[Insight-users] equivalence of Gaussian and low-pass-FFT

Roman Grothausmann roman.grothausmann at helmholtz-berlin.de
Fri Nov 13 04:31:48 EST 2009


Thanks every body for the answers. With Insight-book I mean Insight Into 
Images: http://www.akpeters.com/product.asp?ProdCode=2175

So it's only equivalent if I multiply the FFT with the FFT of the 
convolution Kernel I use in real space.
I think though there's one difference: Using the FFT allows to reduce 
the discretization (a better discrete interpolation) of the result since 
it's calculated with floating point precision. Convolving with a (wide) 
Gaussian gives me very few grey values.

Regards,
Roman

C.Cagatay Bilgin wrote:
> Your question is actually independent of the ITK
> usage. Convolution in time domain corresponds
> to multiplication in frequency domain, so gaussian
> filtering your image is actually multiplying the FFT
> of your image with the FFT of the gaussian filter you
> used (you might need some padding as most probably
> your gaussian filter is was smaller than your image).
> 
> If you want to clean up your image a little, I would say
> do not go to FFT just do it in time domain using the
> gaussian filters provided in the toolkit as you will
> need less number of lines of code.
> 
> By the way, is the Insight-Book are you referring
> the software manual or something else ?
> 
> Regards,
> ---------------------------------------------------------------------------- 
> 
> Cemal Cagatay Bilgin,
> Ph.D Candidate in Computer Science Department at RPI
> bilgic at cs.rpi.edu - http://www.cs.rpi.edu/~bilgic
> Lally 01A, 110 8th Street Troy, NY 12180-3590, USA.
> Tel: (518) 276-8489, Fax:(518) 276-4033
> 
> On Nov 12, 2009, at 10:46 AM, Roman Grothausmann wrote:
> 




More information about the Insight-users mailing list