Canon DSLR Question: In-camera darkframes vs masterdarks

[Christoph Puetz]

Dear community,

during deeper inspection of the theory of image processing I found a question concerning DSLR.
Most of the CANON DSLR allow to create and subtract a darkframe by camera operating system.
This is often found in the camera's menu under "long exposure noise reduction".
If you activate this, the camera takes an internal darkframe of the same exposure time after each lightframe
and subtracts this from the last taken light frame.

So my question is (from theory):

From my understanding this should be less efficient, as subtracting each single darkframe from each light frame
should add noise to each light frame (as described in the handbook of astronomical images processing).
From this point of view it should be better to create enough darkframes (of same temperature), store them,
create a masterdark and calibrate the images later.

Am I right ? What dou you think ?

kind regards
christoph

[naavis]

You are correct. You should not use the "long exposure noise reduction" function when doing astrophotography.

[pfile]

yes, external master darks are always better because you can drive the noise in the dark signal down by stacking darks.

there have been quite a few threads regarding this topic on cloudynights.  there is one poster in particular on cloudynights who knows a little bit about statistics and keeps making the claim that stacking lights that have been in-camera dark subtracted is the same as stacking dark subs and subtracting that master dark from each light. some business about "paired differences" and other jargon. but this neglects the fact that the lights have to be registered, and so the dark signal noise that's been injected into the in-camera darks no longer lies on top of itself when the lights are stacked. it's been shifted around.

rob

[Ignacio]

Yes, use a master dark built from a library of darks of same temperature. But mostly because you will make much better use of the session's time. 

On Rob's point on statistics, it made me scratch my head, but I will have to agree with the guy in cloudynights, as long as the number of lights and the number of darks used in building the master dark are equal. The way I see it is that the dark noise that is attenuated by stacking is random, and I assume spatially invariant across the sensor. So averaging out a set of darks before or after registration should give similar results, and the fixed pattern noise component would be equally well calibrated. Am I thinking this right?

Ignacio

[pfile]

well that guy has never made a mathematical argument why he's correct. he just keeps talking about paired differences, and has certainly never brought up the randomness in the dark signal noise as a reason why it might work.

maybe the lack of correlation after registration is not important if the noise in the dark signal is random. in theory it should be like any other shot noise. you could certainly write a simulation to compare the two techniques if you could characterize the dark signal's noise properly.

you are right though, the only way to improve SNR is to collect more signal. wasting 1/2 your sky time taking darks certainly does not make sense in that context.

the people that do use in-camera dark subtraction seem to mainly be having problems with hot pixels - if they use in-camera, the hot pixels are perfectly cleaned up. on the other hand, proper dithering and rejection during integration should solve this problem just as well, so to me it's a non-issue. just make your darks offline and calibrate DSLR images as you would with a CCD.

of course, whether or not DSLR images can ever be properly calibrated is another topic - it's well known that canon (and others) play all kinds of trick to supress the dark signal.

[Christoph Puetz]

Thank you very much for your answers !!