Elitism in intensity-based image registration

Elitism is a variant of genetic algorithms which enables quicker convergence to the global optimum by preserving the best solutions of the current generation and passing them either unchanged or slightly changed to the next one. This guarantees that in each generation the best elements will be, at least, as good as those of the previous one. As it happens with many stochastic algorithms, Elitism has been used in image registration software and systems. Yet, to the best of our knowledge, no one has investigated its optimization potential in intensity-based image registration with respect to the number of the elites that is allowed to be reserved in the next generation and its connection to the mutation rate. In this paper, a series of experiments are conducted with respect to the number of the best solutions that are preserved unchanged into the next generation and the mutation rate with the purpose to study its effect on convergence, and more specifically whether it converges to the global optimum as well as its stability. Our results indicate that increasing the number of elites as well as the mutation rate will most likely improve the convergence of the registration method to the global optimum.