MI-Based Robust Image Registration Using a Novel Histogram Estimation Scheme

Mutual information (MI) has developed into an accurate measure for rigid and affine mono- and multi-modality image registration. The robustness of the measure, however, is questionable. To improve its robustness, in the present paper, a novel joint histogram estimation scheme is proposed in the calculation of MI for image registration. The new estimation scheme updates the histogram continuously by using an exponential function. Experimental results show that, compared with the partial volume interpolation (PVI), the suggested histogram estimation scheme yields a better MI registration function. The MI registration measure derived from the proposed scheme significantly reduces the interpolation-induced artefacts, encounters less local maxima and improves the registration robustness.