Hierarchical approach for synthetic aperture radar and optical image coregistration using local and global geometric relationship of invariant features
Hierarchical approach for synthetic aperture radar and optical image coregistration using local and global geometric relationship of invariant features
复制标题
DOI：
10.1117/1.jrs.11.015002
复制
发表时间：
2017-01
影响因子：
1.7
通讯作者：
Mehdi Salehpour;A. Behrad
中科院分区：
文献类型：
--
作者：
Mehdi Salehpour;A. Behrad
抽象的。针对合成孔径雷达(SAR)图像与光学图像的配准问题，提出了一种分层配准方法。我们使用二进制稳健不变可伸缩关键点(BRISK)算法提取的局部不变显著点用于SAR和光学图像的联合配准。然而，由于合成孔径雷达图像中的相干斑噪声以及合成孔径雷达图像和光学图像的不同结构，使得匹配点具有很高的误差。为此，采用自适应椭圆双边滤波法去除相干斑噪声。此外，利用活跃特征的局部和全局几何关系，采用分层方法进行联合配准。对于光学图像中的每个显著点，在SAR图像中找到三个最接近的匹配点。在突出点和匹配点周围的局部区域确定匹配点的几何关系，并去除几何匹配分数较少的匹配对。在算法的最后阶段，利用稳健的统计方法得到光学图像和合成孔径雷达图像之间的投影全局模型，并对剩余的假匹配进行细化。实验结果和与已有方法的结果比较表明，该算法具有更高的效率。
Abstract. A hierarchical method is proposed for synthetic aperture radar (SAR) and optical image coregistration. We use local invariant salient points extracted by the binary robust invariant scalable keypoints (BRISK) algorithm for SAR and optical image coregistration. However, the matched points are highly erroneous because of the speckle noise in SAR images and the different structures of SAR and optical images. Therefore, an adaptive and elliptical bilateral filter is used to remove the speckle noise. Additionally, a hierarchical approach is used for coregistration using the local and global geometrical relationship of BRISK features. For each salient point in the optical image, three closest matched points are found in the SAR image. The geometrical relationship of the matched points is determined in the local areas around the salient and matched points, and matched pairs with fewer geometrical matching scores are removed. At the final stage of the algorithm, the projective global model between optical and SAR images is obtained using a robust statistic and the remaining false matches are refined. Experimental results and the comparison of the results of the proposed algorithm with those of the existing approaches show that the proposed algorithm is more efficient.