文献类型：期刊文献

中文题名：尺度不变特征变换法在SAR影像匹配中的应用

英文题名：Application of Scale Invariant Feature Transformation to SAR Imagery Registration

作者：陈尔学[1] 李增元[1] 田昕[1] 李世明[1]

第一作者：陈尔学

通信作者：Chen, E.-X.

机构：[1]中国林业科学研究院资源信息研究所

年份：2008

卷号：34

期号：8

起止页码：861-868

中文期刊名：自动化学报

外文期刊名：Acta Automatica Sinica

收录：CSTPCD;;EI(收录号：20083711541387);Scopus(收录号：2-s2.0-51149095802);北大核心:【北大核心2004】；CSCD:【CSCD2011_2012】；

基金：国家重点基础研究发展计划(973计划)(2007CB714404)资助~~

语种：中文

中文关键词：尺度不变特征变换法;特征点匹配;影像配准;合成孔径雷达

外文关键词：Scale invariant feature transformation （SIFT）, feature point matching, image registration, synthetic aperture radar （SAR）

分类号：TP721.2;TP751.1

摘要：通过几组可代表合成孔径雷达(Synthetic aperture radar,SAR)影像配准主要实际应用情景的实验,对尺度不变特征变换法(Scale invariant feature transformation,SIFT)在SAR图像配准中的应用能力进行了系统的评价.发现SIFT方法可以实现同轨获取的多时相斜距影像之间、斜距与地距影像之间、地距影像与经过地理编码的斜距影像之间的精确配准.为了利用SIFT实现整景遥感影像间的配准,提出了分块处理的方法.实验发现分块寻找特征点虽然可引起特征点总数的降低,但特征点的重复出现率仍大于76%,可满足大影像间配准的需要.同时也发现SIFT匹配过程过于耗时是阻碍其在遥感领域实际和用的技术瓶颈.本文指出了解决该瓶颈的技术方向,并对不变特征匹配法在遥感领域的应用进行了展望.
Some experiments were designed to systematically assess the applicability of the scale invariant feature transformation （SIFT） to synthetic aperture radar （SAR） image registration. The experiments showed that the SIFT can accurately register slant range to slant range, slant range to ground range and ground range to ellipsoid geo-coded SAR images, provided that all the multi-temporal images are acquired in the same ascending or descending orbit. A subset processing based method was suggested to adapt SIFT to the registration of remote sensing images. The experiment showed that although the total number of feature points detected through image subset processing decreases, the feature point repeating rate is still higher than 76 % and can meet the need of the registration of two big images. However, the huge time required for matching was shown to be the major technic bottleneck prohibiting its application to remote sensing. Some possible solutions to this problem were pointed out, and the future applications of invariant feature matching to remote sensing field were suggested.