Help with noise reduction - use my images

[joelshort]

I feel like I'm starting to go backwards with PI.   

I'm having trouble figuring out noise reduction so that it doesn't give me the dreaded blotchy, plastic look.  Yet it seems like if I don't push the noise reduction too far I get black holes and a salt/pepper kind of image after noise reduction.  I've mostly been experimenting with MLT and using the linear mask. 

I would love if some of the gurus here could take my images (narrowband Ha/SII/OIII master, DBE'd) and work your noise magic on them.  Then tell me what you did! 

https://www.dropbox.com/s/tg4gzjayypzz1vd/Sh2-274-1.zip?dl=0

Thanks in advance.

[johnrt]

Well you have captured some nice data there. I had a play and found there to be little noise. I combined the images to SII:HA:OIII for an RGB, then combined all 3 images using Pixelmath to make a mono luminance image.

Once applied over the top with some colour noise reduction in the LRGB Combination tool I didn't think any further noise reduction was required.

A little noise in an image is a good thing, it keeps a natural feel and all photos have that little grain to them. A slightly noisy image beats an image with noise reduction over done everytime.

A mantra quoted to me recently I really liked was - "It's noise reduction, not elimination". 

[joelshort]

Would you mind posting your result?

[MikeOates]

This is what I did, but there would normally be other steps before making non-linear such as deconvolution and a lot more time spent. You don't include the image that you obtained so I don't know how my example compairs.

The process I did was...

Integrate the three images with no rejection to make a luminance image.

Manually HistogramStretch by starting with the SFT, but before applying, I moved the shadow and midtone sliders apart to reduce contrast, which reduces apparent noise. And I think this is the crucial stage, I have only just come to this conclusion that perhaps I was over stretching my images.

LinearFit with Ha as reference.

PixelMath to combine the colour images:
R: SII_DBE*0.5+ 0.5*Ha_DBE
G: 0.15*Ha_DBE + 0.85*OIII_DBE
B: OIII_DBE

BackgroundNeutralization with a reference image of a small preview of the background and set Upper limit to 0.003

Manual HistogramStretch as with the luminance.

Noise reduction with ACDNR on the colour image as below:



Make a copy of the luminance, use convolution of 2 to smooth it. Then make more contrasty with Curves to produce a SN mask which is then applied to the colour image.

With this mask use Curves to increase saturation in the highlights, then invert the mask and decrease the colour saturation in the background but not as much.

Using the same SN mask apply to the luminance. At this point one would perform sharpening etc but for now I just increased the contrast and used HDRMultiscaleTransform.

Now do a LRGBCombination and final contrast tweaks. I know this could be improved a lot, possibly with a little noise reduction on the luminance and/or on the final image.



The full image is here: http://www.mikeoates.org/pi/final_joelshort.jpg

I hope this helps.

Mike

[topboxman]

Hi Joel,

Here is my quick and dirty process of your image.

Peter

[joelshort]

Thanks Mike and Peter.  I'll have to try these settings out.  I think perhaps I am over stretching and over doing the noise reduction.  Here's my processing with "that other program":
http://www.buckeyestargazer.net/Pages/Nebulae/Sh2-274.html

I couldn't get close to this with pixinsight. 

[pfile]

color-wise you can definitely get there with Rick's color mask script: http://pixinsight.com/forum/index.php?topic=7751.0

on the NR i suffer the same problems, can never get a result that i like, at least from the more modern methods of TGVDenoise and MMT. ACDNR on stretched images usually goes okay for me.

rob

[joelshort]

I have used Rick's color mask script before with good results, but for some reason on this image it didn't work as well.  The colors of the Medusa are unlike most other narrowband targets I've imaged.  I didn't get the typical magenta cast I usually get when stretching.  In fact the stars were actually green after stretching. 

[msmythers]

Joel,

Did your supplied images have DBE applied already? I hope someone can talk to this but when I stretched those images it looks like DBE was not applied to well. The edges on 2 of the images have problems. Normally I would crop this type of problem area before using DBE. I'm not sure you could get a smooth, strong stretch with these images. I may be way off base.


Mike

[joelshort]

I can't remember if I cropped the images before DBE or not, but that would certainly make a difference.  I didn't see those edges because I didn't stretch it that hard.