Originating from information theory, mutual information, as a measure for image registration, has drawn much attention and has been shown to be successful. Mutual information should be maximal when the two different images are perfectly aligned. There exist many optimization schemes applied to mutual information matching problems, most of which are local and require a starting point. In this paper, we propose an improved genetic algorithm as a search engine to overcome this problem. To reduce the search data size, multi-resolution optimization strategy is adopted; meanwhile, adaptive sizes of the crossover and the mutation pools along with the changes of resolution are proposed to prevent the process from stalling at a local maximum, accordingly improve the capability with the local search of genetic algorithm. Experiments show our algorithm is a robust and efficient method which can yield accurate registration results.