Multiresolution Image Registration Using Spatial Mutual Information

Abstract

The maximization of mutual information has been very successful at the registration of images. Unfortunately, mutual information takes only into account the relationship between individual pixels and not those of each pixel's neighborhood. Mutual information ignores spatial information. In this paper, we propose a new similarity metric called enhanced mutual information (EMI), which combines mutual information with a weighting function based on the absolute difference of corresponding pixel values. In order to reduce computational cost we also use multiresolution wavelet decomposition as a search data strategy. Experimental results show that the enhanced mutual information is not only more robust to noise than mutual information but more reliable in the registration of multitemporal images.