Negative pixels on dark subtraction

[v4169sgr]

Hello everyone,

I am somewhere in the foothills of my learning curve on how to acquire and process images. My main aims at the moment are: (a) acquire nice round stellar images; and (b) properly implement dark subtraction and division by flats, both for the purpose of getting into photometry of variable stars.

My equipment is an 8" Newt (f4.5, 900 mm focal length) on a NEQ6 and a Canon 450D, unmodded, with coma correction. I use Kstars to download the images and write to FITS, and Siril (v 1.2.0) for data processing.

I've chosen M45 for the purpose of learning: a nice wide bright star field, not far from the zenith, with plenty of interest in the background, and which moat important is a very well recorded target. My mistakes and wrong turns should be all the more obvious.

I am conducting star tests at the moment, with the aim of understanding & being able to eliminate as much as possible any deviations from circular stars. On Monday I'd decided to acquire 120 10 second subs of the target, to contrast them with the 60 second subs I'd taken of the same target a week ago. I would not normally suggest 10 second subs for this target, but wanted to reduce any contribution  of tracking errors.

To reduce the data, as a default, I am using the 'OSC_Preprocessing.ssf' script naively without modifications, to see what I can learn. Along with the lights, I also took 120 darks of the same exposure with the telescope cover on but pointing in the same direction, 120 flats of 0.01 seconds each of a light box placed over the end of the telescope tube, and 120 'biases' of the same exposure, as flat darks. 12 GB of data, but so worth it for the learning experience! 🤣

While running the script I noticed Siril complaining that calibrating the lights with the dark and flat masters the script made produced 20 % negative pixel values over the frame, consistently. Investigating further, I found that the median pixel value in the master dark was 1025, while the same on a representative light was between 1030 and 1045, over the RGB channels.

The scripted calibration command is:

calibrate light -dark=../masters/dark_stacked -flat=../masters/pp_flat_stacked -cc=dark -cfa -equalize_cfa -debayer

As far as I can tell, my method for acquiring the calibration frames seems OK to me. Similarly, there are so many eyes on this script that if something was off them it would have shown up by now. The only reasonable explanation I can find is that my subs were so short that systemic contributions to the dark from the bias made a more than is usual contribution to the data levels in the dark.

Next opportunity I get I am going to try some 30 second exposures to see if this makes any sense.

Any thoughts?

[Stefan73]

I'd guess the dark isn't as dark as it should be.  When I've done darks it hasn't been good enough to just put the telescope cover on, or even to take the imaging train out and put a cap on that, due to light leaks.  I've ended up just having the camera with the cap on in a dark garage or the flats end up over correcting.

That being said I think usual advice is not to do darks with a DSLR and use a constant bias with siril.  For my 550D I used

preprocess light -bias"-2048" -flat=pp_flat_stacked -cfa -equalize_cfa -debayer

I'm not sure what number in place of 2048 you'd use.  I'm not sure you'd really need as many as 120 of the calibration frames but it won't do any harm

 

[alacant]

On 17/01/2024 at 10:11, Stefan73 said:

not sure what number in place of 2048

Hi

450d: offset 1024

We do not recommend using in-camera bias or dark frames with this camera; simply subtract the offset. Hence...

... perhaps best to learn to use Siril manually so that you are in control. You'll probably find manual usage is quicker anyway.

You can use EKOS' flats planner to set the desired adu. Place sheets of paper between the telescope aperture and the light panel. To avoid introducing more artefacts, flat frames using a typical light panel need to be at least 1s duration.

No theory here. Just hands on.

Cheers and HTH

 

 

Edited January 18 by alacant

[v4169sgr]

Thanks for the responses.

I've inspected the stretched master dark and can find no evidence of light leak. Additionally I've noted the median statistic of the master dark and the master flat dark - they are very close to each other, confirming this conclusion.

The 450D body is relatively old, and does not have built-in dark current subtraction.

The EXIF CCD Temperature varied in a uniform way between +3 and +7 C over the course of the multiple frames, in the same way both for the lights and the darks, confirming there's no problem with systematic thermal noise offset between the lights and darks.

Using Siril manually is slower for me than running scripts, and there is nothing to suggest so far that  default processing is giving me issues (other than the nomenclature for flat darks, which I can easily adjust).

My working assumptions are that dark subtraction is sensible and necessary, and that I simply didn't rack up enough sky background in the lights of 10 seconds to make a difference from the darks. My Bortle score is 3.6 - skies are very dark.

I hope to go out again and take some longer exposures tonight or tomorrow (while conducting other tests). We shall see!

[dph1nm]

I think this is unavoidable with short exposures or with narrow band filters. I get this in the red channel when subtracting darks from my 1000D 120s exposures taken through an L-enhance. There is virtually no sky signal, so subtracting the dark+bias has to lead to some -ve pixels. One might hope that the software could cope with this!

NigelM

[pipnina]

one workaround that could be implemented is to "flash" the light frames by simply adding ADU to them in post. I belive PixInsight has this option (called I think "output pedestal") but I am not sure if Siril has this feature.
I DID have a python program someone wrote for me that took FITS files and would let you add/subtract ADU from them singularly or in a batch but I can't get it to work any more.

By adding a flat ADU to the light frames you do not upset the effect of dark frame subtraction but get to avoid the black clipping.

This works as long as your initial image is not black clipped. One way to avoid this with narrowband I should think is to shoot a high ISO like 12800, which will cause even small samples of light to push all of the pixels above 0 in a reasonable exposure time.

[v4169sgr]

Thanks again for all the replies ! 😎

I did get some time at the scope again last night & while messing about with the focuser clamp collected some 60s subs + all the calibration frames. No good at all for lights stacking because of the (quite deliberate) errors I was introducing, but it did give me a chance to corroborate my working assumption.

Longer subs - especially near the moon 😆 - do definitely make the negative pixel warnings on dark subs go away & that's with the same image reduction script.

That's OK because I would have no intention whatever of using 10s subs. I was only using them at the time to get a feel for how tracking errors might be influencing the image (they aren't, at least not up to a minute).

On other points raised: it was quite notable that random noise increased in the darks for longer exposure times (as expected) but also that the median values hardly budged, which - combined with a visual inspection of the master dark - confirm that merely capping the scope doesn't let light in. And I was pointing almost directly at the moon - it was BRIGHT!