[Insight-users] Re: Reg Registration

[Luis Ibanez]

Hi Prasad,


When you use MeanSquares as image metric, the
optimizer must be set to MaximizeOff() [ not
to MaximizeOn() ], please look closely at the
ImageRegistration3.cxx example in line 381.

The MeanSquare metric is optimal when its
value reaches zero.  This is described in the
SoftwareGuide

http://www.itk.org/ItkSoftwareGuide.pdf

section 7.9.1 pdf-page 224.



Regards,


    Luis


------------------------------------

Prasad Sarma wrote:
> hi Luis,
> 
>  Thanks for your response. for the the Images of same
> modality I used the MeanSquaresImageToImageMetric and
> I set the optimizer to Maximizeon(). (This is
> illustrated in the  Registration example 3). 
> I tried with Minimizeon(), The translation seems to in
> the opposite direction and it failed after 150
> iterations, I used the step size .01,0.2 , I think for
> this maximizeon() seems to converge before 200
> iterations as we saw before. for minimizeon() 
> I get after 150 iterations the following o/p
> 
> ------
> 150 = 9279.42 : [-18.8754, -14.9157, -18.0816]
> 
> ExceptionObject caught !
> 
> itk::ExceptionObject (00EDFDA0)
> Location: "Unknown"
> File: C:\Documents and Settings\Prasad\My
> Documents\Insight\Insight\Code\Algorit
> hms\itkMeanSquaresImageToImageMetric.txx
> Line: 91
> Description: itk::ERROR:
> MeanSquaresImageToImageMetric(00231140): All the
> points
>  mapped to outside of the moving image
> -------------------
> I am thinking even though the means squares values 
> increase for MAxmizeon() the optimizer converges
> before 200 iterations for the values of 0.01 and 0.2 ,
> I tried to use still smaller step lengths but the
> optimizer is completing 200 iterations so ,I feel all
> the parameters ( MAximizeon(),.01,.2 etc) I have
> chosen are reasonble parameters. What is your opinion?
> Reg. Muldimodality I have to try with the MattesMutual
> information and the RegularStep gradient descent
> optimizer. I wil do it soon.
> 
> Thanks
> Prasad
> 
> 
> 
> 
> __________________________________
> Do you Yahoo!?
> The New Yahoo! Search - Faster. Easier. Bingo.
> http://search.yahoo.com
>