An equivalence graph cut method based on wavelet transform for SAR image registration

This paper proposes a novel equivalence graph cut method based on wavelet transform for SAR image registration in order to improve robustness and real-time performance. First, the equivalence graph cut model is constructed in low-frequency sub-images after wavelet transform of image, which can reduce speckle noise. The proposed model can not only avoid NP-complete problems but also provide a solution to the choice of mapping function. Then, scale invariant feature transform (SIFT) is exploited to find the feature matching in the object accurately segmented from the original image so as to reduce the feature point description of search space and improve real-time performance. Finally, the accurate SAR image registration is achieved based on the transformation parameters found by matching relationship. The experimental results show that the proposed method can achieve fast and accurate image registration.