Under ideal circumstances the problem tomographic reconstruction is well
posed, and measured data are sufficient to obtain accurate estimates of
volume densities. In such cases, segmentation and surface estimation
from the {\em reconstructed volume} is justified. However, in other situations
the tomographic reconstruction is an ill posed problem, and the reconstruction results are
corrupted by incomplete sinograms, noise in the measurement process, and
misregistration of the views. Such is the case in electron beam
tomography (EBCT).
This talk presents a direct approach to EBCT segmentation and
visualization. The strategy is to impose a fairly simple model on the data,
and treat segmentation as problem of estimating the interface between two substances
of somewhat homogeneous density. The segmentation is acheived by
simultaneously deforming a surface model and updating density parameters in
order to acheive a best fit of the data in the sinograms. The deformation
is acheived via a level-set surface model at the resolution of the input
data, and several computational innovations make the approach feasible on
state-of-the art computers.
Copyright. All Rights Reserved.