Fast Fourier Transform

[Astrocava]

Hi all!

Was reading this in the website of QSI:

http://www.qsimaging.com/ccd_noise_measure.html

And want to try it with my camera, but I've sarched for a FFT process in PI and is ... missing?

Sergio

[Nocturnal]

I seem to remember it's implemented in PCL but not exposed as a module. You could search the forums as this has come up before. I used AIP4WIN to do the FFT on my images after seen Kevin Nelson at MWAIC '10.

[Philip de Louraille]

If the goal is to detect the dominant frequencies in some datasets then a FFT tool is quite useful.
If the goal is then to use that knowledge to remove some frequencies then a FFT (and its inverse) are not that useful because that technique will not differentiate frequencies found in noise from that found in coherent signals.

[Juan Conejero]

Hi Sergio,

That's very easy to implement with a little script:

Code: [Select]

#include <pjsr/UndoFlag.jsh>

/*
 * Active view
 */
var window = ImageWindow.activeWindow;
if ( window.isNull )
   throw new Error( "There is no active image window." );
var view = window.currentView;

/*
 * DFT of the active view image
 */
var C = Image.newComplexImage();
C.assign( view.image );
C.FFT( true ); // center DFT

/*
 * Modulus (absolute value) of the DFT as a new image window
 */
var window = new ImageWindow( 1, 1, 1, 32, true );
var view = window.mainView;
view.beginProcess( UndoFlag_NoSwapFile ); // do not generate any swap file
view.image.assign( C );
view.image.rescale();
view.endProcess();
window.show();
window.zoomToFit();

Just select the image you want (including any preview) and run the script. It will compute the discrete Fourier transform (DFT) of the image and will provide its absolute value as a new image window. In most cases you'll have to apply automatic STF stretch to the DFT in order to view something useful. DFTs are computed with central symmetry (the DC component is at the center of the transform matrix).

Enjoy!

[Philip de Louraille]

Cool! Thanks Juan!

[Astrocava]

Thanks Juan for the code!  

Well, this is the result of the SCRIPT to a single BIAS MasterBIAS substracted. There is a lot non random noise there. As the QSI web site states, will be it impossible to removal by combining multiple images? But can be removed by BIAS substraction?

btw, nice pattern.

[Astrocava]

Sorry for the up. Any idea?

Sergio

[Carlos Milovic]

I don't follow... why do you want to substract a bias from a bias?

[Ioannis Ioannou]

Well, I guess what you try to do is the procedure described at Iris' manual pages , http://www.astrosurf.com/buil/iris/new410/new410_us.htm, scroll down to "Fixed pattern noise removal". I was also looking for something similar last summer but never had the time to really implement it (I started a script with Juan's guidance, but never finished it).
 
(BTW it will be a great tool IMHO if ever been implemented into Pix - and should not be difficult. All functions are there, just somebody knowing how to combine them )

But, I do not understand why you need it. Your FFT is not a single dot,  but rarely is. The pattern you have is almost textbook. Look at Iris' page above to see which patterns you should worry about. IMHO any residual fixed pattern from your camera will never be visible with proper calibration (you should see what my camera produces, horror !! )

[Juan Conejero]

which patterns you should worry about.

The DFT you've got does not show periodic features that you should worry about IMO. It shows patterns corresponding to dominant orientations of significant structures in the image.

What you are looking for are features well localized on the Fourier transform space. Think of the Fourier transform matrix as if it was a 'normal' image: you're looking for relatively small, bright and isolated features, because they are easy to identify and remove with a 'point and shoot' procedure (known as notch filters speaking more technically). For example, an isolated feature on two/four quadrants of the (centered) Fourier transform matrix denotes a periodic pattern, such as the typical interference pattern, which is very easy to remove working in the Fourier domain. The frequency of the pattern is a function of the distance to the center of the matrix. There are no such features easily identifiable in the DFT of your bias frame.

BTW it will be a great tool IMHO if ever been implemented into Pix - and should not be difficult. All functions are there, just somebody knowing how to combine them Wink

Indeed it wouldn't be difficult —although the required user interface is not trivial if it is well implemented—, but there are other priorities. A Fourier transform tool is very nice and interesting from an academic point of view (and it's the kind of things I love to write ), but it would find little practical use in astrophotography —correct me if I'm wrong.

[Ioannis Ioannou]

A Fourier transform tool is very nice and interesting from an academic point of view (and it's the kind of things I love to write ), but it would find little practical use in astrophotography —correct me if I'm wrong.

Not with my camera when you want to bring out every tiny bit of detail 
I'm dealing with the noise from these patterns with ACDNR (and it does a pretty good job), but I'm loosing information also.
But indeed, the QSI in question is clean, no need for it.

[Astrocava]

Thanks all for the input!

Have fun!

S-