[Insight-users] Multiresolution registration shrink factors

[Luis Ibanez]

Hi David,

When using the registration framework it is
always useful to think in terms of millimeters.

Given that your images should have consistent
spacing, and the pyramid structures corrects
for spacing when subsampling, it actually
doesn't matter how you select the shrink factors.
... at least in theory...

In practice, you probably want to make sure that
the top levels of the pyramid are not too degenerate.
E.g. not using a too small image like 4x4x4 pixels...

Also you will have a penalty in performance for
using images with larger number of pixels.
(not that much in MutualInformation metric, but
mainly with MeanSquares, PatternIntensity and
NormalizedCorrelation).

---

That being said, from the point of view of
making good use of the number of pixels available
in each image, it makes sense to use the first
levels of the piramid to bring the two images
to the same resolution.

For example in your case, do not decimate
the moving image in the first level (use
factors 1 1 1 for the moving image and
2 2 1 for the fixed one).

After that you could progress up the pyramid
by trying to keep both images with approxim
the same number of pixels along each dimension.
This is just an intiuitive reasoning,... so
you may actually want to preform a couple of
experiments.

We will appreciate if you let us know what
you find.


Thanks


    Luis




---------------------------------------------
David Wikler wrote:
> Dear ITK Users,
> 
>    I have a question about shrink factors in
> multiresolution MI registration.
> If we have 2 datasets with different resolution,
> is it better to choose shrink factors so each
> resolution level registers volume of the approximatly
> same image size and voxel size.
> For example, if my fixed volume is 256,256,20 and
> my moving volume is 128,128,10
> 
> Should I preferably choose
> 
> 4 4 1
> 2 2 1
> 
> as shring factors
> 
> instead of for example
> 2 2 1
> 2 2 1
> 
> Thanks for your help.
> 
> David
> Brussels
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users@public.kitware.com
> http://public.kitware.com/mailman/listinfo/insight-users
>