Registration of Optical and Radar Satellite Images Using Local Features and Non-rigid Geometric Transformations

Abstract

Optical and radar sensors provide us with different physical properties which require to be combined for different remote sensing applications like change detection. In this paper, a feature-based method for registering RapidEye and Pleiades optical images with TerraSAR-X radar images is presented. Different feature detectors/descriptors have been tested during the keypoint extraction and matching processes. In the extraction phase, a keypoint filtering process is proposed in order to control the number of extracted keypoints from both optical and radar images for a real time processing on high resolution satellite images. A speckle noise removal is performed on TerraSAR-X radar images as a pre-processing step in order to reduce the number of false detections. After getting the matches between the input images (list of tie points), a full registration process is followed by testing and applying different non-rigid image transformation models (e.g. affine and polynomial) in order to warp the sensed image on the reference one. The registered version is obtained after resampling both images using interpolation methods. The results showed a good performance for some detectors/descriptors like SURF, GFTT and Harris on optical images and SIFT on radar images. There is also a varying performance of the tested warping and resampling methods in addition to a full registration framework using SIFT and global homography transformation with linear blending. The tested optical and radar images are in different spatial resolutions and covering the area of Berlin Brandenburg airport in Germany.