[Insight-users] non-gradient base optimizer for 3D registration with versors

[Luis Ibanez]

Hi Moti,


Here are a couple of suggestions:


1) You can use the OnePlusOne evolutionary optimizer.
    This optimizer does not require gradients in the
    cost function. Note that the versor rigid 3D transform
    will map its parameters to the Versor space, so that
    you stay in the unit sphere.




    but.. before you get rid of the Powell optimizer,
    note that you may be simply using a bad set of
    parameters for it:



2) The message that you are getting from the Powell optimizer
    can be related to:

      A) Poor initialization of the transform.

         How are you initializing your transform ?
         Have you verified that the initial transform
         give you a large overlap of the fixed and moving images ?


      B) Using steps that are too large

         How many iterations does the optimizer performs before
         you get the exception ?
         What parameters are you passing to the optimizer



      C) Lack of using a right combination of parameter scaling.

         What values are you setting for the parameter scaling?
         This is very important, since it compensates for the
         difference in the dynamic range of the versor components
         versus the translation components.
         You may find useful to read the description of this
         issue in the ITK Software Guide:

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




    Regards,


        Luis


---------------------
Moti Freiman wrote:
> Hello,
> I'm trying register two 3D volumes within the ITK framework using a new 
> similarity measure which has no analytical gradients.
> Since the 3D versor optimizer in itk is based on a gradient decent 
> optimizer. i cannot use it with my similarity measure.
> I tried to use the Powell optimizer with 3DRigidVersor transform but I 
> got the following message:
> "All the points mapped to outside of the moving image"
> It seems that powell is not optimized for the versor space.
> 1. Is there any option to build a non-gradient based optimizer for the 
> versor space as it done for gradient decent optimizer?
> 2. Any other suggestions to which elements i should use in my 
> registration process given that my similarity measure has no gradients?
> Thanks,
> Moti
> -- 
> __
> Moti Freiman, Ph.D Student.
> Medical Image Processing and Computer-Assisted Surgery Laboratory.
> School of Computer Science and Engineering.
> The Hebrew University of Jerusalem Givat Ram, Jerusalem 91904, Israel
> Phone: +(972)-2-658-5371 (laboratory)
> WWW site: http://www.cs.huji.ac.il/~freiman 
> <http://www.cs.huji.ac.il/%7Efreiman>
> 
> -- 
> __
> Moti Freiman, Ph.D Student.
> Medical Image Processing and Computer-Assisted Surgery Laboratory.
> School of Computer Science and Engineering.
> The Hebrew University of Jerusalem Givat Ram, Jerusalem 91904, Israel
> Phone: +(972)-2-658-5371 (laboratory)
> WWW site: http://www.cs.huji.ac.il/~freiman
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users