PDF version of this document

Adaptive Precision

The source of this useful strategy came from the registration algorithm. It was found that coarse and hastily-chosen transformation sufficed at the former stages of the optimisation. It was premature to require a high precision from the optimiser at these stages. Putting this idea in different terms, the machine chooses to loop laboriously looking for very meticulously-chosen warps when, in fact, since all elements in the set is dynamic, a slack choice of a warp serves the overall aim. Sets are manipulated one instance at a time and the problem appears quite different every time a loop is completed. By lowering the precision demands, more iterations can fit within a tantamount time period (see Figure cap:Adaptive-precision-requirement).

The subject of precision and its impact on speed will be revisited in Section . It is an essential strategic choice which is suitable for a problem where its nature is dynamic^7.21.