计算机工程与应用 ›› 2012, Vol. 48 ›› Issue (36): 171-176.

• 图形、图像、模式识别 • 上一篇    下一篇

基于SIFT特征的异源遥感影像匹配方法研究

吕倩利1,邵永社1,2   

  1. 1.同济大学 测量与国土信息工程系,上海 200092
    2.中国科学院 光电研究院,北京 100094
  • 出版日期:2012-12-21 发布日期:2012-12-21

Research on matching algorithm for multi-source remote sensing images based on SIFT features

LV Qianli1, SHAO Yongshe1,2   

  1. 1.Department of Surveying and Geo-informatics, Tongji University, Shanghai 200092, China
    2.Academy of Opto-Electronics, Chinese Academy of Sciences, Beijing 100094, China
  • Online:2012-12-21 Published:2012-12-21

摘要: 由于不同传感器、多时相、多分辨率、多波段的遥感图像的光谱特征、空间特征、纹理特征等存在较大差异,为影像匹配带来了困难。针对异源遥感影像成像机理的不同特点,从影像特征角度,引入尺度不变特征变换(Scale-Invariant-Feature-Transform,SIFT)方法,实现光学影像、SAR影像和多光谱影像间的匹配;针对SIFT单向匹配算法的不足,引入匹配约束,采用双向匹配策略对其优化,提高了匹配的可靠性。实验表明,该算法具有稳定、可靠、快速等特点,适用于存在光谱特征、空间特征、纹理特征等差异的异源遥感影像的高精度匹配。

关键词: 合成孔径雷达(SAR)影像, 多光谱影像, 尺度不变特征变换(SIFT)特征, 异源影像匹配

Abstract: Since multi-source, multi-temporal, multi-resolution and multi-band remote sensing images are too different in spectral characteristics, spatial characteristics as well as texture features, it is full of difficulty to match these remote sensing images. According to the different characteristic of imaging mechanism for multi-source remote sensing images, a new matching algorithm based on Scale Invariant Features Transform(SIFT) features is proposed from the perspective of image features, to match these remote sensing images, among optical images, SAR images and multispectral images. The match constraint is proposed to make up for the insufficient of SIFT based unidirectional match algorithm. And a bidirectional matching strategy on its optimization is used to improve the matching reliability. The experimental results demonstrate that this approach is robust, reliable, fast and efficient for the high-precision matching among the multi-source remote sensing images which are too different in spectral characteristics, spatial characteristics, as well as texture features.

Key words: Synthetic Aperture Radar(SAR) image, multispectral image, Scale Invariant Features Transform(SIFT) feature, multi-source image match