PDF version of this document
The appearance models currently used are not ideal in any sense and a solution to this flaw would be highly desirable. While it is not clear how to optimise models or how to evaluate a model [44], there are measurable means for arguing about the quality of these models comparatively. Amongst the main problems that are ordinarily seen in appearance models is their inferior performance, although this depends on the functionality required. Automation could have a significant contribution to such a model, but correspondence needs to be achieved first. Luckily, issues of correspondence have been investigated largely in the past decade so this should not be a peril. What is worth investigating even further is the ability of warps to improve models and at the same time encapsulate several analysis steps together. What is even more reassuring is the proven ability of models to improve warps selection and improve on existing group-wise registration methods. This improvement relies on the fact that a large-scale collectiveanalysis replaces the weaker yet computationally inexpensive pair-wise scope.