Help with M31 NBAccent (Ha) processing

[decay]

Hi all,

I’ve tried to accent my M31 ( viewtopic.php?f=11&t=2703 ) with some Ha using NBAccent like Jochen and Ivo suggested in some posts here. Therefore I captured about 1.5 hours data with my L-eNhance duo band filter (other specs like in the above mentioned post regarding M31 visual spectrum image).
Somewhere I read, that duo band imaging would be possible with moon on stage, but possibly was not meant that the 50% moon shines straight into your tube. So Ha data shows a pretty obvious gradient.

Both ASTAP stacks were aligned with DSS. Processing was quite straightforward. Compose module [L + Synthetic L From R(2xG)B, RGB (Color from OSC/DSLR)], the visual stack loaded into the RGB slots and duo band stack into NB Accents File, NB Accents Type set to Ha/S-II from NB filter. Then AutoDev, Bin, Crop. Due to the gradient in Ha data I had to do a more aggressive WIPE then necessary for the visual stack only. (This was recently discussed in Stefan’s thread: viewtopic.php?f=6&t=2687 .) But I think, the more aggressive WIPE did no harm. Then AutoDev with a large ROI leaving out only the borders, Contrast, HDR, Sharp, SVDecon.
Yes, and then NB Accent. Ha is very noisy and so I had to use Threshold to cut off the noisy background. Unfortunately I was not able to get a decent result. I tried a lot over and over again with different ROIs and tweaking all parameters on both NBAccent steps. It turns out, that best results are achieved with lowering Brightness Correlation and Gamma, but pushing Strength. The result always looks more or less like this (only luminance for demonstration purposes):

tamed-nr.jpg (107.08 KiB) Viewed 4962 times

(I already posted that image in my other thread viewtopic.php?f=4&t=2716 )

The Ha knots came out quite fine I think, but there is always a strong red hue between the outer arms and the bulge (like a pink donut) . The bulge/core itself is pretty white, as expected.

I’ve no idea, what’s going wrong here. Of course, Ha data is weak and noisy, but is that the problem in this case?

Has someone an example of an image of M31 processed with ST and NBAccent?

If someone would like to take a look, the stacks may be found here:

visual RGB stack
https://c.web.de/@334960167135216273/NK ... NaZwQEyrhg

duo band
https://c.web.de/@334960167135216273/c9 ... 43KikGukgQ

Any help would be much appreciated.

Thanks & best regards, Dietmar.

[hixx]

Hi Dietmar,
I used NBA on M33 before.
To avoid the red hue in the arms, You need to limit detail size --7to 100 px or so- experiment)
This will limit the NBA to the knot size only. Also, You can set a higher threshold to avoid "red noise" creeping in....
If You have a Duoband, You could also try to set accent type to duoband Ha/O-III. This should also bring down the red hue in the core and spiral arms, as these should also carry energy in G&B in a duoband shot, resulting in a more neutral NBA tint
Cheers,
Jochen

[decay]

Thank you, Jochen.

To avoid the red hue in the arms, You need to limit detail size --7to 100 px or so- experiment)

Yes, I did that, but reducing to much wipes out everything, including the 'knots' as well

Also, You can set a higher threshold to avoid "red noise" creeping in

I already use a high threshold, more would wipe out the Ha areas. I thought about noise being the problem as well, therefore I posted my question lately with my other thread. But probably noise is not the problem? M33 works well, as we can see in documentation as well. I wonder if with M31 there's something special, which prevents NBAccent from working well.

If You have a Duoband, You could also try to set accent type to duoband Ha/O-III.

I've just tried, but the result is even worse In order to demonstrate what I'm talking about, here's a screenshot with intentionally pushed Strength and Gamma settings:

pink-donut.jpg (313.39 KiB) Viewed 4946 times

Thanks again, Dietmar.

[hixx]

Hi Dietmar,
I think I used detail sizes around 20-30 px or so. You want to figure out the size until the knots are the only objects left. Obviously size will be depended on pixel resolution, object size etc. Balancing Threshold vs. Detail size until You are left with the knots only, seems to the best strategy probably. Start with detail size 100 or so, then slightly shift the threshold first until You see some areas disappearing a bit , then slightly reduce detail size etc. Repeat the process 2 or 3 times until most of the areas are gone, just the knots left.

The only thing special about M31 I can think of is its huge brightness. Maybe the NBA dataset should use a bit less exposure length / ISO or weaker stretch. You should get rid of most of the bright areas. The brightness might also be the problem why the duoband / Ha & O-III trick doesn't work well. The green channel seems to have too much energy, but remember this can be adjusted in NBA as well, just set the color rendering to red for Ha and chose between teal and cyan for O-III. I'd suggest red & teal (=yellow) to avoid washing out colors into white (=red& cyan).

In duoband mode, there is also a color balance parameter in NBA where You can shift towards Ha, so the teal/green does not become too prominent.
It might just need a slight dose of it to pride the galaxy a natural looking yellow.

I am yet to find a clear night or two for my actual M31 version. The best shot I have so far is an LP filter stack using an unmodified Sony A7 II from 3 years ago or so.

Cheers,
Jochen

[decay]

Thank you much for your detailed reply, Jochen.

I will give your processing advices a try - tomorrow evening I hope.
I really tried a lot already, but the result always tends to be that pink donut - red hue in the 'wrong' areas, between the outer arms and the core. So I've little hope that this is a matter of (wrong) settings in NB Accent.

The only thing special about M31 I can think of is its huge brightness.

Yes, that's my thought a well, therefore my question if someone has an example of M31 processed with NB Accent.

The brightness might also be the problem why the duoband / Ha & O-III trick doesn't work well.

I'm not sure, if O-III should play here any role at all? But yes, O-III is really strong in the inner parts a the galaxy. Maybe this is a problem for the internal processing of NBAccent?

As said, I will give your processing advices a try and I will also think over your remarks concerning OIII / duoband processing - maybe I'm missing or misunderstanding something right now.

Thanks & until then, Dietmar.

[hixx]

I'm not sure, if O-III should play here any role at all? But yes, O-III is really strong in the inner parts a the galaxy. Maybe this is a problem for the internal processing of NBAccent?

Hi Dietmar,
O-II actually doesn't play any role in the center of the galaxy, I guess. But the energy in G & B channels passing through the filter's "O-III" band does. The idea is not to capture real O-III in the galaxy but to balance out the pink/red donut into a more Yellow/white donut.

Thinking of the donut as such, it might result from a setting in local stretching (Contrast / HDR), which gives the midtones too much dynamic headroom.
I remember having the same donut in my shot (without NBA) You may want to try different HDR presets and include both core and outer arms as well in AutoDev's ROI - I know setting ROI can drive one mad on M31...

So my strategy to fight the donut would be 2-fold: 1) Mitigate donut energy by optizing stretch 2) Mitigate donut visibility by shifting hue towards yellow using the "O-III" trick.
Good luck,
Jochen

[decay]

Hi Jochen,

Thinking of the donut as such, it might result from a setting in local stretching (Contrast / HDR), which gives the midtones too much dynamic headroom.

Just a preliminary feedback: I omitted Contrast and HDR from processing and the result is: a larger donut. Of course, this looks better, but I'm not sure, if this is good news, as it shows that NB Accent processing depends on the (local?) stretch. I'm not sure, if this should be the case?

pink-donut-2.jpg (267.55 KiB) Viewed 4877 times

I'll keep on trying ...

Best regards, Dietmar.

[Mike in Rancho]

Hi Dietmar,

Hmm, I'm not super sure what actual processing issue there is here, though I do follow the steps and arguments you are making to get there.

Throwing NB Accent at a pre-Color mono screen isn't terribly realistic, though I guess it could be interesting for some kind of specialty image. I don't really think NB Accent is meant to be used in that way, though I suppose it could be with some elbow grease.

I ran your files through a fairly typical processing workflow for RGB plus duoband NBA, and my (colored!) M31 had a few nice little Ha knots strewn about, as fully expected. Of course that can be altered greatly by the settings. But...the result kind of seemed like the point for galactic NB Accent.

The scattered background Ha Accent, regardless of whether in a donut or not, mostly hides away and gets lost in the color. In fact I'm not sure I could see a whole lot of effect, other than with stars (and that matter has been discussed in prior threads).

After finishing NBA, I used Layer and undo buffer to run a subtraction and see the difference. Indeed lots of scattered red, but mostly where expected (plus stars) and not really bothering the final full color image, IMO.

I would have done more testing, like also post-DN of the two, if the undo allowed more than just one step you could go back.

This particular combo of data may work better in NBA by extracting the duo file into just the Ha first, and composing it accordingly. One thing I noticed is that the duo file here is very strong in OIII, which we are not going to use, and that OIII strength kind of overwhelms the ability to properly set the first NBA screen. The selection of which duo band we are interested in, however, is not available until screen 2. So for now, pre-extraction of Ha by itself may allow better thresholding in screen 1.

Though, I haven't tried it yet and don't know how it would hold up still when gauging NBA against a mono background.

It kind of makes sense, perhaps? Would need to run it by Ivo, but I'm not sure NBA ever contemplated applying accents where every pixel is starting out with identical RGB values. Or maybe that doesn't matter?

[decay]

Hi Mike,

thank you very much for taking some time to have a go with my data sets.

First, some words regarding my mono / luminance only screenshots. I saw that Ivo did it in order to demonstrate and emphasise the impact of the NBAccent module, for example here viewtopic.php?p=13403#p13403 or in the documentation as well: https://www.startools.org/modules/narrowband-accents (M27 example). So I thought it would be a good idea to explain what I’m talking about. It’s not my intention to get NB accented mono images.

Regarding my processing issue / problem / question: Yes, I got a somewhat decent output with accented knots visible, but:
- there are areas (the donut ) where the colour balance is shifted towards purple
- I have the impression, that it may be possible to boost the Ha knots even more, but – yes - the pink donut gets simply too strong.

m31-6-nba.jpg (528.15 KiB) Viewed 4828 times

But of course, I’m not sure how it should look like. Therefore my question, if someone has an M31 example processed with NB Accent.

I would like to explain why I’m talking about this pink donut all the time: As far as I understand, this donut results as follows: The outer, dark areas are dark enough as result of the AutoDev stretching curve or clipped off by Threshold setting, so that here Ha does not make it into the resulting image. The inner, bright areas are masked by some kind off continuum extraction Ivo has implemented in NB Accent module (as described in his post, pls. see link above). This continuum extraction depends on – well, I guess – both stretching curves - of the visual data set and of the Ha data set. So the pink donut represents the area where accenting takes place or can take place, according to the settings. The position or size of this donut can be controlled by modifying the stretching curves – reasonably the curve of the Ha data set (‘AutoDev’ on screen 1 of NB Accent module).

(Does all this make any sense? )

Having all this said: I have the impression that this area, where accenting takes place (the pink donut), is too much towards the centre and it is not possible to get it more towards the outer regions by adjusting the ‘AutoDev’ of the Ha data set. And this is maybe the case due to the huge brightness of the core of M31 ?!

Thanks again, Dietmar.

[Mike in Rancho]

Hi Dietmar,

Well, what is it that is the issue here, a donut? Meaning you think there is too much accenting ending up in the, um, donut area, or you think there isn't enough accenting in the center...i.e. the donut hole?

The donut hole being the galaxy core, very bright, and of course there is a brightness correlation component here.

I extracted R (Ha) from your duo file and used that in the NBA slot, as Ha/SII from NB. It perhaps took some playing around especially in Screen 1, because that's going to be the basis of what transfers over, but I think I ended up with Ha sort of everywhere, including the donut hole, albeit less. Depends on settings. Also I did in fact do this after skipping color, so same mono underlying image (though I still wonder how useful that is), and I completed normal Contrast and HDR too.

I think there could be some limitations here. One, the L-eNh is decently wide, compared to say 3 or even 6-7 nm Ha bandpasses, no? And M31 is particularly bright-cored. And so as has been discussed elsewhere perhaps a fair amount of "Ha" ends up in the file when it's really just a portion of something bright and white.

BTW I noticed your first post claims that you are using that same compositing choice of Ha/SII from NB, but you do not say that you've extracted. I am not aware that the NBA slot will self-extract on its own in the way that R, G, and B will extract when noticing it's been given an OSC file. The docs and user notes seem slightly vague on this, neither explicitly saying it does so, or specifically denying it. So I would lean towards no, unless we hear otherwise. And if that's the case, that setting in Compose might be feeding the entirety of your L-eNh file (probably converted to an L of sorts), meaning all the bright stray Hb/OIII too, into NBA as if it is Ha.

Unless I am just getting lost.