Multi-view SAR image registration based on feature extraction

Abstract

Aiming at the problem of SAR image interpretation of the same target from different angles, the performance of registration algorithms based on different feature operators on multi angle SAR images is compared. Firstly, this paper introduces two kinds of universal registration methods. Combined with the characteristics of multi view SAR images, feature-based matching methods are selected for research; Secondly, sift, brisk and surf features are used as descriptors for feature selection and extraction, and the feature points extracted from SAR images with different viewing angles are matched by Euclidean distance matching method and random sampling consistency algorithm (RANSAC); Finally, the affine transformation model is used to realize the image transformation, and the registration accuracy of the three operators is quantitatively analyzed by calculating the maximum mean square error (RMSE). The results show that the registration algorithm based on surf feature has higher tolerance to scene complexity and stronger robustness,; Brisk based feature is the second; The registration algorithm based on SIFT features is most easily affected by the scene, and the matching effect is the worst.