A Learning-Based Optimization Algorithm: Image Registration Optimizer Network

Remote sensing image registration is valuable for image-based navigation system despite posing many challenges. As the search space of usually nonconvex, optimization algorithm, which aims to find optimal parameters in space, a challenging step. Conventional algorithms can hardly reconcile contradiction rapid convergence and global optimization. In this letter, novel learning-based algorithm named optimizer network (IRON) proposed, predict optimum straightforwardly. The IRON trained by 3-D tensor ( $9\times 9\,\,\times9$ ) consists similar metric values. Each value corresponds initial parameters’ notation="LaTeX">$9 \times 9 9$ neighbors space. tensor’s label vector points from parameters. Because special design, our could directly. experimental results demonstrate that proposed performs better than other classical as it shows higher accuracy, lower root-of-mean-square error (RMSE), more efficiency. code publicly available at https://www.github.com/jaxwangkd04/IRON .