Multi-sensor image automatic registration using mutual information

Abstract

A novel multi-sensor image registration algorithm was introduced, which was based on wavelet transform and the maximization of mutual information. Firstly, the region of interest was determined by search of wavelet magnitude maxima in the preprocessing step of images, and the affine transform was chosen as the image transform model. Secondly, the probability distribution of grey value was estimated by interpolation of pixel values to calculate image intensity histogram. Finally, the simulated annealing algorithm was used to approximate the global optimum. The experiment data show that the algorithm was effective and feasible.