Testing Texture/Shape Weighting for Different Objective Functions

The assumption here is that weighting was suited to work well with a certain type of objective functions (ssd, abs_diff wtd_abs_diff). Therefore, it is interesting to see if other methods don't work or perhaps just weighting is not correct. It would be interesting to see if MI and NMI work, as they should.

MI with varying weights was attempted.

• 25112004-4: weight is set to 0.05 (note that 0.5 was tried yesterday and was not successful)
• 25112004-5: weight is set to 0.005
• 25112004-6: weight is set to 0.0005
• 25112004-7: weight is set to 0.00005

The experiment was not as long as 2.1, but as the parameters in the appendix indicate, it was long enough to allow good results.

Results: the comparison which can be found under /MI_Weight-25112004 shows no change at all. The weight does not seem to affect the result. There is a change during MI registration though.

Conclusions: MI is not affected by shape/intensity weighting. It does not get good results in any of the weight settings at this moment.