Optical image and SAR image registration based on linear features and control points

This paper presents a method for automatically registrating multi-sensor images based on linear features and control points. Taking optical images and synthetic aperture radar (SAR) images containing man-made objects as examples, we propose an adaptive multi-scale fast discrete Beamlet transform to extract the common features for the same man-made objects shown in both images, i.e., the salient linear features. Then, we construct the control points based on the linear features and design the matching function based on the control point features. Automatic coarse-to-fine registration between images of different sensors is realized by the feature consensus based coarse registration and the control points based fine registration. The experiment results show that the proposed method has high registration accuracy for the optical image and SAR image which differ in intensity, rotation or translation.