A hybrid SAR image registration algorithm base on SURF and mutual information

摘要

Aim at the problems of features points is difficult to extract, image deformation is difficult to estimate and low registration accuracy in synthetic aperture radar (SAR) image. This paper present a hybrid SAR image registration algorithm base on speeded up robust features (SURF) and mutual information. The hybrid registration algorithm consists of coarse registration and fine registration, respectively. In the coarse registration stage, use SURF algorithm finds the regional extreme value feature points, due to the SURF algorithm ignore structural features, we use Harris corner detection algorithm to extract corner feature, thus extend the feature points to improve the accuracy of coarse registration. Acquire image transform parameters by affine transformation model. In the fine registration stage, through maximize mutual information (MI) to complete the further registration, acquire the more accuracy of image transform parameters, and achieve the high accuracy SAR image registration. The experimental results shows that the method presented in this paper can improve the accuracy of the algorithm.