Multimodal image registration using multiresolution genetic optimization

Image registration is an important preprocessing step in medical imaging applications. It can be formulated as an optimization problem where the associated energy to be optimized is a non-convex function that often shows local optima. Unlike classical numerical optimization algorithms frequently used in image registration, evolutionary optimizers involve search strategies preventing the algorithm from getting stuck in local optima and do not rely on a starting solution. However, they may suffer from slow convergence speed and lack of accuracy. In this paper we propose a new multimodal intensity-based image registration technique based on a specific design of real-coded genetic algorithm. The proposed approach provides a higher convergence speed than conventional genetic algorithm and superior alignment accuracy related to the use of multiresolution strategy with three image complexity levels. The experimental results show the outperformance of our method compared to a well-known registration method for real multimodal registration scenarios.