AN IMPROVED ROBUST HIERARCHICAL REGISTRATION ALGORITHM

This note describes an improvement to an accurate, robust, and fast re gistration algorithm (Alexander, M.E. and Somorjai, R.L., Mag. Reson. Imaging, 14:453-468, 1996). A computationally inexpensive preregistrat ion method is proposed, consisting df simply aligning the image centro ids, from which estimates of the translation shifts are derived. The m ethod has low sensitivity to noise, and provides starting values of su fficient accuracy for the iterative registration algorithm to allow ac curate registration of images that have significant levels of noise an d/or large misalignments. Also, it requires a smaller computational ef fort than the Fourier Phase Matching (FPM) preregistration method used previously. The FPM method provides accurate preregistration for low- noise images, but fails when significant noise is present. For testing the various methods, a 256 x 256 pixel T-2()-weighted image was tran slated, rotated, and scaled to produce large misalignments and occlusi on at the image boundaries. The two situations of no noise being prese nt in the images and in which Gaussian noise is added, were tested. Af ter preregistration, the images were registered by applying one or sev eral passes of the iterative algorithm at different levels of preblurr ing of the input images. Results of using the old and new preregistrat ion methods, as well as no preregistration, are compared for the final accuracy of recovery of registration parameters. In addition, the per formances of three robust estimators: Least Median of Squares, Least T rimmed Squares, and Least Winsorized Mean, are compared with those of the nonrobust Least Squares and Woods' methods, and found to converge to correct solutions in cases where the nonrobust methods do not. (C) 1997 Elsevier Science Inc.