Skip to Main Content
Accurate registration of multitemporal remote sensing images is essential for various applications. Mutual information has been used as a similarity measure for registration of medical images because of its generality and high accuracy. However, its application in remote sensing is relatively new. In this paper, we introduce a registration algorithm that combines a powerful search strategy; named Simulated Annealing based Marquardt-Levenberg (SA-ML), with mutual information, together with a wavelet-based multiresolution pyramid due to Simoncelli. We consider images, which are misaligned by a three parameter rigid transformation, consisting of rotation and/or x- and y-translations. It is shown that the SA-ML search combined with mutual information produces accurate results and magnificently extends convergence region of Marquardt-Levenberg (ML) method, which previously has been developed for medical data, when applied to synthetic, as well as multitemporal sets of satellite data. We evaluate several hybrid Simoncelli pyramids (low-pass, high-pass) for the best results in terms of accuracy and convergence. It is found that 4-level pyramid SimIB1B2B3 (band-pass pyramid with the original image in the finest level) performs best for multitemporal images.