Abstract: An improved image registration method is proposed based on mutual information. Firstly, a new registration measure function, combining mutual information with morphological gradient information of the images, is used to take the advantages of these two different indices to achieve the global optimization. Secondly, a hybrid optimizer based on particle swarm optimization (PSO) and Powell is applied to efficiently restrain local maximums of our new registration measure function. Lastly, multiresolution data structure based on wavelet transform is used to expedite the registration process and increase the algorithm's robustness. Experimental results demonstrate that this new algorithm can efficiently yield a good registration result and can achieve subvoxel accuracy, while meeting the real-time need of diagnostic and research purposes.