Trouble colour balancing in SHO

[StuartT]

I've got a few hours of SHO data which I am having a hell of a job processing. I am a total noob to narrowband, so be gentle...

I integrated the Sii, H alpha and Oiii to get three masters. I then linear fitted them to the weakest (Oiii) and did a basic combine using pixel math (Sii -> R, H alpha -> G and Oiii -> B). But (unsurprisingly since the H alpha signal is super strong) I get a very green image. I've tried background neutralisation and color calibration but nothing seems to fix it. I think my data is pretty much ok, but I just can't figure out how to get a useable RGB image from it. A linear fit to H alpha is even worse!

Can anyone put me out of my misery and get a nice picture out of this?

Here are the three masters 

 

[scotty38]

Assume no further joy with the fancy pants dynamic pixelmath stuff then?

[vlaiv]

I downloaded the data, but no dice.

Don't do xisf as I don't have PI.

Here is something that you can try instead of your normal workflow:

Stretch each channel individually until you get pleasing image / you bring out the detail. Keep in mind that Ha is strongest so don't over do it with Ha.

Then do RGB combine for SHO palette.

 

[powerlord]

SCNR it and see where that gets you.

I don't know pixinsight, but when you say you linear fitted them, not sure what that means but as Vlaiv suggests - open all 3 masters next to each other, and aim to make the histograms peak around the same amount. usually this will mean stretching the Sii and Oiii up to where the Ha is. then combine them.

Also, you may want to try the script thing that does this someone mentioned - (google search later) - SHO-AIP.

 

[StuartT]

20 hours ago, scotty38 said:

Assume no further joy with the fancy pants dynamic pixelmath stuff then?

sadly no. I tried a number of fancy expressions.

20 hours ago, vlaiv said:

I downloaded the data, but no dice.

Don't do xisf as I don't have PI.

Here is something that you can try instead of your normal workflow:

Stretch each channel individually until you get pleasing image / you bring out the detail. Keep in mind that Ha is strongest so don't over do it with Ha.

Then do RGB combine for SHO palette.

 

Yes, I have tried to do that. I tried linear fitting them first. I also tried stretching them first to try and balance them out a bit more. But the Oiii signal is so faint that stretching it to get anywhere close to the Ha merely makes the sky bright. Here they are saved as FITS files @vlaiv
https://drive.google.com/file/d/15XUubOxT-LyIgbFM3MAFSsb5SmRfjR8z/view?usp=sharing

19 hours ago, powerlord said:

SCNR it and see where that gets you.

I don't know pixinsight, but when you say you linear fitted them, not sure what that means but as Vlaiv suggests - open all 3 masters next to each other, and aim to make the histograms peak around the same amount. usually this will mean stretching the Sii and Oiii up to where the Ha is. then combine them.

Also, you may want to try the script thing that does this someone mentioned - (google search later) - SHO-AIP.

 

Thanks. Linear fit is a Pixinsight process that tries to even up the images to each other (selecting one as a reference). SCNR on green makes a small difference but still nowhere near enough

[scotty38]

I can make varying different looking versions just by combining what you have but what are you looking for, what is it that you are not getting?

[vlaiv]

Here is what I've managed to do with the data and Gimp/ImageJ for processing:

OIII is very very faint.

We tend to spend same amount per filter - but that is often not good decision for NB imaging as Ha tends to be much much brighter than other wavelengths, and needs much less total exposure.

In this case, image would benefit from more OIII time.

[tomato]

I guess the crux is what constitutes a nice picture. I have tried to create more of a dual colour image by loading the channels into StarTools to their respective HST palette channels then after default processing I used the "Interpolate G, 100R,(50R+50B),100B in the Matrix option in the colour tool. Then into PI, separate the stars with StarXterminator, adjust the levels and channel saturation with Curves Transformation, reduce the stars with Morphological Transformation, recombine with PixellMath (~(~starless)*(~stars)), then a good dose of NoiseXterminator.

There are some vertical bands in the background which I have not attempted to remove, I have probably accentuated them with my pretty aggressive processing approach.  

What is the object? I would like to have a  go at imaging it.

 

 

[StuartT]

Thanks @vlaiv and @tomato - these are both nice images. I had another attempt today and managed slightly less greenness. But I am going to try and get a lot more Oiii data for this target (which is NGC6357, by the way)

 

[tomato]

There is a lot of interesting structure in this nebula, alas I think it is a bit too far South for me to image successfully.

[carastro]

I could tell you how to do it in photoshop.  Basically you need to adjust the green channel.  I have a tou tube tutorial for Hubble Palette on my website in my signature.  
 

Carole

[StuartT]

26 minutes ago, carastro said:

I could tell you how to do it in photoshop.  Basically you need to adjust the green channel.  I have a tou tube tutorial for Hubble Palette on my website in my signature.  
 

Carole

Thanks Carole. I did try dialling down the green channel in Pixinsight and also in photoshop. But to only limited avail.

The basic problem here is that my H alpha signal is drowning out the other two filters. I'm going to get a load more Oiii data (and maybe slightly less Sii)

[carastro]

Ha does gave a tendency to do that but fiddling around in selective colour will eventually sort it.   Just takes practice.   
 

Carole

[Elp]

Not quite SHO, I used HSO and a quick 15 minute process in PS with Siril PP prior, the data is exceptionally clean:

 

[StuartT]

3 minutes ago, Elp said:

Not quite SHO, I used HSO and a quick 15 minute process in PS with Siril PP prior, the data is exceptionally clean:

Thanks for this. I must say, I was hoping there was some genuine goodness hiding in there and it was only my processing ineptitude that isn't getting it out!

I didn't think of HSO. 

[Elp]

As everyone's said the O3 is very weak so will need more time, or you could try using just the H and S2 to create a fake O3, depends on the result you're looking to get.

[StuartT]

2 minutes ago, Elp said:

or you could try using just the H and S2 to create a fake O3, depends on the result you're looking to get.

you mean something like (Ha*0.6+Sii*0.4) to make a fake Oiii ?

[Elp]

Try it, maybe less Ha as it's quite strong and will overlap the existing Ha data which is where most of the signal is. There isn't really an exact rule to follow I find, as long as it looks good and is close to reference images.

[powerlord]

Well, couldn't resist having a go.

Here's my SHO with the data. Processed in Siril and Affinity Photo.

I've attached the affinity photo (AP) file too if anyone wants to have a look (workings in disabled groups at the end). affinity photo free to trial if you don't use it.

week oiii wise - this is usually just a sort of background shade in most stuff, so if you have week data, stretch and denoise it and guassian blur it - it then merges in better while still giving you the blue highlights where it's strong.

process:

process each channel is siril: histogram stretch, save as tifs.

use starnet2 cli to remove stars in each. save as tifs.

pull ha tif with and without stars into AP and subtract no stars from stars one. merge visible - name it stars layer.

now use levels and curves on ha no stars. to pull it up and move black point to just below clipping, noiseX it.

for sii, do the same thing

for oiii, do the same thing, but then use levels to pull down until you get rid of most of outer noise by clipping a bit. Then gaussian blur it.

save each layer (ha no stars, sii no stars, oiii no stars) as greyscale tifs.

back in siril: do an rgb composition - no luminance, with R=sii, G=Ha, B=oiii

do a green noise removal.

save RGB as tif.

open in affinity photo and copy layer into your composition with stars layer above it.

switch stars layer to add blend mode.

turn off stars for now, and use curves, levels, contrast, vibrance and a bit more noise X on nebula.

when happy, turn on stars. finally crop a bit.

jobs a good un 🙂

stu

lobster.afphoto

[ollypenrice]

I don't do false colour imaging but colour weighting is what it is in any scenario.  Firstly I'd follow Vlaiv's suggestions out of pure common sense. When you want to reduce the impact of the Ha you have two simple options, bringing in the black point and adjusting the stretch. You can do both in Levels. I'd begin by looking at the colour balance in the background sky. If you want a neutral dark grey you need parity between channels. I'd probably use black point adjustment to get there.

Next you want to balance the colours above the background sky level, so I'd use Curves.  You could pin and fix the background at a chosen level then lower the Ha curve above that. In a civilized program like Ps you can do that while looking at the three-channel image in real time as you adjust the stretch in one channel only.

But..  if you colour map as per Hubble and shoot an Ha-dominated image it's going to come out very green, surely? Why not try HOO instead?

Or...  do a nicely colour balanced image in which the Ha is greatly under-used and save that as a colour layer. Then use the Ha as luminance to restore the Ha interesting parts.

Olly

Edited August 5, 2022 by ollypenrice

[StuartT]

20 hours ago, ollypenrice said:

I don't do false colour imaging but colour weighting is what it is in any scenario.  Firstly I'd follow Vlaiv's suggestions out of pure common sense. When you want to reduce the impact of the Ha you have two simple options, bringing in the black point and adjusting the stretch. You can do both in Levels. I'd begin by looking at the colour balance in the background sky. If you want a neutral dark grey you need parity between channels. I'd probably use black point adjustment to get there.

Next you want to balance the colours above the background sky level, so I'd use Curves.  You could pin and fix the background at a chosen level then lower the Ha curve above that. In a civilized program like Ps you can do that while looking at the three-channel image in real time as you adjust the stretch in one channel only.

But..  if you colour map as per Hubble and shoot an Ha-dominated image it's going to come out very green, surely? Why not try HOO instead?

Or...  do a nicely colour balanced image in which the Ha is greatly under-used and save that as a colour layer. Then use the Ha as luminance to restore the Ha interesting parts.

Olly

thanks. This was very helpful. I carefully stretched each channel first so as to ensure the H alpha wasn't overpowering. Then when they were combined I got a much more pleasing result. As you suggested, I then added (a more fully stretched) H alpha channel as luminance to add the interesting bits back in. 

I have learned a lot in the process. Much obliged Olly

[ollypenrice]

4 hours ago, StuartT said:

thanks. This was very helpful. I carefully stretched each channel first so as to ensure the H alpha wasn't overpowering. Then when they were combined I got a much more pleasing result. As you suggested, I then added (a more fully stretched) H alpha channel as luminance to add the interesting bits back in. 

I have learned a lot in the process. Much obliged Olly

That's a thriller! The background is a big part of its success.

Olly