[Insight-users] MultiResolutionImageRegistrationMethod: passing Tform parameters between levels

[Karthik Krishnan]

On 2/27/08, Felix Bollenbeck <bollen at ipk-gatersleben.de> wrote:
>
> Hello all,
>
> I have I question regarding the MultiResolutionImageRegistrationMethod
> and passign the current-level parameter to the next level:
>
> At the end of one resolution level the transformparameters of the
> current level are stored:
>
> MultiResolutionImageRegistrationMethod.txx  line 312:
>
>       if ( m_CurrentLevel < m_NumberOfLevels - 1 )
>         {
>         m_InitialTransformParametersOfNextLevel =
>           m_LastTransformParameters;
>
> And in the next level iteration these are used for intialization:
>
> MultiResolutionImageRegistrationMethod.txx  line 108:
>
>   m_Optimizer->SetInitialPosition(
> m_InitialTransformParametersOfNextLevel );
>
> I do not understandt how  the transform parameters adapted to now larger
> image
>

Since most (see below) transforms in ITK represent transformations in
physical space and not in pixel space, nothing needs to be done to adapt
them to a higher resolution level.

This is not true for several deformable transformations. For instance if you
used a deformation field, you would have to resample your deformation field
at each resolution from a lower resolution field.  Or, if you used
itk::BSplineDeformableTransform, you would interpolate a high resolution
grid from a spline grid of lower resolution. In such cases you would use
itk::ImageRegistrationMethod instead of
itk::MultiresolutionImageRegistrationMethod and perform the interpolation
yourself.

For most transforms, itk::MultiResolutionImageRegistrationMethod works just
fine, no adaptation needed.

HTH
--
karthik
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://public.kitware.com/pipermail/insight-users/attachments/20080227/e850fb08/attachment.html