Correlation and mutual information based image registration

In this investigation we shall look into the co-registration ofradar (TerraSAR-X) and optical (Pleiades and SPOT) data.When performing geometrical co-registration of images,correlation is often used as a measure of association betweenvariables. Correlation is not necessarily a good choice fordata of very different modalities (as here with radar and optical data). Inspired by practise in medical image analysis,in this work, we supplement correlation as the measure ofassociation between variables with the information theoretical measure mutual information (MI). Information theoreticalwork was pioneered by Claude Shannon in his now classical article from 1948.MI allows for the actual/sample joint and marginal distributions of the variables involved and not just second orderstatistics. While correlation is ideal for Gaussian data andlinear associations, MI facilitates analysis of variables withdifferent genesis, different modalities and therefore differentstatistical distributions and nonlinear associations.