PDF version of this document
Some of the main relevant concepts and techniques in existence have been explained and numerous examples have been given, although their number was restricted to allow for a broader survey. Most such techniques directly relate to the problems which need to be tackled and their utilisation in past and present has been thoroughly explained. As future experimentation is expected to rely on recent research and is most likely to involve similar ideas, algorithms and paradigms, continuous reading of technical reports, alongside reproduction of the experiments, will be an essential portion of the research approached.
The problems with current techniques were found to vary from the interest in efficiency to possible flaws and gaps, a part of which being driven by insufficient correctness arguments and lacking ground-truth. Without a doubt, there are phases in current research where heuristics take over on the expense of valid implementation that can be reasoned straight-forwardly. Many areas are still controversial and common assent is missing and might never be reached. As instances for the aforesaid claims, a group-wise brain analysis algorithm devised a wide range of domain-specific facts. Moreover, a major undecided issue is the most advantageous warp type and its corresponding complexity that strives to give ideal results per permanent time unit.
This project, much like other projects in this area, attempts to find some answers to the questions raised and resolve uncertainties and disagreement by providing empirical proof or implementing a convenient tool for quickly evaluating and profiling different ideas and approaches. Whether it will be successful in the sense that it should provide inarguable answers and discover new techniques that are ingenuous, this project should draw conclusion regarding performance, feasibility and validate or invalidate some results of previous work. Preferably it should surpass previous work that it shall build upon. No results will be taken for granted and a critical approach will be dispensed at all times.
Within the first year of the project it is hoped that an implementation of a warp and model testing prototype will be available to achieve dense correspondence across a set of synthetic (and hopefully medical) images in 2-D and 3-D. Software should be capable of looking into the behaviour of warps regardless of the nature and scale of the data. It must also respond within sensible time period, although the notion of suitable time period is loosely-defined. There is a growing belief that such tool can be of great interest to these who use and facilitate active appearance models. Nonetheless, there is a real snag as the data under consideration should be modified to approve the successful application of the techniques to data of higher dimensionality. The run-time and the results that can be retrieved within a limited time-frame is then the main impediment.
Although a partial time-line was specified for this project and its intermediate objectives, it is not yet clear where the project will turn and what it will eventually accomplish successfully. It is known, however, what should ideally be accomplished. Semiannual reports and documents will clarify the emerging plans and intentions as they become more concrete.