PDF version of this document

Milestones

From the large pool of experiments which were performed, the next few sections, which include much of the 'meat' of this report, select and focus on a few which inferred important information and conclusions that ought to be highlighted.

The following experiments can be seen more or less as milestones. These are the catalysts for the more interesting and/or perplexing results which could be identified amongst the entire set of experiments. Each of these will be explicitly or implicitly mentioned later on as minor milestones are dealt with roughly chronologically in Section (page ), Section (page ) and Section sec:Present-Work.

The milestone are:

1. Registration target and approach of the model-based evaluation to it.
2. Comparison and benchmark of different registration methods.
3. Finding the correlation between the size of the set and the performance of the model-based objective function.
4. Point insertion to compensate for the change in bump height.
5. Use of the residual of the model to better perform (4).
6. The optimisation refusing to improve steadily, fixed by dynamically changing the precision required from the optimiser.
7. Finding out that optimisation can go below target even when initialised at the correct solution described by a piece-wise linear warp.
8. Considerable speed-up of the algorithm.