A local image registration method is proposed specially for the conditions when images

in medical image registration

are largely similar or the differences are not significant. The transformation function is a radial basis function with compact support and its locality can be conveniently controlled by distributing the feature points into the desired regions

which especially allows us to deal with local changes in medical images. Mutual information is chosen as cost function in order that the transformation function can be achieved accurately. In the process of the optimization

the image registration is treated as optimal problem and niche genetic algorithm is employed to optimize the transformation function parameters because it can overcome the drawbacks of premature and weak exploitation capabilities compared with genetic algorithm. The experiments on the simulated image with the known transformation function and the real image are conducted by using the proposed method. The results show that the optimal transformation function can be found and its action domain is controlled within a relatively small region. The presented method

which is a feasible and robust medical image registration approach

exploits the advantages of both feature points and intensity information and can obtain the accurate transformation function by the efficient optimization strategy.