[Insight-users] gradientRecursiveGaussianImageFilter

[Luis Ibanez]

Hi Martin,

This filter performs the equivalent of a Gaussian Smoothing
followed by a gradient operation.

http://www.itk.org/Insight/Doxygen/html/classitk_1_1GradientRecursiveGaussianImageFilter.html

It is not enough to compute the convolution with the derivative
of the Gaussian along one direction because that will not be
consistent with the model above.

As with many other operations in image processing this is a
trade off between memory consumption and computing time. If
memory is a concern, then you may want to use the Gradient
Image filter that computes the gradient using convolution
kernels.

http://www.itk.org/Insight/Doxygen/html/classitk_1_1GradientImageFilter.html
Note however that this filter does not perform smoothing...




The Recursive Gaussian is particularly interesting if you
need large Sigmas for your Smoothing and you need high speed.

A couple of speed versus memory analysis have been performed
for the various Gaussian smoothing filters available in ITK.
If you are interested in those analysis, you may want to search
the mailing list archives.

For more details on the theory behind the RecursiveGaussian
filter, you should read the report

"Recursively Implementing The Gaussian and Its Derivatives",
by Rachid Deriche
INRIA, 1993

ftp://ftp.inria.fr/INRIA/tech-reports/RR/RR-1893.ps.gz


Regards,


    Luis


-----------------------
Martin Urschler wrote:
> hello all,
> 
> I was looking at the GradientRecursiveGaussianImageFilter to find the 
> reason why the standard rigid registration allocates lots of memory. I 
> get an out of memory exception when registering two 512x512x231 images 
> (I have to say that I'm holding some more images in memory, however the 
> registration method seems to allocate more than 1GB RAM for gradient 
> calculation, which seems to me like being too much).
> Could anyone explain to me, what the mini pipeline consisting of 
> NrDimensions ZeroOrder gauss images and one FirstOrder gauss image 
> (derivative) is used for? Naively I would say one only needs the 
> derivative filter for each dimension.
> 
> thanks,
> Martin
> 
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users
> 
>