Techniques for working under LP skies?

[Joel Shepherd]

I'd like to trawl the collective wisdom of SGL for the techniques folks use to get the best AP results possible under significantly light-polluted skies (think Bortle 6 and up). I mostly interested in (L)RGB DSO imaging with a CCD but, hey, take it where you will.

To give an example of why I think I'm missing some clues, see the attached. This is M63, composed from an hour each (12x300" each) of RGB, 24x300" of L and 6x600" of Ha for "sparkle": so, six hours of light. I processed it in PixInsight following standard routines for stacking each channel, combining and noise-reducing. The background is (to my eye) unpleasantly mottled, which I think is due to the RGB stacks being relatively noisy despite my best efforts to use the rejection algorithm that gave the lowest overall noise. I could clean it up by moving the dark point, but that would also eat into the "glow" around the galaxy which -- having reviewed some other photos of the same subject -- seems legitimate: M63 seems surrounded by a light unstructured fog of stars.

I feel that for 6 hours of light, it should be possible to get better, less grainy and mottle results, which to my small mind means either the LP is wreaking havoc or my processing technique is poor. It's quite possibly the latter, but I'm especially interested in what I can do about the former: what can I do to improve the quality of the original data?

(Technical details: taken with a SW Esprit 80ED APO, Atik 460ex, Baader LRGB filters, Astrodon 5nm Ha filter, poor man's dithering (manually tweaking scope position every 3 exposures), bias subtracted as dark.)

[des anderson]

Look... I don`t know anything about deep sky photography,but if i took an image like that,I would be the proudest bloke on EARTH.Des

[Floater]

Des and I agree.

??

[MattJenko]

Mottling can come from noise reduction in processing, especially if you used ACDNR, which is tricky to use and avoid the effect. The noise reduction in MultiScaleLinearTransform might be a better bet.

Also, try MaskedStretch for the stars to prevent core bloating. You can combine the tighter stars it produces with the galaxy stretched using the standard HistogramTransformation if you don't like what is has done to the galaxy.

Also, get the noise reduced in the L layer and don't worry too much about it in RGB layers, those can be processed more aggressively.

[Pompey Monkey]

You've got a similar setup to me*, but you don't say what mount you have?

Why do you use "poor man's dithering"? If you are guiding, then you must have some software control somewhere?

*Check my signaturere, and update you own for best response to any questions

 

[Joel Shepherd]

Thanks for the thumbs up, Des and Floater. :-)

Signature updated.

@MattJenko : thanks for the tips. I'll give the details of the processing below, but ... it occurred to me on the third go-around that noise in the L channel was a much bigger problem than noise in the other channels -- i.e. a really clean L image would probably make up for a lot of noise in the RGB layers -- and that's one of the things that's making me wonder whether I need to do something differently during data collection or processing.

In a nutshell, processing was:

• Calibrate lights in each each channel.
• Cosmetic-correct black/white pixels lights (gently, something like 0.05% pixels affected).
• Register all lights together.
• Stack each channel. I tried multiple rejection methods for each and generally picked the one that resulted in lowest overall noise, except for Ha where I traded a bit of noise to get rid of some odd hot pixel artifacts.
• Crop all stacks.
• DBE on each stack.
• Combine RGB with background neutralization and color calibration.
• Re-extract channels and integrate Ha with R the "Vincent Peris way". Re-combine RGB.
• Deconvolve L
• Noise reduce L with MultiscaleMedianTransform.
• Gently noise reduce L with ACDNR (three scales IIRC), adjust L dark point.
• Stretch RGB-combined and L with histogram and curves.
• Gently blur RGB-combined with MultiscaleLinearTransform (three scales disabled).
• Apply HDRMultiscaleTransformation to RGB and L.
• Combine L with RGB, with chrominance noise reduction and a bit of saturation.
• SCNR and a bit more putzing with the dark and median point.

In general, stretching seemed to increase noise levels significantly. Maybe that's expected, but it was really hard to get the fainter detail in the luminance without making the background increasingly grainy. ACDNR seemed to reduce noise by an order of magnitude with a little blobbing in the background, but the result with ACDNR was better than without.

Would "real" dithering help significantly? More lights, even with shorter exposures (e.g. am I better off taking 20x180" than 12x300", especially given the LP is going to drown the faintest details anyway)? More aggressive noise reduction across the board?

 

[MattJenko]

Few things you could try. 

• Do SCNR after colour correction, at linear stage.
• Stretch earlier, worry about noise after.
• Don't use ACDNR, use TGVDenoise with no mask, use previews to ensure it doesn't hide galaxy details before applying to whole image.
• HDRMultiScaleTransform can re-introduce noise as well, so be careful.
• I also tend to combine my L earlier in the process, but have had success doing it later like you too.

 

[brrttpaul]

sorry to jump on this thread but I still dont understand how an hours exposure isnt blowing out the stars, if I do 5 mins exposure my stars are all blown out I must be missing something. I would be a very happy person with that

[MattJenko]

2 hours ago, brrttpaul said:

sorry to jump on this thread but I still dont understand how an hours exposure isnt blowing out the stars, if I do 5 mins exposure my stars are all blown out I must be missing something. I would be a very happy person with that

The OP's hours exposure time is not all in one go, but multiple 5 + 10 mins subs

[Joel Shepherd]

Those sound like useful ideas to try: thanks Matt. I haven't played with TGVDenoise as yet. Also, if you use either, do you have a preference between MultiscaleLinearTransform or MultiscaleMedianTransform for general noise reduction? I've been using MMT due to a warning I've read that MLT can be hard on fine detail ... but after playing with them side by side last night it seems as though MLT deals with background noise (at least with the non-trivial amount I have) more effectively than MMT.