Simultaneous Perturbation Stochastic Approximation Algorithm for Automated Image Registration Optimization

Automated intensity-based image registration approaches become popular and urgent when facing today´s increasing data mining and frequent fusion demands. As a core part of an automated image registration system, many kinds of gradient-based optimizers were proposed in the past decade. In this paper, a local gradient-free optimizer, namely the simultaneous perturbation stochastic approximation (SPSA) algorithm, was firstly applied for the automated multi-source image registration using the mutual information as a similarity measure. Results of rigid experiments on the image pairs of ASTER-ASTER, ASTER-Map and SAR-SAR showed the SPSA optimizer has much more flexibility and efficiency than the traditional gradient ascent optimizer. It is more suitable as a local optimizer to the automated image registration system. The main shortcoming of this algorithm is too many control parameters needed during the execution process.