next up previous
Next: Multi-scale in NRR Up: Experiments Previous: Affine

Optimising Scale and Translation ``Together''

Experiment 21112004-3

The previous set of experiments suggested returning to translation one scale was 'fixed'. The effects were investigated by performing 6 stages: translation, scaling, translation, scaling, translation and finally scaling.

The results appear very interesting.

After the first 2 iterations, which is what normally gets done, improvement continue to be made. These improvements are quite significant too.

To find out how long such improvements can be made, a long such experiment was performed, namely:

Experiment 21112004-4

Same as above, but 18 stages rather than just 6. The result is very interesting to look at. Improvement are made during the first 4 iterations or so and then the brain 'wobbles' between two states (optimal scale versus optimal translation). It still seems that before that stage of 'wobbling', actual improvements are made. It was decided to try the same procedure on two completely different brain data which leads to the next experiment.

Experiment 21112004-5

Two very different brains were chosen. In this case, after the first 2 stages, the 'wobbling' state was reached. This would imply that optimising translation and scale 'together' might not always be helpful.


next up previous
Next: Multi-scale in NRR Up: Experiments Previous: Affine
2004-11-23