[Insight-users] better not use image pyramid?

[Ashish Poddar]

Just one more comment in addition to what has been discussed here. I also
used the pyramid filter earlier for registration of 3D images. but there was
a problem that the Z-dimension was only 7 pixels. so even  for second step,
it was divided by 2, giving <4 pixels. but since the image pyramid filter
internally uses gaussian filter (correct me if i am wrong) it needs more
than 4 pixels for approximation, and hence it didnt worked out for me.
However in the pyramid we can set different scales for different dimensions,
by which I was able to achieve the desired results.

regards,
Ashish.

On 11/27/05, Luis Ibanez <luis.ibanez at kitware.com> wrote:
>
>
> Hi Grace,
>
> You can use the pyramids for 3D registration.
>
> There is nothing wrong about it.
>
> The reason why the example in 3D is not using
> the pyramid is that the example is intended to
> be very simple.
>
>
> The fact that you attempted to use the pyramid
> and didn't performed better, just doesn't mean
> anything.
>
> This anecdotic information is useless. Saying
> that you try something once, and it didn't work
> is not a technical way of reporting a problem.
>
>
>
> In order to be useful, Your report should mention
> details such as:
>
>
> 1) Image modality
>
> 2) Image size
>
> 3) Components in your registration framework
>     3.1) Optimizer
>     3.2) do you use a Transform initializer ?
>     3.3) What Transform are you using
>
> 4) How many levels did you set in your pyramid
>
> 5) What parameters did you set up on ever level.
>
> 6) When you say that it "performed worsely"
>
>     what do you mean ?
>
>     6.1) it did not converged ?
>     6.2) it did converged but took longer ?
>     6.3) it converged but the images are still
>          misaligned ?
>
>
> Please be more specific when you report problems.
>
>
>
> It is very dangerous to generalize ad-hoc rules
> out of anecdotic information. Doing so is more
> along the lines of developing superstitions rather
> than working in scientific research.
>
> Such practices are fine for publishing papers on
> Decadent Journals where author only need to be able
> to say that "my method is better than any other one".
> This is not acceptable when you really want to solve
> problems in image processing.
>
>
> Please report your experience, with enough amount of
> detail for allowing somebody else to reproduce what
> you attempted.  Only in this way we can provide you
> with useful advice.
>
>
>
>
> Regards,
>
>
>
>      Luis
>
>
>
> ---------------
> Grace Chen wrote:
> > Hi all,  good day!
> >
> > I use the transform and the optimizer (VersorRigid3DTransform,
> > VersorRigid3DTransformOptimizer, and CenteredTransformInitializer) to
> > perforom 3D rigid registration on my data, and I notice that in itk
> > example, the image pyramid isn't used for the 3D registration.  I tried
> > adding that to my registration program, but it produced worsely
> > registered output.  Does anyone one know why?  Does that mean I
> > shouldn't use image pyramid technique for registering 3D data?
> >
> > Thanx!!
> >
> > Grace
> >
> >
> > ------------------------------------------------------------------------
> >
> > _______________________________________________
> > Insight-users mailing list
> > Insight-users at itk.org
> > http://www.itk.org/mailman/listinfo/insight-users
>
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users
>



--
Ashish Poddar
<http://ahpoddar.bizhat.com>
Its high time we standardize everything!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://public.kitware.com/pipermail/insight-users/attachments/20051128/5f10f0fc/attachment.htm