点附近的图像区域进行特征提取,根据提取出的特征向量进行角点匹配,并提出一种简单有效的抽样一致性算法以剔除错误匹配点对,最后在一对多角点的匹配结果中选出最佳的匹配作为最终的角点匹配结果.实验结果表明,该算法能有效地克服分辨率的差异给匹配带来的困难,算法简单,可靠性好.In order to solve the problem of matching two images with different resolutions, a oneto-many image matching strategy is proposed, employing the Gaussian image pyramid. Modified scale invariant feature transform（SIFT） descriptor is used to extract the image feature of the local region around each corner point, and used to match the corner. A simple and robust method based on sample consensus is presented to eliminate spurious matches. After all the spurious matches are eliminated, the best one-to one image matching result is chosen from all the matching results as the final exact result. Extensive experiments showed that the proposed method is able to match images with different resolutions effectively, and that it is robust and easy to compute.

关 键 词：图像匹配 特征描述算子 高斯金字塔 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]