A new automatic registration method for InSAR image based on multi-step strategy

Interferometric Synthetic Aperture Radar (InSAR) technology has been widely used in various applications. The registration of SAR (Synthetic Aperture Radar) images is the first step in interferometric processing therefore accurate registration is essential for the successful creation and interpretation of interferometric products. However, with the growing number of SAR satellite launch and the amount of data acquisition, the degree of automation of image registration have become increasingly demanding. In this paper, we propose an automatic registration approach based on multi-step matching strategy. In the first step, key points are detected and matched using modified scale invariant feature transform (SIFT) operator which modified by us reducing the influence of speckles. In this step, owing to the existing of speckle and the defect of matching strategy of SIFT operator, the expected level of matching accuracy is about 2 to 3 pixels. In the second step, correlation matching (CM) is used to exclude the matched points with low correlation. In the third step, the probability relaxation (PR) algorithm based on global matching is used to induce consistency constraint and ensure reliability of the matching result. Finally, corresponding transformation function is determined through the relationship established by matched point pairs. In order to verify the applicability of proposed methodology, two SAR images acquired over mountainous regions are used in our experiment. The experiment results show that subpixel registration accuracy and good efficiency have been achieved, which demonstrates the correctness and feasibility of proposed method.