By expatiating image registration method based on wavelet transformation and mutual information, an improved image registration algorithm was proposed. The low-frequency images of wavelet decomposition are used to register; the mutual information was used to be a scale of similarity to two registered images. A new optimization algorithm, which is a hybrid of particle swarm optimization and Powell algorithm, is used in this paper. The experiment shows that this algorithm reduces the amount of computation largely and improves the speed of image registration when the registration precisions have not notable change.