PDF version of this document

New Algorithms

At the stage when results described in Chapter resembled these which are described in Chapter on MDL for shape models construction, it was decided to come up with new and more advanced ideas. In fact, parallel work by Twining on group-wise registration promoted collaboration and suggested that ideas should be exchanged in order to form a new, more powerful registration algorithm. As part of the discussions on issues of commonality and possibly duplicated effort, new schematics were drawn. This section is intended to explain them in some level of detail since they might at some point be implemented.

Two possible views on how the problem can be tackled are shown below. What is common to both is that they attempt to encapsulate the model discrepancy in one cunning way or another. They need to express a way in which discrepancies relate to the data and to the warps which are applied to that data.

Figure: One possible proposal for the further development of the registration algorithm. The main idea to take is that residuals should drive warps that in turn affect the model.

Figure: A second reasonable proposal for algorithm extension. MDL is the main driver of warps here.

For explanation of these, the aforementioned document needs to be looked at. These are rather long-winded to explain and the notation needs to be understood in advance. These notation is not final either and it is soon going to be changed as most recently agreed.