Noise Reduction in Workflow

Hello Mark,

Just for reference, this is what I am doing in Photoshop:

Moon images
Apply a Gaussian Blur (0.5) before unsharp mask, but after a high pass.

DSO images
Duplicate the layer and apply a Blur/Average filter, then blend the layers with Lighten.

Cheers,
Robert

Hi Mark,

I find that it depends on the noise reduction algorithm and your workflow. 

For me:
- If I have DSLR images and I do the nonlinear workflow, then applying a modest denoising with Rawtherapee even before the individual subs have been stretched is a must. Mind you, if you are just curious you can actually extract a crazy amount of signal from the sub using other tools in Rawtherapee e.g. with Retinex. But the subs become less "integratable" (not sure that's a word, ). 

- If I do a standard calibrated workflow:
-> SCNR helps removing excess green. Sometimes it works better if you apply it before stretching, sometimes afterwards. Sometimes it is even needed before any kind of color calibration. I guess it depends on the amount of excess green.
-> MultiscaleLinearTransform works on a linear image. MultiscaleMedianTransform works on linear and non-linear. But I haven't been using any of those two for years.  
-> I find that ACDNR works better after the image has been stretched, but you have to raise the star protection thresholds because the defaults are for linear images where star brightness begins at 5%. I do not use that process too often.
-> Darktable's Astrophoto Denoise reigns supreme, but obviously it is applied at the very last stage, after the integrated image has been stretched exported from PixInsight as a 16 or 32-bit TIF. 

Usually with Pixinsight I do only SCNR at some point that I determine by trial and error, followed by Astrophoto Denoise when the image has been brought into Darktable.

Hi Mark!

I don't make any claim to expertise, so caveat emptor! At this early stage of my AP development, I tend to do whatever I'm going to do quite early on. I do an SCNR almost immediately to get rid of the noisy green pixels. I do that in Siril. Once I'm in PS and quite early on in that workflow,  I do some minor NR using ACR--specifically the 'Texture" slider and then I pop into the Detail tab of ACR and make a few additional adjustments there. I rarely go more than '10' on the NR slider and sometimes as high as '50' on the CNR slider. These are global changes, of course. I find I rarely need anything more, but if I do I can mask and adjust as needed.

I have also used Topaz Denoise, and sometimes still do, but that would also be early on in the flow. 

In any case, I figure that noise needs to go as early on as possible, before I start enhancing and modifying it and getting it confused with real signal.

Daniel

Hi folks:

Thanks for all the responses! I should have added, I'm not using PixInsight, so all the PI-specific workflows are a bit outside the realm of possibility for me right now. I'm renting Astro Pixel Processor, which currently doesn't have any NR functionality. I guess what I was wondering was more theoretical and less application-specific.

I just finished reading the post-processing chapter in Michael Covington's Digital SLR Astrophotography, and he claims that it doesn't make much difference to reduce noise in linear before gamma-corrected, but I'm sensing from your responses that it helps to get it in, at least minimally, early on in post-processing and I'm inclined to agree.

If we are being application-specific, I do have Siril, which I used before getting APP, but I find I can't shift FITS files between the two, so Photometric Colour Calibration & SCNR is off the plate at least until I get that figured out. Opening a FITS file in APP which it didn't generate itself, or which was edited in another app, produces a black image which cannot be stretched. I've been approximating SCNR with Star Colour Calibration & by reducing green saturation in Photoshop using the Camera Raw filter.

Thanks to everyone who has chimed in!

CS,
Mark

I have the advantage of reasonably datk skies (mag 20.4), and I do most of my noise reduction early in the process: I gather more data. During post processing I hardly do any noise reduction, except for removing a green cast with scnr in pixinsight. On occasion I remove chrominance (colour) noise in the rgb image prior to lrgb combination.

Hi mark, I also use Photoshop ( v5) for my post processing.  The main PS tool is Filters-Noise - Reduce Noise.

Your question was addressed to when to apply noise reduction and you got a lot of responses to that question.  But I would like to be more general in my response.  I looked at some of your images and see you are using a non-cooled un-modified Canon 3Ti on and iOptron Sky Guider.   Most of your images seem to be in the 1 to 3 total hour accumulation range.

You mentioned your post that you might get varied responses, so, here it goes.  Signal/Noise ratio will vary based on acquisition approach and equipment used. Not to mention the most important, the level of sky glow.  Your question is more directed at short acquisition using a non-cooled un-mortified DSLR.  Your acquisition will render a poorer S/N ratio than someone using long total exposure time with a less nosier system.

So yes, if you S/N ration is poor, then stretching it early in post processing will likely create a S/N problems making it difficult to bring out the signal latter.  IF your S/N is very good, then doing a early noise reduction may be counter productive because you may over soften the fine detail and not be able t recover it latter.

I have learned that one size does not fit all with image post processing.  Data can vary a great deal and will demand different approaches.  It can vary even with the same equipment at the same site based on transparency and seeing.   But given different equipment, different acquisition techniques and different sites, what works for one may not work for another.

I have used a DSLR rarely and use a modified Canon T3.  So, I am fairly un-familiar with processing OSC DSLR data.  I use cooled CCD cameras and primarily use narrow band filters.  But some general approaches may apply.   If the S/N is very poor them one must accept that the data will not allow much stretching of the data, and you discovered early noise reduction may help.  If the S/N is good then the data may allow aggressive stretching and little noise reduction latter in the post processing steps.

BUT, I would like to rephrase you question to not when, but where.   The S/N will not be equally distributed throughout the image.  The stars will likely have much better S/N, as the brighter areas of the subject.  The fainter/darker areas along with the background will likely have poor S/N.  By processing all equally, you are trying to turn right and left at the same time.  You want to stretch/bring out  the better S/N and smooth/hold back the poor S/N. The image must be thought of in zones and each zone processed differently.    Certain algorithms in Pix Insight and other software can do this.  The  PS action plug-ins do this.  But you can do very effectively yourself using Mask in Photoshop.  Mask can seem overwhelming, but is actually fairly easily ones you get the hang of it.  Mask allows you to selectively sharpen areas of the image and smooth (reduce Noise) in other areas.

My suggestion for handling the noise in your images is to first, if possible obtain longer  total session data.  Second, I would suggest learning to use Mask rather relying on when to apply a noise reduction to the Whole Image.  Adam Block's PS tutorial are excellent.  I know it is hard to lay out $100 on a download, but it may be the most beneficial thing one can do to improve image processing.

I hope this helps some, Lynn K

I use PS, my workflow usually looks like this in some fashion:

Initial levels
Levels with threshold mask 
Curves
Gradient (maybe a level adjustment)
RAW Camera filter (minor adjustments)
Noise (Topaz Denoise)
RAW medium adjustments 
Space & Deep Space Noise (Astrophotography Tools Actions)
then I check star reduction, star color, etc

So.. I typically do noise reduction early but always after my initial curve adjustment. I also use layers so I sometimes go back and start over and use noise earlier and later just to see if there is a major difference