Overlap insensitive image registration combining MI and feature points
Wolfgang Jacquet · VUBIR (Vrije Universiteit Brussel) · 2007
In this paper we introduce a two step method to reflne the entropy based registration approaches used in medical image registration. Entropy is deflned over the variable overlap of the reference and test image. When images are made with different acquisition techniques, the difierences in the flelds of view, and the relative position of the objects being studied, can be large, resulting in relatively large overlap changes in the registration process. To reduce the sensitivity to these changes in overlap, and to minimize the resulting misalignments, Normalized Mutual Information (Studholme et al., 1999) and the Entropy Correlation Coe‐cient (ECC) (Maes, 1998) have been introduced. Although these criteria are known to be rather robust