A PCA-aided EV-EGI Method for Registering Volumetric Datasets

Abstract

A method for volumetric dataset registration that utilizes principal component analysis (PCA) and volumetric extended Gaussian image (EGI)-based processing is presented. The method uses PCA to determine an initial coarse estimate of orientation difference between two volumetric datasets. The PCA is based on certain automatically selected (i.e., significant) voxels. The coarse estimate then is refined by a three-stage process that utilizes enhanced volumetric extended Gaussian images (EV-EGIs). These final EV-EGI stages also provide the translational component. The method's combination of steps allows for faster processing at roughly similar accuracy versus prior work based solely on EV-EGIs. Experimental comparisons with Globally optimal Iterative Closest Pointset (Go-ICP) registration are also reported and analyzed.