Automatic coarse-to-precise subpixel multi-band SAR images co-registration based affine SIFT and radial base function(RBF)

摘要

Because of the local deformation of SAR images Registration of Multiband SAR images with high accuracy(sub-pixel accuracy) is a very difficult problem. This paper present an automatic Coarse-to-Precise(global-to-local) subpixel registration framework to track it. Firstly, affine SIFT method is adopted to get affine parameter of global transformation between registered images. After global registration, can be obtained, some control points(CPS) arranged in a unform format from common region of two registered images are used for block-matching, and a lot of CPS are obtained. In order to character the deformation of local region, Radial basis functions(RBF) are used to model the local deformation and parameters are estimated. The image are resampled to get registratered image with sub-pixel accuracy.