[Insight-developers] Gaussian filtering.

[Miller, James V (Research)]

FilterDimensionality allows you to limit what dimensions of an 
image get filtered.  Sometimes you have a volume but only want
to blur in x and y.  To do this, you would set the filter dimensionality
to 2.

As for the variance vs sigma and pixel units vs physical units, the 
answer is "yes, we should be consistent".  In most cases, it really 
doesn't matter what coordinate system (pixel vs physical) you specify the 
parameters in as long as it is documented.  However, there are cases where 
an algorithm must take the physical units into account to get the correct
answer.
This is particularly important when the pixels are not isotropic.  If your
pixels are not isotropic, you either need to code all your algorithms to 
take this into account, or you resample your images to make them isotropic.
Both of these approaches have been taken in ITK.  You can argue that it is 
wrong not to properly handle the pixel spacing but some algorithms would be 
HORRENDOUSLY slow if the pixel spacing was explictly taken into account.
In these cases, we are better off documenting when isotropic pixels
are assumed.

Truthfully, I tend to work in pixel units more often than physical units.
It usually bites me and I convert over to physical units.

Jim




> -----Original Message-----
> From: Mark Foskey [mailto:mark_foskey@unc.edu]
> Sent: Monday, April 28, 2003 4:31 PM
> To: Miller, James V (Research)
> Cc: insight-developers@public.kitware.com
> Subject: Re: [Insight-developers] Gaussian filtering.
> 
> 
> Thanks for the info -- I've added some of it to the documentation for 
> those classes.  For the Discrete filter, I also documented 
> some of the 
> parameters, which had not been done.  (See the get/set methods.)  I 
> couldn't figure out FilterDimensionality (how can that be different 
> from the image dimensionality?) so if someone wants to add a comment 
> for that, it'd be great.
> 
> Miller, James V (Research) wrote:
> > RecursiveGaussianImageFilter measures Sigma in world coordinates.
> > 
> > DiscreteGaussianImageFilter measures Variance in pixels.
> > 
> > I used the DiscreteGaussian for small kernels and when I 
> will be taking
> > many different types of digital derivatives of the smoothed image.
> > 
> > I use the RecursiveGaussian for large kernels and when I 
> want to take the
> > derivative of a smoothed image in one step.
> > 
> > 
> > 
> > 
> > 
> >>-----Original Message-----
> >>From: Mark Foskey [mailto:mark_foskey@unc.edu]
> >>Sent: Monday, April 28, 2003 11:24 AM
> >>To: insight-developers@public.kitware.com
> >>Subject: [Insight-developers] Gaussian filtering.
> >>
> >>
> >>In itk::RecursiveGaussianImageFilter, is Sigma measured in pixels? 
> >>Assuming that it is, do we want to have an easy way select 
> a sigma in 
> >>world coordinates?  With anisotropic voxels, it might be nice 
> >>if there 
> >>were an easy way to simulate a spherical convolution kernel.
> >>
> >>Also, when would you want to use DiscreteGaussianImageFilter 
> >>instead of 
> >>the Recursive version?  Is there a rule of thumb, along the 
> lines of 
> >>"if you aren't sure, use this one"?  Are there other Gaussian 
> >>blurring 
> >>filters I haven't noticed?
> >>
> >>-- 
> >>Mark Foskey    (919) 843-5436  Computer-Aided Diagnosis and 
> >>Display Lab
> >>mark_foskey@unc.edu            Department of Radiology, CB 7515, UNC
> >>http://www.cs.unc.edu/~foskey  Chapel Hill, NC  27599-7515
> >>
> >>_______________________________________________
> >>Insight-developers mailing list
> >>Insight-developers@public.kitware.com
> >>http://public.kitware.com/mailman/listinfo/insight-developers
> >>
> > 
> > _______________________________________________
> > Insight-developers mailing list
> > Insight-developers@public.kitware.com
> > http://public.kitware.com/mailman/listinfo/insight-developers
> 
> 
> -- 
> Mark Foskey    (919) 843-5436  Computer-Aided Diagnosis and 
> Display Lab
> mark_foskey@unc.edu            Department of Radiology, CB 7515, UNC
> http://www.cs.unc.edu/~foskey  Chapel Hill, NC  27599-7515
>