[Insight-users] Difference between DiscreteGaussianImageFilter & RecursiveGaussianImageFilter?

[Luis Ibanez]

Hi Parag,

The basic difference between these two filters is that
DiscreteGaussian applies a convolution with a gaussian
kernel while RecursiveGaussian approximates this convolution
with IIR filters.

RecursiveGaussian has the drawback of being implemented
as a basic component for building up other filters.
It is applied along a single dimension (out of the ND).
So, in practice you may have to enchain N of these filters
in order to smooth an ND image.  In that sense, the
DiscreteGaussian is more convenient to use.

We may want to implement a variant of the RecursiveGaussian
that encapsulates the elements necessary to smooth an image
along all the dimensions. It shouldn't be too complicated
to implement.

If you are considering large sigmas, you are probably better
with the RecursiveGaussian filter since its number of
computations is independent of the sigma value.


---

The SoftwareGuide.pdf have a discussion on smoothing filters.
Please take a look at section 5.5. You will find the classical
gaussian convolutions in section  5.5.1 and the edge-preserving
smoothing filters in section 5.5.2.

There is a minimal examples that you can run in the command
line under:

Insight/Examples/Filtering:

- DiscreteGaussianImageFilter.cxx

you can feed it with Meta, PNG or DICOM images.

The GaussianFilter and GaussianFilter2D demo-application
use the RecursiveGaussian filters, so you can get a feeling
of its performance.


  Please let us know if you have further questions


   Thanks


    Luis



------------------------------------------------------------------

  Chandra wrote:

> I have a 2-D image that I want to blur using a Gaussian, and I know the 
> value of sigma for the Gaussian I should use. Which of these two filters 
> is best for that purpose? If both are suitable, I would like to know 
> which one will result in a smoother image.
> 
>  
> 
> Thanks,
> 
> Parag Chandra
>