PDF version of this document
EVERAL results have now been shown and discussed, but there are less technical matters that must be at least mentioned. The establishment of a naïve and simple model-based objective function was not the main implementation aspect that the author can take credit for. It was already in place at the beginning of the year, but it lacked many of the components that presently make it actually work and achieve good results which drive this research onwards. Results can now also be obtained rather quickly and flexibly since a front-end to the console-based functions was established. Technical details about the implementation can be found in the following sections on the WWW:
Experiments were often performed on several machines to collect different results of large registration processes simultaneously. Much of the computational power used resided in the Department of Computer Science where many of the strong and modern9.1 computers laid idly.
There were some difficulties at getting persistent computational power at the very late stages of the year when long-standing clusters of computers were stored due to refurbishment work. However, that was the point when more theoretical ground was sought, as opposed to the analysis of large and cumbersome experiments. Such experiments were mainly used to provide apodictic proofs (as described in Chapter cha:Experiments). Also at that time, work on smaller experiments, which did not concern images, became much more appropriate.