Image registration using uncertainty coefficients

Abstract

The use of information theoretic cost functions in image registration is investigated, focusing on uncertainty coefficients. The theoretical derivation of the popular normalised mutual information cost function is presented and shown to be equivalent to a particular symmetrisation of relevant uncertainty coefficients. The use of uncertainty coefficients in image registration is investigated by recovering simulated deformations applied to MRI images showing that they may be used for image registration. A comparison of popular information theoretic cost functions is produced.