On 27/08/15 19:23, Jonathan Gibbs wrote:
Has anyone played with OIIO's FFT functions? I've not worked with this
much, but what I'm used to seeing out of the FFT is an image with a
white pixel in the center and some matter of pixels around those.
(For instance: http://homepages.inf.ed.ac.uk/rbf/HIPR2/fourier.htm)
This is one convention of looking at the spectral content of an image,
i.e. [-N/2 -> N/2-1] with the DC component at the center...
What I get out of OIIO is an image with interesting bits in the 4
corners and nothing in the middle. Am I right to interpret pixel (0,0)
as the lowest frequency in X and Y and pixel (max,max) as the highest
frequency in both X and Y?
But OIIO uses the convention used by FFTW which is [0 -> N/2-1; -N/2 ->
1]. You would need to do an FFT shift (swap quadrants wrt the center) to
get the data in a centered layout.
MATLAB's or Numpy's fft / fftshift documentation gives you further
rationales behind the centered vs fftw layouts.
Ghis
_______________________________________________
Oiio-dev mailing list
[email protected]
http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
