Abstract: An improved multi-model image registration method was proposed for visible and infrared image registration, which was based on optimal region mutual information (MI) and edge orientation correlation (OC). Firstly, with proper principle, local regions with rich information were selected by image partition. Then, in order to include spatial information, image edge orientation was introduced to construct an edge orientation map. Finally, by integrating the correlation of edge orientation map and MI of optimal regions with a trade-off weight, a new similarity metric was constructed. Experiment results show that by combining edge orientation correlation, this new metric can effectively improve the accuracy of MI based method, and processing with local optimal regions instead of whole image can improve the efficiency.