Abstract:

Medical image registration using mutual information can be viewed as a optimization problem. In optimization problems, the objective function is often trapped into a local minimum. In this paper, differential evolution is proposed for medical image registration, and applied to registration of multi-modality medical images. The proposed method can avoid local minimum due to its converging characteristics. Experimental results show that, using mutual information and affine transformation as objective functions, the differential evolution algorithm outperforms the Powell algorithm and the simulated annealing for medical image registration.

Key words:

differential evolution, image registration