SIFT改进算法及其在遥感中的应用  

Improved SIFT algorithm and its application in remote sensing

作　　者：肖浩威 邵振峰[1] 李从敏[1] 

机构地区：[1]武汉大学测绘遥感信息工程国家重点实验室,湖北武汉430079

出　　处：《计算机工程与设计》2016年第8期2077-2081,2189,共6页Computer Engineering and Design

基　　金：重大设备专项基金项目(2012YQ1601850);国家科技支撑计划基金项目(2013BAH42F03);湖北省杰出青年基金项目(2013CFA024);武汉大学拔尖创新人才基金项目(2042014kf0212)

摘　　要：针对SIFT算法结果中存在错误匹配的现象,分析错误产生的原因后,提出一种SIFT改进算法。利用SIFT提取特征点并进行初始匹配,基于封闭图形进行仿射变换时面积比不变性,从初始匹配中提取若干个局部最优匹配,构建局部最优匹配中特征点与其它待研究匹配中特征点之间的几何关系并进行分析,利用分析结果判断该匹配的优劣,剔除判断为错误的匹配,提高匹配精度,即正确匹配数量占总匹配数量的比例。实验结果表明,该方法有效提高了SIFT的匹配精度,改善了匹配效果。For the phenomenon that mismatching still exists in the result of SIFT algorithm,an improved SIFT method was proposed based on the analysis of the causes of the mismatching.Keypoints were detected and matched into pairs using SIFT,and several local optimal matching pairs were selected on the basis of the theorem that the area ratios of enclosed figures in 2Dremain invariant when conducting affine transformation.The geometry relationship between the keypoints in the selected local optimal matching pairs and the keypoints in the pairs that about to be evaluated was constructed.By analyzing the geometry relationship,the correctness of the pair was judged.The results of experiments show that the proposed method can significantly improve the accuracy of SIFT,thus leading to better performance on matching results.

关 键 词：特征匹配 SIFT算法 尺度空间理论 遥感影像 几何关系 面积比优化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]