Multimodal image registration using stochastic differential equation optimization

Abstract

An approach to image registration is outlined based on a new stochastic differential equation optimization method. The proposed method requires the use of the numerical solution of a particular stochastic differential equation to determine the iterative update of the transformation variables. A comparison to Differential Evolution optimization was carried out to establish the rate of convergence and the quality of result, as measured by the number of cost function evaluations and the size of the standard deviation of the optimal variables. Experimental data shows that the new technique is robust in terms of computational speed and convergence. The method is validated on magnetic resonance and histology images of mouse brain.