[Insight-users] Problem fft 3D

[Luis Ibanez]

Hi Alex,

Follow Angela's example:
http://www.itk.org/pipermail/insight-users/2008-July/026867.html

Pad the image with the value that is representative
of the background in your dosimetry images.

You may want to get familiar with the effect of
padding input images before you apply Fourier
Transforms to them.


For example:


   a) Change in resolution in the Fourier Domain

   b) Risk of introducing spurious frequencies
      if the padding is discontinuous

   c) Therefore, eventual need for using
      Windowing in the input signal (the image).



Before you proceed further, you should consider
spending some quality time with a book like

   "The Fourier Analysis and Imaging"
   by R. Bracewell
http://books.google.com/books?id=kucSV0SFaF0C&printsec=frontcover&dq=bracewell+fourier&sig=ACfU3U1jq4f4pCCLZPvkI6P4N7n51Yb7MQ

   or

   "The Fourier Transform and Its Applications"
   by Bracewell
http://www.amazon.com/Fourier-Transform-Its-Applications/dp/0073039381



Practice with 1D Fourier Transforms in Octave
(or Matlab), and get familiar with:

     a) Zero-Padding

     b) Aliasing

     c) Hamming, Hann, Cosine and Lanczos windows
http://en.wikipedia.org/wiki/Window_function#Spectral_analysis



Understanding the Fourier Transform, and its discrete
implementations will save you a lot of time and
unnecessary suffering.




     Regards,


        Luis



---------------------------
alexandre govignon wrote:
> i have found the problem,
> it s about the dimensions of my image, i don't know that i need power
> of two image for vnl FFT, which is a very bad news.... Have you an
> idea... without using FFTW, i think about resampling my image data to
> a correct dimensionned image? but maybe there lost of information...
> and it's about medical imaging system for dosimetry... maybe it
> shouldn't be a good idea???
> What do you think?
> 
> Kind regards
> Alexandre Govignon.
>