基于相位特征的可见光和SAR遥感图像自动配准

Automatic registration of optical and SAR remote sensing image based on phase feature

针对可见光和SAR遥感图像存在非线性辐射差异和几何差异,加之SAR的斑点噪声,使得可见光和SAR图像配准十分困难的问题。本文提出了一种基于改进相位一致性的可见光和SAR图像配准方法。首先,分别计算相位一致性的最大矩和最小矩,将二者叠加,利用Harris算子在叠加图上提取特征点,得到稳定的角点和边缘点作为待匹配的特征点;接着,分别构建相位一致性的方向图和基于多尺度融合的最大幅值索引图,借助于(Histogram of Oriented Gradi⁃ents,HOG)模板,利用相位一致性方向对基于多尺度融合的最大幅值索引图进行投票,建立一种新颖的局部特征描述符;最后,利用欧式距离作为特征向量的度量,计算最近邻比率实现特征匹配,采用快速采样一致性算法剔除误匹配点。在四组图像数据上的实验结果表明,本算法相比于基于梯度的OS-SIFT算法具有更多的正确匹配点对和更高的匹配精度,正确匹配点数分别提高了11,8,15和11对,均方根误分别提升了57.5%,57.9%,23.5%和58%。

Optical SAR image registration is highly difficult because of the geometric and nonlinear radia⁃tion differences between optical and SAR remote sensing images,as well as the speckle noise of SAR.Thus,this paper proposes an automatic algorithm,based on phase congruency,to register optical and SAR images.First,the maximum and minimum moments of phase consistency are calculated,and the re⁃sults are superimposed.The feature points are extracted from the superimposed image by the Harris opera⁃tor,and then,the stable corner points and edge points are obtained as the feature points to be matched.Subsequently,the phase-consistent orientation and the maximum amplitude index map,based on multiscale fusion,are constructed with the help of the HOG template.The maximum amplitude index map based on multi-scale fusion is voted using the phase consistency orientation,and a novel local feature de⁃scriptor is established.Finally,Euclidean distance is used as the measure of the feature vector,the nearest neighbor ratio is calculated to realize feature matching,and the fast sampling consistency algorithm is used to eliminate mismatched points.Experimental results on three sets of image data show that the proposed algorithm has more correct matching points and a higher matching accuracy than the gradient-based OSSIFT algorithm.The number of correct matching points is increased by 11,8,15 and 11 pairs,and the root mean square error is increased by 57.5%,57.9%,23.5%and 58%,respectively.