SAR image registration based on extremal scattering regions in multiscale framework

This paper propose a novel concept of the Extremal Scattering Regions (ESR) for registration of satellite SAR images without prerequisite acquisition parameters or projection correction. It is implemented in the multiscale framework. First, the ERS is extracted and used for primary registration in the coarse scale. Then, the mutual information is used to fine-tune the registration of the images in the fine scale. The performance of this algorithm is demonstrated in the experiment with TerraSAR-X and Radarsat-2 SAR images, and validated by the results with RMS errors in the range and azimuth directions of 0.338 and 0.134 pixels, respectively.