[Insight-users] Re: On MIRegistration

[Luis Ibanez]

Hi Valli,


Your new matrix looks much better. The numeric
values I suggested were the theoretical result
of the registration. It is sure that you will
never get exactly those values since after all
the image you rotated may have been interpolated
and will not necessarily fit back into a 15 degrees
rotation.

The registration process will no go any better
than pixel accuracy.

On the other hand, there is a remanent translation
in the transform, that probably shouldn't be there.
It is likely that the optimizer is compensating a
bit the erroneous translation with a fake rotation.

You may want to play a bit with the "TranslationScale"
parameter that determines how much rotations are
optimized with respect to translations. In your case,
translation should be reduced to a minimum since
you know that the underlying transformation involved
only rotations.

----


What problems did you encounter in defining an
itk::Command for observing the iteration events ?

The example class in the file:
Insight/Testing/Code/Algorithm/itkCommandIterationUpdate.h

should work without any changes. Just instantiate one
in your code and connect it to the optimizer.

Please let us know if you have further questions.


   Thanks


      Luis


=============================================

valli gummadi wrote:

> Dear Mr.Luis,
>          I have tried with different sets of iterations and learning rates.
> Final matrix i could get is
> 0.999998    -0.001080    0.001774
> 0.001025    0.999529    0.030684
> -0.001806    -0.030682    0.999528
> 
> -0.183549    0.204500    -3.955895
> 
> In ur mail u mentioned matrix should be
> 1.00000   0.00000    0.00000
> 0.00000   0.999529   0.030684
> -0.001806 -0.030682  0.999528
> 
> I think 0.999998 is almost equal to 1.00000.
> But,I could not get 0.000000.
> I have tried in many ways.
> 
> I am unable to use itk::Command to see values at exch iteration.
> How to get the actual matrix.
> 
> Thanking U,
> Regards,
> Valli.
> 
> 
> 
> 
> 
>