基于方向相位稠密特征的多传感器遥感影像配准方法和系统

Multisensor remote sensing registration method and system based on dense feature of orientated phase

针对多传感器遥感影像间因显著几何畸变和非线性灰度差异造成的配准困难问题,本文提出了一种基于结构相似性的快速精确配准方法。首先,为构建多传感器影像间的稳健结构特征,本文引入光照和对比度不变性的相位一致性模型,利用相位一致性特征值和特征方向,构建了一种逐像素的三维结构特征描述符—方向相位稠密特征DFOP(Dense Feature of Orientated Phase)。该特征描述符通过捕捉影像的几何结构分布,能够有效抵抗多传感器影像间的灰度差异;其次,基于模板匹配的策略,将DFOP描述符变换到频率域,并进行快速子像素精度匹配,从而研制出了一种快速鲁棒的多传感器遥感影像配准系统;最后,通过利用多种地貌类型的多传感器遥感影像对所提出的方法进行测试。测试结果表明,相较于其他基于灰度或结构特征的匹配方法,本研究提出的DFOP方法能够获得更高的匹配正确率,且所开发的配准系统也优于商业软件ENVI以及ERDAS的配准模块。

To solve the problem of registration difficulty caused by considerable geometric distortion and gray differences between multisensor remote sensing images,this study proposes a fast and accurate registration method based on structural similarity between images.In this method,the phase congruency model with illumination and contrast invariances is introduced to construct robust structural feature descriptors of images.First,the intensity and orientation of phase congruency are used to build a pixel-wise three-dimensional structural feature representation named Dense Feature of Orientated Phase(DFOP),which can effectively resist the grayscale difference between multisensor images by capturing geometric structures of images.Next,the DFOP feature descriptor is transformed into the frequency domain,and the single-step DFT approach is used to achieve fast matching with subpixel accuracy by employing a template matching scheme.In addition,a fast and robust automatic multisensor remote sensing image registration system is developed on the basis of the proposed DFOP.Finally,the proposed method and registration system is validated using multiple pairs of multisensor remote sensing images(including optical,LIDAR,and SAR)covering different scenes.Results show that the proposed DFOP achieves higher correct matching rate,and the developed registration system outperforms the registration module of ENVI and ERDAS in registration accuracy.Our system is available at https://github.com/yeyuanxin110/Remote-Sensing-Image-Registration-system.git.