A novel registration method that incorporates template matching and mutual information

Image registration is kernel algorithm of many system applications such as three-dimensional reconstruction in computer vision area. This paper proposes a registration method that combines mutual information and template matching, firstly carries out template matching with similarity criteria based on mutual information, then gets candidates of sub-images, according to spatial relationship between a image and template and spatial relationship between the other one and candidates, we can get panorama image after registration. After computing mutual information of both big images corresponds to every candidate, we can get final registration result according to the maximal mutual information of both images. This method not only solves defects of local extremes in traditional mutual information method that computes both big images´ mutual information directly but also reduces computation. Experiment results show that the proposed method is robust to images which have great gray scale difference.