This paper presents a method of registering multi-sensor images such as visible and infrared images for image fusion. The registration is achieved by finding the corresponding feature points provided by Harris corner detector. For solving multi-sensor registration problems, mutual information (MI) is often used as the similarity measure for corresponding feature matching but was shown to lack robustness in certain cases. To overcome this limitation, we use a combination of normalized mutual information (NMI) and gradient orientation as a new similarity measure for corresponding pair matching. The proposed algorithm shows improved performance result compared with that of the conventional algorithm.