A Comparison of Similarity Measures for 2D Rigid MR Image Registration Using Wavelet Transform

Abstract

Medical image registration is a decisive step in medical image processsing. In intensity-based image registration methods, multiresolution coarse-to-fine strategy is often used to speed up the registration process. In this paper, several commonly-used similarity measures were compared under multi-resolution wavelet framework. The similarity measures are energy, joint entropy, mutual information, normalized mutual information, correlation ratio, and partitioned intensity uniformity. Experimental results give a guidance to the selection of appropriate similarity measures for registration in a multiresolution wavelet framework.