Optimal Euclidean distance matrix matching method for synthetic aperture radar images

In the application of synthetic aperture radar (SAR) image registration based on the scale invariant feature transform (SIFT) algorithm, a keypoint is always assigned with several dominant orientations. Thought the number of matches is increased, the feature matching performance usually decreases significantly with the infuluence of the feature vectors extracted with different orientations. An optimal Euclidean distance matrix (OEDM) is proposed for two sets of feature vectors to enhance the matching performance. The most similar keypoints are selected from the OEDM. In addition, spatial consistency of the keypoints from the two images is maintained by calculating the transformed distances, and the incorrect matches are eliminated effectively. Comparison with traditional dual matching (DM) methods is performed. The experimental results demonstrate the superiorities of the proposed method in both accuracy and efficiency.