An optimization algorithm based on Bi-SIFT for SAR image registration

Registration is an effective pre-processing for amounts of applications like image mosaic, mapping, remote detection, and change detection. The scale-invariant feature transform (SIFT) and block matching are two widely algorithms for registration in computer vision. However, as a result of speckle noises, the two algorithms perform unsatisfactorily on synthetic aperture radar (SAR) images. In this paper, we introduce an algorithm, which is named Bi-SIFT optimization algorithm, combing SIFT in two different scales and optimization model in block matching. Bi-SIFT ensures enough correct matching pairs rather than avoiding mismatching-pairs, which is used in traditional ways, to make it more effective. We illustrate the improvement brought by Bi-SIFT by both qualitative analysis and comparing test. Besides, we present serious registration applications of Bi-SIFT.