PDF version of this document
A main goal, which relates to the big picture that is the GC, is the merger which involves (non-rigid) registration and statistical models. In both cases, some dense correspondence across some or all of the images is involved and must eventually be determined. Reuse of the information that is incorporated in each of this two techniques (which are believed to be inherently the same) would make the overall analysis task more powerful, flexible and well-integrated. If even a moderate combination of the two is obtained, then new ways of building and using models will be open for investigation. The parallel development in both fields, especially the need to identify homologous structures, is what makes this GC so suitably arranged and increases its potential of resulting in success.
In NRR, lower-level inspection of image pixels identifies similarity using mutual information (or any other similarity measure for this argument's sake), whereas in statistical modelling, the correspondences are often marked by hand (as explained in previous sections this is no longer quite the case) or gathered in an ill-chosen fashion. It is imperative that effort is made to reuse the segmentation from NRR so that models can be constructed more quickly and fitted to targets before feature extraction takes over and does its part of the analysis job.