[SciPy-user] Data convolution.

Lance Boyle lanceboyle at cwazy.co.uk
Thu May 27 23:46:41 CDT 2004


When doing (linear) convolutions (of any dimension) with FFTs, does 
SciPy take care of aliasing in the "other" domain, usually time domain 
and spatial domain? This is done by zero-padding and with the so-called 
overlap-add and overlap-save methods. This is a different issue from 
edge effects (as I understand the earlier posters) which are Gibbs 
phenomena.

Jerry


On May 27, 2004, at 2:45 PM, Stephen Walton wrote:

> On Wed, 2004-05-26 at 16:01, Travis E. Oliphant wrote:
>> Andrea Riciputi wrote:
>>>
>>> Thanks, but my data set are typically 1024x1024 so I think I need 
>>> FFT.
>>> Nevertheless I'll look into scipy for testing and verifying my own 
>>> code.
>>>
>
>> It could be as simple as
>>
>> ifft2(fft2(a)*fft2(b))
>
> I've got quite a bit of experience doing this with astronomical images.
> Travis is right about boundary issues;  you almost certainly want to
> extend your images to twice their original size in each direction (four
> times their original area) with the mean of your images.  In my case,
> when I do this with full disk images of the Sun, I extend the image 
> with
> a model aureole to avoid edge effects.  This may be overkill.
>
> All my code for doing this is in Fortran and built on top of the PORT
> library FFT routines, so you probably don't want a copy.
>
> -- 
> Stephen Walton <stephen.walton at csun.edu>
> Dept. of Physics & Astronomy, Cal State Northridge
>
> _______________________________________________
> SciPy-user mailing list
> SciPy-user at scipy.net
> http://www.scipy.net/mailman/listinfo/scipy-user
>



More information about the SciPy-user mailing list