On search space and search data strategies in image registration

Image registration is the process of determining the correspondence between all points in two or more images of the same scene. Image registration requires intensive computational effort. For high accuracy, robustness as well as low computational cost, a suitable similarity metric and reduction in search data and search space is required. In this paper, we look into subimage selection as reduction in search data strategy. We compare the reliability of mutual information, correlation coefficient and sum of absolute differences as similarity metrics using this search data strategy. We propose a measure, called alignability, which shows the ability of a subimage to provide reliable registration results, using mutual information as similarity metric. We also investigate alternative selection methods, such as entropy and gradient. We further investigate the effect of bin size on mutual information. A search space strategy using gradient techniques to maximize mutual information and our first results are also presented.