A model for nonrigid mammogram registration using mutual information

This approach to mammogram registration uses a similarity-measure known as mutual information as the cost-function used to determine the degree of similarity between two corresponding mammograms. An initial registration is performed to remove global differences and bring the mammograms into approximate alignment. The mammograms are then partitioned into sub-images and each of the corresponding sub-images is matched independently using normalized mutual information. The centroids of each of the matched sub-images are then used as control-point pairs in association with a radial basis function to produce a global nonrigid registration.