Background noise guidance...

[footbag]

Hello,

I have been using PI for a couple of months and I'm getting used to it.  One issue I'm having is that I'm finding my RGB data to have a noisy background.  I have been taking 10 subs per channel with my STF-8300m.  I see lots of people getting better results with only a couple of RGB subs, so I think I'm doing something wrong. 

What would be the best way to get a smoother background?  I suspect it's a matter of using either ACDNR or TGVDenoise, likely with a luminance mask, but I'm not sure.  Or at least it hasn't given me the results I've been expecting.  I can get rid of the small scale noise, but it tends to get blotchy.   Is it better to do background NR on a linear image?  Is there a tutorial or walk-through?


I've posted a link to my Astrobin.  The three most recent images are all processed using PI.  I desperately need a strategy for backgrounds. 

http://www.astrobin.com/users/Footbag/

Thanks,
Adam

[sreilly]

I'm assuming the last three are actually the first three posted on the top of the page? Looking at these I'm not sure how smooth you're expecting the images to be. Plus it's not only the amount of time but the quality of the frames being used that have a lot to do with the end result. Use frames that have say a FWHM of 4-5 and the image won't be so good. Use HWFM of 2-3 and you'll see a big improvement. Are you culling your images? Have you tried using the SubFrameSelector under Scripts/Batch Processing? I've just started using this as in the past I've used CCDInsoector and get very similar results.

-Steve

[footbag]

OK.  I do have a lot of light pollution so my original frames do have gradients.  But DBE gets rid of these pretty well.  After DBE, though, my background gets very noisy.  I have read that it's because the STF tool stertched it slightly differently after DBE. 

But, this is where I'm stuck.  I should be using ACDNR or TGVDenoise, right?  My background noise seems random, but when I try to remove it, it just gets blotchy.  I also have trouble getting it dark enough.  Blotches tend to come out just as I'm darkening the background. 

[pfile]

to be clear, what's going on is that most of the signal you've collected in the background is skyglow. when you subtract a smoothed version of that skyglow from the image, you're essentially left with background signal; the background signal is a small component of the signal you collected and thus it's really noisy.

actually, i don't think your backgrounds look all that awful for an LP situation. i guess you can always go for more and more integration time.

rob

[bianry]

I would be very happy I could get my backgrounds as smooth as yours 
What Steve writes I find interesting. If that is correct it means it would be better to spend 500 frames on 1 subject and not chase after 'the next one'. Must remember that.
And while waiting for clear skies you can always hone your PI skills on the ultimate (free) 'Rent-A-Scope', Hubble.
http://hla.stsci.edu/hlaview.html

regards

Mats

[footbag]

OK.  So the noise I'm talking about is a very granular chroma noise in the background after my RGB combine.  It very well may be caused by light pollution.  But, what tool would you use to remove this noise from the background?  TGVDenoise or ACDNR?  I used ACDNR on the last image, and it did OK.  I think I can get TGVDenoise working better but I cannot get local support to work without ringing the stars. 


Here is a picture of the background noise immediately after combining the RGB channels.  So if I shoot from darker skies, I own't see this?  I just thought it was a CCD RGB combination thing.

https://www.dropbox.com/s/s7m68zvqte6k3m4/chroma%20noise.jpg

[Alejandro Tombolini]

Hi Adam,

I would use ATWT. See "Markarian's Chain with DSLR" in "Processing Examples" that could be helpfull.

In the crop you have posted try deleting layer 1 and 3. Depending on your object you may need a good mask to protect object and stars. 

Saludos,
Alejandro.

[pfile]

ya - alejandro's method is the way to go here. you have to very aggressively mask the foreground and stars, because this method really kills color.

rob

[footbag]

Thanks guys!  I'm having luck now.  I guess my masks weren't aggressive enough.  I was generating lots of artifacts in my background.

[jerryyyyy]

Yes, looking at your images on Astrobin I think you are doing very well.  I am at about your level of experience and here are a few tricks I have tried.  [Your mileage may vary]

1.  I agree that TGVD is hard to use.  I have had better luck with ATWT as described above.
2.  A little trick that seems to help me is that I have taken to using the MaskedStretch script to median of 0.12 with 100 iterations.  Then I use ATWT on layer 1 and you will see the histogram take a big move to the right, then you can clip to 0% or 1%.  This is a bit brainless.. but I am getting a bit old and do not need to rediscover the wheel with each image.
3.  When I do use a custom mask I have learned I better be pretty aggressive to get the crud out of the background in the mask.  If there is crud in the mask, it will leave holds in the masked image and seems to mottle. 
4.  When I started doing this Pfile suggested I start on a bright object (M101) and collect 5X more images than I thought necessary for LRGB.  I collected 17h and finally got a decent image... someone said more time on one target is better than less time on multiple targets... I sorta learned this after getting through the Messier List.
5.  Finally, I just bought a used Tak 180ED at F2.8.... 4X the light bucket than my old scope.    Ask me about collimation...
6.  Oh, I also do DBE on separate channels before combining RGB...

Cheers and keep at it. 

[bianry]

3.  When I do use a custom mask I have learned I better be pretty aggressive to get the crud out of the background in the mask.  If there is crud in the mask, it will leave holds in the masked image and seems to mottle. 

When you talk about an aggressive mask in this thread, do you mean a mask that is more binarized so to speak. A mask made from the luminance is always kinda pink over the background and more red on the stars and the core. I tend the use histogram to clip the shadows part and stretch it to get a more black/white mask. Is this what you are referring to? Or do you mean to make a more smooth mask?

regards

Mats

[blackdragon72]

Hi Adam,

I would use ATWT. See "Markarian's Chain with DSLR" in "Processing Examples" that could be helpfull.

In the crop you have posted try deleting layer 1 and 3. Depending on your object you may need a good mask to protect object and stars. 

Saludos,
Alejandro.

Alejandro,

Do you recommend this step in linear or non-linear, or both?

[Alejandro Tombolini]

You can do it in both with ATWT, but for chrominance it is beter if you have already stretched and saturated the image as the noise become more evident.
Saludos
Alejandro.

[jerryyyyy]

One question (the devil is in the details).  Why levels 1 & 3.  I usually do 1.  Why skip 2 if you do 3?  Maybe I do not understand the Procedure. 

[Alejandro Tombolini]

Hi jerryyyyy
I downloaded the crop posted and by try and error saw that layers 1 and 3 removed quite well the noise.
Depending on the kind of chrominance noise you may want to work over different layers, even the biggest.
Saludos,
Alejandro