A Fast Algorithm to Estimate Mutual Information for Image Registration

We propose an extension of mutual information and a new fast algorithm to accelerate the evaluation of mutual information of images. This algorithm adopts gauss function as kernel function, and then uses fast gauss transform to reduce time complexity and improves the fast gauss transform by adaptive k-center clustering. The new algorithm can estimate smoother curves of mutual information function with registration parameter than original mutual information and evaluate the mutual information in linear time complexity. Experiments show it is very efficient.