A Generic Approach towards Automatic Image Co-registration

This paper presents a mathematical framework for dealing with area-based image co-registration problems in a generic and modular way. Based on this framework an efficient implementation is devised allowing “plug & play” support for a whole gamut of geometric transformations, image similarity measures (criteria) and optimisation methods. A typical drawback of area-based methods is the exhaustive use of memory and the slow optimisation speed. Therefore, this paper outlines an effective subsampling strategy that considerably speeds up the registration process. The final part of the paper is devoted to an extensive evaluation of the algorithm using both ground-truth data and a series of challenging real examples.