Canon 600D - Darks or No Darks?

[Aramcheck]

Last week I mentioned in a reply that I'd previously tried stacking a set of images from my Canon 600d with & without Darks, and that with Darks the noise level was reduced by 50%. I've now tried to quantify things a bit better & have found that I'm now only getting about a 7% improvement in noise levels when using Dark frames.  I'm not sure why there is such a discrepancy, unless I did something wrong with my early comparison, but that would have back before I'd started guiding /dithering & using longer subs.

To quantify the effect of Darks with my Canon 600D (astromodified), I processed 21 x 180 sec Lights in PixInsight with:-
a) No Master Dark
b) a 97 sec Master Dark
c) a 180 sec Master Dark
(All with a SuperBias and Master Flat).

I first weighted the 21 x Lights using PixInsight's SubFrameSelector process, & then applied ImageCalibration, CosmeticCorrection (with Auto Detect set to 5 for Hot & Cold pixels), Debayer & Alignment, before Integrating, cropping & applying DBE. Only the ImageCalibration was applied differently, with the different Dark options.

I then cloned 9 background previews & combined them with the Preview Aggregator script, and ran the Statistics process on each of the results.

With the 180sec Master Dark applied, the Mean & Median noise levels came out at 93% compared to the No Dark, with a similar reduction in the (averaged RGB) Max value, and a 2-3% improvement in the std/avg deviations. The 97 sec option had minimal effect.

Cheers
Ivor

[Budgie1]

I'm watching this one with Interest.

I've always taken darks and included them when stacking but I have read a few comments recently, not just on SGL, that when using a DLSR's the darks don't make a great deal of difference and don't waist time with them.

Flats & bias, yes.

If that's the case then I'm happy to spend more time taking an extra 10 or 20 lights at he end of a session to improve the image than I am taking darks.

[rnobleeddy]

I'm by no means the expert, but:

- the advice not to bother with darks usually comes paired with the advice to dither instead (and large dithers at that!)

- if you do take darks, they need to be temperature matched very closely, and plenty of them

 

For this analysis, were the lights dithered, and how many darks went in to the master dark?

[Aramcheck]

2 hours ago, rnobleeddy said:

For this analysis, were the lights dithered, and how many darks went in to the master dark?

Yes I dithered between subs & there were 50 darks used in the master darks. I don't think temperature is that significant with a DLSR? (I vaguely remember somebody doing a comparison with darks taken in a fridge (with camera in a poly bag) & room temp.)

I re-use the Master Dark & Superbias & just take Lights & Flats on  the night. Usually 30 Flats & as many Lights as I can manage.

Cheers
Ivor

[Budgie1]

2 hours ago, rnobleeddy said:

- the advice not to bother with darks usually comes paired with the advice to dither instead (and large dithers at that!)

- if you do take darks, they need to be temperature matched very closely, and plenty of them

I forgot about dithering, yes that was mentioned, as was increasing the dither.

The temperature was one of the reasons for not using darks with a DSLR, because there is no way accurately match the temp.

I tend to take all the darks at the end of a session but on my last session the sensor temp in ATP was showing 16°C at the start and 13°C by the time I came to take the darks. The only way I can think of to get near the requirement it to alternated lights & darks during the session.

[alacant]

Hi

Siril has a nice dark optimisation (for non temperature matched darks) algorithm which works well reducing banding with our 450d, but adds more noise with both our 1200d and 700d. Camera dependent?

I'm a fan of losing the darks and going with dither.

Best to experiment perhaps.

Cheers

 

[Seelive]

As far as I understand, darks will only reduce the coherent noise of an image  but will always add (statistically) to the random noise of an image. Any sort of movement between images (dithering) and some form of statistical stacking seems to me to be the better option.