Application of Steerable Wavelets in Medical Image Registration

This paper presents a new method of medical image registration for affinely deformed images using steerable wavelets. Mutual information based methods are often used in medical image registration due to their accuracy. Multiresolution image registration approach has a potential to speed up the computation of mutual information based methods. In this paper, use of steerable wavelets is proposed for multiscale and multiorientation image representation to overcome the limitations of conventional wavelets. Registration parameters estimated in the coarsest resolution scale are passed to the finest resolution scale to speed up registration. Similarity measure of mutual information is used in the coarsest resolution scale and normalized cross correlation is used in the finest resolution scale. Experimental results with medical datasets are provided.