An Algorithm for Object Function Optimization in Mutual Information-Based Image Registration

Image registration is widely used in remote sensing image processing, medical image analysis and many other fields. Mutual information-based image registration technique gained much attention recently. However, during the optimization procedure, because of the influence of interpolation and the changes of overlapping areas in the images, the object function of the mutual information method may converge at local maximums so that wrong transformation parameters are generated as a result. In this paper, an improved simulated annealing–simplex method is proposed to solve this problem. Experiment results show that registration precision can reach to the sub-pixel level.