Heuristic search for maximum of mutual information

Mutual information (MI) and Normalized MI has proved to be a reliable tool to measure the similarity between two images as they are chancing through a registration process. In this paper we present an improved registration algorithm based on Normalized MI similarity measure, incorporating spatial information of both images, using Genetic Algorithm (GA) capability of optimizing a multi-parameter optimization problem. For this to happen we employed gradient vectors of images to represent spatial information and made some modification to normalized MI. This modification made NMI of two images smoother with fewer fluctuation, enabling the GA to play the role of main optimizer. Slow moving and softening effects of some functions like natural logarithm is used to hamper constant changes in MI between two images.