基于结构信息的可见光和SAR图像两阶段配准方法
A Two-Stage Registration Method for Visible Light and SAR Images Based on Structural Information
贾蕾蕾 1史林 1刘利民 1史冬冬1
作者信息
- 1. 陆军工程大学石家庄校区,河北石家庄 050003
- 折叠
摘要
针对可见光和合成孔径雷达(synthetic aperture radar,SAR)图像配准中存在的配准精度低和非线性辐射差异问题,提出了一种新的基于结构信息的可见光和SAR图像两阶段配准算法.在粗配准阶段,利用改进的最大自不相似性检测器(maximal self-dissimilarity detector,MSD)分别检测可见光和SAR图像中的特征点,基于图像局部自相似性提取图像结构信息构建最大自相似性索引图,利用最近邻与次近邻之比和快速抽样一致性识别正确匹配点对,估计变换模型.在精配准阶段,分别建立可见光和SAR图像的像素化结构特征表示,在可见光图像中以特征点为中心选取模板窗口,利用变换模型估计SAR图像中相对应的搜索区域,采用欧氏距离作为模板匹配的相似性度量,并利用快速傅里叶变换(fast Fourier transform,FFT)加速计算模板匹配过程.在9对真实可见光和SAR图像上的实验结果表明,该算法在可见光和SAR图像配准方面具有明显优势.较 RIFT、HAPCG、OSS、LNIFT 和 ASS 算法,正确匹配数(number of correct matches,NCM)分别提高了 7.95、10.86、29.32、14.75 和 9.84 倍,定位精度分别提高了 0.44、0.42、0.35、0.41 和 0.40 pixel.
Abstract
To address the problems of low registration accuracy and nonlinear radiometric differences in the alignment of visible light and Synthetic Aperture Radar(SAR)images,a novel two-stage registration algorithm for visible light and SAR imagery based on structural information is proposed.In the coarse reg-istration stage,the improved maximal self-dissimilarity detector(MSD)is used to detect feature points in visible light and SAR images;then,based on the local self-similarity(LSS),the maximum self-similarity index map(MSSIM)is constructed;finally,the correct matched point pairs are identified using the nea-rest neighbor to second nearest neighbor distance ratio combined with a rapid consensus sampling approach to estimate the transformation model.In the fine matching stage,the pixelwise structural feature repre-sentations of visible light and SAR images are established respectively;then,in visible light images,the template windows centered on feature points are selected,and the transformation model is used to estimate the corresponding search areas in SAR images;finally,Euclidean distance is employed as a similarity measure for template matching,and the Fast Fourier Transform(FFT)is utilized to accelerate the tem-plate matching process.The experimental results on nine pairs of real visible light and SAR images show that the proposed method has significant advantages in the registration of visible light and SAR images,compared with the other methods,such as RIFT,HAPCG,OSS,LNIFT,and ASS methods.The num-ber of correct matches(NCM)has been increased by 7.95,10.86,29.32,14.75,and 9.84 times,respec-tively,and the positioning accuracy has been improved by 0.44,0.42,0.35,0.41,and 0.40 pixels.
关键词
可见光图像/SAR图像/两阶段配准/结构信息/局部自相似性/像素化结构特征表示Key words
visible light images/SAR images/two-stage registration/structural information/local self-similarity(LSS)/pixelwise structural feature representation引用本文复制引用
基金项目
军内科研项目(LJ20232B030191)
出版年
2024
NETL
NSTL
万方数据