[SciPy-user] shift FFT2D...

[Francesc Altet]

El dc 09 de 05 del 2007 a les 16:40 +0200, en/na fred va escriure:
> Hi all,
> 
> I use an FFT 2D on a matrix to compute a convolution like this:
> 
> a = fft2(input_data)
> b = fft2(output_data)
> c = real(ifft2(a*b))
> 
> The problem is that c should look like this:
> 
> c1 | c2
> -----------
> c3 | c4
> 
> but it looks like this:
> 
> c4 | c3
> -----------
> c2 | c1
> 
> How  can I get efficiently the right result ? (something like shift ?)

Something like:

N4 = N/4
c1 = c[:N4];  c2 = c[N4:2*N4];  c3 = c[2*N4:3*N4];  c4 = c[3*N4:]
tmp = c1.copy(); c1[:] = c4;  c4[:] = tmp
tmp = c2.copy();  c2[:] = c3;  c3[:] = tmp

should do the trick.

As c1, c2, c3 and 4 are views and not copies, the above snippet should
be fairly optimal in terms of both speed and memory usage.

HTH,

-- 
Francesc Altet    |  Be careful about using the following code --
Carabos Coop. V.  |  I've only proven that it works, 
www.carabos.com   |  I haven't tested it. -- Donald Knuth