On the development of a fully automated cryosection image registration system

The CCD cryosection images for the male collected in the National Library of Medicine's Visible Human Project were registered through a procedure involving significant human intervention. In addition, several large misalignments were visible in the registered CCD data for the male. To expedite the exploitation of the visible human cryosection data sets, there is a need to develop a highly accurate yet efficient image registration tool. In this paper, we present the current status of the development of such a fully automated registration system. This system is based on the maximization of mutual information criterion. One key issue of this approach is a robust optimization scheme. The method developed in this paper is to estimate the initial search point through segmentations of regions of interest for each cryosection image slice. Experimental results using the 70 mm cryosection image data for the male are presented in this paper.