Hi Fabrice,
one important point when processing FFTs:
Do *not* process the 8-bit image displayed as power spectrum in the
image window of an FFT. It is for visual inspection only; it is
logarithmic and it will have different scaling for each FFT. E.g. it
might comprise 7 orders of magnitude for one image, but 13 orders of
magnitude for another one, all mapped to the same 8-bit scale. This
means that 'Find Maxima' will need a different noise tolerance for
each image.
Instead, always use the raw Power spectrum that you can get from the
FFT options, and divide it by the image size^2 to get nicer numbers.
Usually you have to discard the lowest frequencies: set an area
around the origin, i.e., (size/2, size/2) to a very low value (not
exactly zero if you want to run 'log' later on).
Now you should get decent results with 'Find Maxima'.
You might also use a log scale now (Process>Math>Log); then you need
not care about the absolute height of the peaks when setting the
noise tolerance for 'Find Maxima', and typical values for the noise
tolerance might be around 5. If you want to apply smoothing, do it
before conversion to log.
You might also try 'Subtract Background' with the log image (then you
should not have changed the pixels at the origin to a low value).
Hope this helps,
Michael
________________________________________________________________
On 10 Nov 2011, at 14:34, fabrice senger wrote:
> Hi again,
>
> I work on an analysis where I need to define maxima in FFT power
> spectra.
> I saw that some people allready worked on taht.
>
> Somehow I'm not happy with the tool "Find maxima".
>
> Any suggestions are welcome.
>
> Thank you,
>
> Fabrice.
Locked
