An algorithm for automatically estimating the spatial misregistration between two synthetic aperture radar (SAR) images of the same scene is described. This algorithm consists of a multi-resolution hierarchical search which drives a multi-subarea cross-correlation of the logarithm of the SAR image intensities. The multi-resolution/multiple subarea correlation approach is suitable for removing very large translational uncertainties and minor rotational uncertainties in SAR image coordinates due to inertial navigation system errors. The resulting algorithm is potentially highly reliable, highly accurate, and computationally efficient. The organization of this paper is as follows. First, the general problem of image registration is introduced. Second, existing techniques for image registration are surveyed. Third, the multi-resolution/multi-subarea correlation approach to SAR image registration is discussed. Finally, experimental results for a two-resolution version of the algorithm are presented.
Robert T. Frankot,
"SAR Image Registration By Multi-Resolution Correlation", Proc. SPIE 0432, Applications of Digital Image Processing VI, (9 January 1984); doi: 10.1117/12.936661; https://doi.org/10.1117/12.936661