一种基于改进SIFT算法的遥感影像配准方法  

A registration method of RS image based on improved SIFT algorithm

作　　者：雷小群[1] 李芳芳[2] 肖本林[3] 

机构地区：[1]湖南省第一测绘院,湖南衡阳421001 [2]武汉大学遥感信息工程学院,武汉430079 [3]湖北工业大学土木工程与建筑学院,武汉430068

出　　处：《测绘科学》2010年第3期143-145,共3页Science of Surveying and Mapping

基　　金：湖北省自然科学基金资助项目(2008CDB388)

摘　　要：本文首先分析了传统SIFT算法在遥感影像配准应用中存在的不足,从特征点匹配精度上进行了改进。通过分析SIFT特征点匹配的主要误差来源,逐步消除可能误差,提取尽可能多且精准的匹配点对。接下来利用该匹配点对作为配准用控制点对,分别对不同时相、不同分辨率遥感影像进行仿射变换和小面元微分纠正配准。最后与传统SIFT算法进行比较,试验结果表明本文算法具有更高的匹配精度。This paper firstly analyzed the limitations of traditional SIFT algorithm in remote sensing image registration, and improved the algorithm on the accuracy of feature points matching. By analyzing the main sources of error in SIFT feature points matching, possible error pairs were eliminated gradually, and precise pairs were extracted as much as possible. Then using these matched points as registration control points, affine transformation and tiny facet primitive rectifying were carried out on different time and different resolution images. Experimental result showed that this algorithm can get higher accuracy in registration.

关 键 词：SIFT算法 小面元微分纠正 影像配准 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]