基于改进SIFT的SAR图像与可见光图像配准

Registration Algorithm for SAR and Optical Images Based on Improved SIFT

针对SAR图像与可见光图像的自动配准问题,提出了一种基于尺度不变特征(SIFT)的SAR与可见光图像配准算法。算法首先对SIFT主方向的检测进行了优化,利用特征点邻域内边缘局部极大值点的梯度方向作为特征点的方向,并以去均值归一化互相关系数为相似性度量进行特征点对匹配,然后通过随机抽样一致性算法(RANSAC)剔除误匹配点对,最后利用剩余的特征点对实现SAR与可见光图像的自动配准。实验结果表明,本算法对不同分辨率图像和不同旋转角度图像具有较好的适用性,在正确匹配点的比率和定位精度方面都优于原始SIFT算法和Harris算法。

Focusing on the automatic image registration problem of SAR and optical images,a new algorithm based on SIFT is presented in this paper.Firstly,the detection of SIFT dominant direction is optimized,in which the gradient directions of local maxima edge points are used as the directions of key point.Secondly,the normalized cross-correlated algorithm is used in the features matching.Thirdly,the RANSAC algorithm is applied to remove false matching points.Finally,according to the correct matching points,the images are registered automatically.In the experiments,the adaptability of the algorithm is analyzed for the images of different resolutions and different rotating angles.The experiment results show that the proposed method is better than SIFT and Harris algorithms in the correct matching probability and features matching precision,and also presents the applicability.