[Insight-users] Re: Multi Resolution Registration

[laia sales]

Hi Luis,

thanks for the hints,

I am using Windows XP and Visual Studio 8. I used the release mode, and to
initialize the transform that is perform with itkVersorRigid3DTransform and
the initializer is itkCenteredTransformInitializer.

For the optimizer* *I used itkVersorRigid3DTransformOptimizer

optimizer->SetMaximumStepLength( 4.00 );

optimizer->SetMinimumStepLength( 0.01 );
I tried to fix these parameters, but maybe in the metrics I am using
itkMeanSquaresImageToImageMetric
I have to initialize some parameters to fast it up.

Laia




On 9/21/07, Luis Ibanez <luis.ibanez at kitware.com> wrote:
>
>
> Hi Laia,
>
>
>                        Welcome to ITK !
>
>
> It should be possible to register, in about 5 minutes, images
> of the size that you describe, using modern desktop computers.
>
>
> You may want to give it a try to fine tune a normal single-resolution
> registration before you move into the multi-registration framework.
>
>
>
> A couple of questions:
>
>
>   1) What is your platform ? (compiler, operating system)
>
>
>   2) Did you make sure to build ITK in Release mode ?
>      (a Debug build will be a lot slower).
>
>
>   3) How are you initializing the Transform ?
>      (starting with a poorly initialized transform will
>       make very hard for the optimizer to find a solution
>       in a short time).
>
>
>   4) Did you produced a log of the transform parameters
>      at every iteration of the optimizer ?
>      This is very useful for fine tunning the parameters
>      of the registration.
>
>
>
>   Please let us know,
>
>
>      Thanks
>
>
>        Luis
>
>
> -------------------
> laia sales wrote:
> > Hi All,
> >
> > I am quite new in ITK and I am trying to use the Multi-Resolution
> > registration from ITK to register my images.
> >
> > I am currently using CT images from femur bones of aproximately
> > 250x250x500 with different resolutions of around 0.6x0.6x0.1 mm.
> >
> > The Optimizer, i am using is the itkVersorRigid3DTransformOptimizer and
> > the metric itkMeanSquaresImageToImageMetric.
> >
> > I was firstly implementing the rigid registration with the code provided
> > with the 3D registration, and it resulted pretty slow. Therefore, I
> > wanted to try with the pyramidal registration to speed up the
> > registration, but it is not giving a faster process, but slowing down. I
> > don't get well the parameters SetMaximumStepLength and
> > SetMinimumStepLength. How these parameters can reduce the number of
> > iterations for each level and achieve a faster and optimal registration.
> >
> > Does anybody use that kind of registration before and can give me some
> > hints abut it?
> >
> > Regards,
> >
> >
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://public.kitware.com/pipermail/insight-users/attachments/20070925/a1334e2a/attachment-0001.htm