Molecular image registration using mutual information and differential evolution optimization

In this work we propose a novel rigid image registration approach to determine the position of high-resolution molecular structures in medium-resolution macromolecular complexes. Mutual information similarity measure is used as an alternative to the cross-correlation coefficient commonly applied in this context. The optimum of the objective function is sought by means of differential evolution algorithm. This global optimization technique yields robust registration, exhibits fast convergence and is easy to use. In order to additionally improve its accuracy we combine it with a local gradient search strategy. The registration framework is tested both on simulated and experimental data sets forcing large rotations and translations. Results in terms of success rate and execution time, indicate the suitability of the proposed approach.