Medical image registration using modified iterative closest points

Abstract

The closest iterative point (ICP) algorithm is commonly used in medical image registration. However, due to its natural limitation, the processing time and registration accuracy need to be further advanced. In this paper, by computing the moments of the reference and floating images, the centroids are computed and thus the initial translation parameters are obtained. The rotation angles acquired respectively by the second-order central moments, inertia matrix, Karhunen-Loeve transformation (K-LT) and singular value decomposition (SVD) are referred to as the initial rotation parameters of the ICP algorithm for image registration. The edges of the reference and floating images are detected by Canny operator and then the binarization images involving the feature points are acquired. The experimental results show that this proposed method has a fairly simple implementation, a low computational load, a fast registration and good registration accuracy. It also can efficiently avoid trapping into the local optimum and is adapted for both mono-modality and multi-modality image registrations. © 2010 John Wiley & Sons, Ltd.