TVG Denoise - creating unwanted contour lines around galaxies

[dpaul]

I've been using TVG denoise to remove the graininess of the background sky. Lets assume for example I have an integrated set of light frames but its still linear and to protect a galaxy structure's sharpness.

So I create a clone and stretch the clone/do background noise removal to get a dark sky background, then invert to create a mask. I then use this mask to protect galaxy structures from being softened by the denoise process.

I'm using only mild settings - strength, edge protection, smoothness all at 2.0.  The background gets smoothed out nicely but I then notice faint contour lines around the galaxy.

How can I avoid this. am I doing something wrong?  Also is there another way of getting what I want with a different process?

Thanks

David

[dpaul]

Further to my note below, I managed to avoid most of the contour evidence with the right background darkness and background extraction.

Attached is an image of M65 - it was with a first quarter moon.

David

[John_Gill]

Hi,

I see in your image there is a contour line around the galaxy core.  Perhaps try a mask on the entire galaxy and then apply the TGVD process.

Look up
John

[dpaul]

Hi John,

I did use a mask - there is actually some subtle banding around the outside of the galaxy too - more obvious when the original higher quality image is opened up.

Regards

David

[Carlos Milovic]

If you are using the mask, it is not the same as removing the object as it wasn't there. Let me explain further, if you have a very strong light source (i.e. a star) and you have no edge protection, it will be blurred and the surroundings will have larger intensities. If you use a mask to protect the star, you'll blur the image, and then replace the star with the old data. You'll end up with the same increase in brightness of the background.
So, what you need to do is to fine-tune the edge protection, to avoid such artifacts. Using the mask as Local support would also be beneficial since it acts internally to avoid the data "leakage". It is different to a standard mask.

[ChoJin]

For a good edge protection value you might start with the following:

- create a preview over a background area without stars or other stuff
- run the statistics process
- in the parameters of the statistics process add the standard deviation.
- make sure the values are displayed normalized
- use the stddev value for your edge protection value

[dpaul]

Hi Cho,

Thanks for the note.

Just to confirm a few thing:

1/.In statistics do you mean avgDev is the standard deviation?

2/. When I then ''use this value as the edge protection value'' do I simply adjust the exponent and the slider to match that number? 

Regards

David

[dpaul]

Hi Carlos,

Thanks for the reply -
Before I ask some questions related to your note, I have a few fundamental questions:

1/. Which of the following is best:
(a) Use TVG denoise on the R, G, B and L separately?
(b) Channel combine the RGB and then use TVG denoise and separately on the luminance?
(c) Complete the LRGB combination and then (whilst still linear) use TVG denoise?

2/. What should be the order of priority out of the following (and why?):
(a) TVG denoise
(b) MT to reduce star sizes
(c) Deconvolution

3/. When creating a mask to protect a galaxy (for example), I make a clone then stretch it, remove background with DBE then invert it. Should the mask be a clone of a color image (whether single or RGB) or just extract the lightness?

4/. When creating a mask (as in question 3), currently I'm not blurring the edges in any way but I probably should be. I think I should use ''range selection'' on the mask and play around with the 'lower limit' and the smoothness (to blur the edges) - is this the best way?  Also what about star masks, do they need to be blurred?


Now my final questions relates to local support:

5/. When I check 'local support' which support image am I picking, presumably the mask?

6/. assuming I leave midtones, shadows and highlights as default, what about ''noise reduction'' should it be left as zero default?


Many thanks in advance.

Regards

David

[ChoJin]

yes, avgDev, you either just copy the value to the box where you can see the full value (the one on the full left), or you copy it as float and exponent. I prefer to just copy it as is to the left box (and the two other boxes will update automatically)

With my very very limited experience:
- using TGV on linear gives me weird blotches, even after tuning the values for hours. I'm probably doing it wrong. I therefore used MLT following the M81&M82 tutorial from http://www.pixinsight.com.ar/en/info/processing-examples/48/m81.html, and got much better results.

- on non linear, I got quite nice results with ACDNR (using lightness mask enabled on both lum and chroma), I could match the results with TGV, but it wasn't really better (at least to my opinion on the data I tried on) to justify how slower it was. So even after hours of TGV Tuning, I went back to ACDNR with my default values.
Also note that I had run MLT during the linear process, hence I probably get better results on non-linear because of that.

[dpaul]

Thanks Cho

Regards

David

[dpaul]

Hi Carlos,

Did you get chance to see my reply below from February 27th -





Hi Carlos,

Thanks for the reply -
Before I ask some questions related to your note, I have a few fundamental questions:

1/. Which of the following is best:
(a) Use TVG denoise on the R, G, B and L separately?
(b) Channel combine the RGB and then use TVG denoise and separately on the luminance?
(c) Complete the LRGB combination and then (whilst still linear) use TVG denoise?

2/. What should be the order of priority out of the following (and why?):
(a) TVG denoise
(b) MT to reduce star sizes
(c) Deconvolution

3/. When creating a mask to protect a galaxy (for example), I make a clone then stretch it, remove background with DBE then invert it. Should the mask be a clone of a color image (whether single or RGB) or just extract the lightness?

4/. When creating a mask (as in question 3), currently I'm not blurring the edges in any way but I probably should be. I think I should use ''range selection'' on the mask and play around with the 'lower limit' and the smoothness (to blur the edges) - is this the best way?  Also what about star masks, do they need to be blurred?


Now my final questions relates to local support:

5/. When I check 'local support' which support image am I picking, presumably the mask?

6/. assuming I leave midtones, shadows and highlights as default, what about ''noise reduction'' should it be left as zero default?


Many thanks in advance.

Regards

David

[wimvb]

Have a look here:

https://jonrista.com/the-astrophotographers-guide/pixinsights/effective-noise-reduction-part-2/

As for the standard deviation in the statistics tool; if you press the wrench button, you get a menu from which you can choose several statistical measures. Std Dev is in there.

Hope this helps,

[dpaul]

Hi Wim,

That was a 'very' useful thankyou - explains why and not just how.

It clearly shows the order of priority with linear images:

Debanding if required
Background Neutralisation
Color Calibration
Noise reduction (SNCR then TVGDenoise)

Still have a few questions:

1/When should Morphological Transformation and Deconvolution take place compared to the above (and which first, Deconv or MT)?
2/.Until now I've done noise reduction, MT and deconvoltion on separate R, G and B integrated frames (e.g, 20 integrated Red frames), except of course color calibration which is done after channel combination. So I'm wonder if I should channel combine ''first'' then do all the above?

Thanks

David

[wimvb]

Deconvolution should be done in the linear stage, star reduction in the nonlinear stage. However, it doesn't hurt to go away from common wisdom, and test tools in areas where they weren't meant to be used. At least you'll learn when (not) to use a tool.

My normal workflow for rgb after stacking:
1. inspect the r, g, and b masters. If there is a substantial difference in average/median pixel values, I use LinearFit to "align the histograms". Otherwise:
2. combine r, g, and b masters in channelcombination.
3. Crop any stacking artefacts with DynamicCrop
4. CanonBandingReduction if needed.
Cooled CMOS cameras can have horizontal bands when used with short exposures at high gain. If I do get this, I use CBR.
5. DBE with normalisation checked in the correction field.
6. Background neutralisation.
7. Colour calibration.
Nowadays I use Photometric Colour Calibration, in which case steps 5 and 6 are one process
8. Deconvolution. If I also have a luminance master, I only do convolution on this.
Sometimes I get weird artefacts (coloured pixels) around stars if I do deconvolution on an rgb image. If I don't have an L master, I extract a synthetic L from the rgb image and do a deconvolution on this. Later on I then do LRGB combination, in the nonlinear stage.
Not all images need deconvolution, but if you finetune the deringing and regularisation parameters, you can only target the stars in deconvolution, making them a bit tighter.
8. Noise reduction with Jon Ristas methods, TGV and MMT, or a variation thereof.
9. Histogram transformation on L, Arcsinh stretch on rgb. (But sometimes I find that Arcsinh stretch clips the blackpoint too much. I need to refine this step.)
10. Refining the stretch of L.
11. HDRcompression.
12. Make sure that the histograms of both L and rgb are roughly the same, then use LRGB combination with chroma noise reduction.
13. Further stretch refinements and colour saturation increase.
14. Star reduction using Morphology Transform.
This step not only reduces stars, it also dims them. I make a new star mask that only targets the center portion of stars and use MLT with a positive bias on layers 1, 2 and 3, to sharpen the stars a little.

If the background shows colour noise, I use ACDNR after stretching on chrominance only, to neutralise this.

[dpaul]

Wim

Thanks for the detailed reply, much appreciated!

I was doing much of your methodolgy except the following:

1/ Never bothered with Linear Fit, but I will take note in future
2/.DBE I've usually done 'after' background neutralisation and color calibration but is probably causing ''splodgy background'' so I'll try later instead
3/. Deconvolution I do on RGB and Luminance but I appreciate the importance of the Luminance for detail so I'll just try that in future
4/. HDR Compression I haven't been using - it gives great contrast but seems to darken images a lot but I'll experiment more!
5/. Morphological Transformation I've alway done very early in the linear stage - I'll try it later after going non-linear.
6/. Currently I'm combing the RGB color calibrated image and the luminance image in linear state and then doing Arcsinh (or Histogram Transformation until very recently). However I agree the RGB and Luminance are not well matched (Lum dominates) so I'll try combining them after stretching.

Regarding Arcsinh, I do like this tool - so far I've set the black point ''just before'' it starts to crop pixels then adjust the stretch factor until. Interestingly some of my data shows much less cropping than others so this might be a useful indicator of how good the image calibration was done with the dark master (e.g. not setting optimisation too high).

What I've also tried is a less agressive Arcsinh and then do a slight bit of Histogram Transformation - the former gives better colors, the latter may help contrast (I think).

One question on using masks - there are two ways I've done this:

(a) Take a clone image and use range selection to get a mono image (of say a galaxy) then blurr it a bit.
(b) Take a clone image, DBE and stretch it

The otion (a) probably protects the areas you want to protects better and softer transitions than option (b). any suggestions on this?

Thanks a lot

David