Efficient Multimodal Registration Using Least-Squares.
Maja Omanovic, Jeff Orchard · 2006
Abstract — Multimodal image registration is a difficult problem in both medical imaging and remote sensing. The least-squares cost function has generally been overlooked for multimodal registration problems due to an underlying assumption that the two images being registered must have corresponding intensities. More recently, methods that employ the least-squares cost function have been developed to efficiently evaluate the globally optimal shift and intensity remapping simultaneously. However, these methods estimate the translation and not the rotation. In this paper we propose a method for using the least-squares cost function efficiently for multimodal registration. By modeling rotation using a linear approximation, we find the globally optimal translation and intensity remapping, and locally optimal rotation angle. In a series of experiments based on registering PD-, T1-, and T2-weighted magnetic resonance images, our method performs better than mutual information.