计算机工程与应用 ›› 2010, Vol. 46 ›› Issue (23): 8-11.DOI: 10.3778/j.issn.1002-8331.2010.23.002

• 博士论坛 • 上一篇    下一篇

结合不同光学影像结构信息的SIFT特征配准法

王瑞瑞1,3,马建文2,陈 雪1   

  1. 1.中国科学院 遥感应用研究所,北京 100101
    2.中国科学院 对地观测与数字地球科学中心,北京 100101
    3.中国科学院 研究生院,北京 100049
  • 收稿日期:2010-03-04 修回日期:2010-06-12 出版日期:2010-08-11 发布日期:2010-08-11
  • 通讯作者: 王瑞瑞

Automatic registration for different optical images based on SIFT algorithm combined with structural information

WANG Rui-rui1,3,MA Jian-wen2,CHEN Xue1   

  1. 1.Institute of Remote Sensing Applications,Chinese Academy of Sciences,Beijing 100101,China
    2.Center for Earth Observation and Digital Earth,Chinese Academy of Sciences,Beijing 100101,China
    3.School of Graduate,Chinese Academy of Sciences,Beijing 100049,China
  • Received:2010-03-04 Revised:2010-06-12 Online:2010-08-11 Published:2010-08-11
  • Contact: WANG Rui-rui

摘要: 由不同传感器摄取的遥感影像由于成像模式、拍摄角度和分辨率不同,给两者之间的配准造成了很大的困难。针对这个问题,提出了基于结构信息的SIFT特征配准法,首先提取具有角度和尺度不变性的SIFT特征点,对其进行归一化处理,降低了不同光学传感器遥感影像色调差异大的影响;然后通过对SIFT匹配点对结构信息的一致性检验,增强了算法的鲁棒性;最后结合最小二乘法实现自动配准。选取了角度和尺度偏差较大的SPOT-5(Pan)与ASTER影像、SPOT-5(XS/XI)和TM影像两组数据进行实验。实验结果证明该算法对配准影像在角度、尺度和色调上的偏差具有较强的鲁棒性,可以取得较高的配准精度。

关键词: 光学遥感, 不同光学影像自动配准, 结构信息, SIFT特征

Abstract: The imaging model,acquired angle and resolution between imagery acquired by diverse remote sensing sensors are different,which throws a big difficulty in the registration between them.Aiming to this problem,the SIFT algorithm based on structural information is proposed.Firstly the SIFT descriptors which are robust to differences in angle and scale are normalized,which can reduce the impact of hue difference between imagery acquired by different sensors,and then the consistency check of structural information of SIFT matching couples improves the robustness of this algorithm,lastly with the help of least square method,the automatic registration is achieved.Two groups of images SPOT-5(Pan)and ASTER,SPOT-5(XS/XI)and TM,which have big differences in angle and scale,are tested.A mosaic is formed from the reference and registered sensed images.In the mosaic,lineal features are well-aligned,which effectively proves that this automatic registration algorithm is robust and has a high accuracy.

Key words: optical remote sensing, automatic registration of remote sensing imagery from different optical sensors, normalized SIFT descriptor, structural information

中图分类号: