I need help with SHO

[Datalord]

I have tried a few times to get a hubble palette image in PixInsight and always end up using something else because I just can't make it look good. Forget about the noise and sharpness and all that, I can deal with that, but getting the colours to that vibrant blue an gold is beyond me.

This is what I have, right after I have masked stretched each of the NB masters and combined with pixelmath:

After a series of ColourMask and CurvesTransformation, I get to this:

Can someone give the first one a shot? Do I have to alter the straight SHO formula?

[wornish]

Not sure if this helps but using your first image I followed the tutorial here using PS.

http://bf-astro.com/hubblep.htm

This is the result.

 

[Datalord]

Yeah, yours is much better. Getting the blue center from yours should be easy. I'll take a look. 

Thanks! 

 

[Datalord]

It got me a lot closer. I think I'll have to change my tactic on the noise side, but the colours are at least much better. Thank you!

[wimvb]

If you post or pm the 3 unstretched masters, I can have a go at it. Mind you, I have very little experience with nb data, but would love to learn and will of course share my experience.

[Datalord]

Hey Wim. I've placed the unstretched, uncropped masters in this folder: https://www.dropbox.com/sh/if37gb0vud8l1ep/AAAze5c7ATSAu8eJdvdJc3Ifa?dl=0

They have been deconvolved and aligned, but not stretched. I'm pretty certain I know what to do, though: de-noise the masters with a very heavy hand to create the colour master, then use my Ha as luminance to add contrast and detail, just as I would an LRGB. But if you are curious, take a look at it. The data is pretty decent.

[wornish]

I had a go with the masters you shared and the results are much better.   Still used the same process steps in PS referred to in the link.

Your data is very good

 

 

[Datalord]

Brilliant result. Thanks for your help! 

[wimvb]

Here are my attempts at NB processing. Used PixInsight

Basic process:

1. linear fit (histogram alignment) with O3 as reference
2. channel combination
3. cropped edges
4. DBE to remove gradients
5. background neutralization (using the dark nebula in the center as a reference)
6. reduced the green a bit (scnr at 50%)
7. removal of purple halos (invert, scnr, invert)
8. arcsinh stretch
9. contrast enhancement with curves stretch
10. colour saturation
11. star reduction
12. removed more of the green with scnr at 50 %
13. resample

For the second image, I did the last scnr at 100%

[wornish]

You have created a lot more feeling of depth to the image.  

Big thanks for sharing the steps, never stop learning.

 

 

 

 

 

[Datalord]

4 hours ago, wimvb said:

DBE to remove gradients

Hmm, I decided to omit that step after trying, simply because I found so few points without nebulosity that the gradients were skewed. Did you place the points on the nebula? Should I not worry about that?

4 hours ago, wimvb said:

arcsinh stretch

I never really use this one. What's the improvement of this over MaskedStretch and HistogramTransformation?

4 hours ago, wimvb said:

For the second image, I did the last scnr at 100%

Big thanks for the efforts and steps. Really appreciate it.

[wimvb]

1 minute ago, Datalord said:

Hmm, I decided to omit that step after trying, simply because I found so few points without nebulosity that the gradients were skewed. Did you place the points on the nebula? Should I not worry about that?

I don't think there's any area with background. It's all nebulosity. If you use DBE, you try to put just a few markers on any poisition where you want to have the same value after correction. Then you use "normalize" in the correction tab of DBE. This will keep the "average" colour and won't do a neutralisation. Many people don't use this feature and let DBE neutralise the background. This is improper use, imo. It makes more sense to me to let the correct process (background neutralization) do that.

6 minutes ago, Datalord said:

never really use this one. What's the improvement of this over MaskedStretch and HistogramTransformation?

Arcsinh stretch keeps the colours intact, even at high values. Histogram transform will brighten all high values to white, while masked stretch works the same, but protects highlights in each iteration. The masking suppresses or delays this whitening effect, but only arcsinh stretch actually keeps colours as they were in the linear image. 

[Datalord]

Interesting. Thanks again!

[vlaiv]

On 19/01/2020 at 19:15, Datalord said:

I've placed the unstretched, uncropped masters in this folder: https://www.dropbox.com/sh/if37gb0vud8l1ep/AAAze5c7ATSAu8eJdvdJc3Ifa?dl=0

Why do you use 16 bit format? You are loosing a lot of precision that way. Let me show you something.

Look at histogram above and my black/white point settings - they are set to 40 and 270 - there is less than 256 levels in there and almost full dynamic of the image in this range. This means that you in reality have 8 bit of data for your nebulosity because you are using fixed point format - 16 bit.

With fixed point format - all pixel values must fall in that range and must be aligned to precision supported by that format. In floating point format - every pixel is free to have it's own range (hence floating point) and you won't run out of number of bits any time soon (precision is something like 10^15 or so).

Could you do another upload - but this time just save image in 32bit format (fits preferably) and don't even do deconvolution - just bare registration (alignment) and stacking.

[Datalord]

2 minutes ago, vlaiv said:

Why do you use 16 bit format?

Inside PixInsight it is always 32bit xisf. It's only when I export to Photoshop it has to be 16bit as PS can't deal with 32bit tiff files.

I uploaded them as fits in 32bit if you want to have a poke at them.

[wimvb]

1 minute ago, Datalord said:

Inside PixInsight it is always 32bit xisf

32 bit floating point, afaIk. 

[Datalord]

1 minute ago, wimvb said:

32 bit floating point, afaIk. 

Right, stored in the project in the xisf format?

[vlaiv]

6 minutes ago, Datalord said:

I uploaded them as fits in 32bit if you want to have a poke at them.

Sure I do

I'm not saying I'll provide any meaningful results, but fiddling around with data is much fun

[wimvb]

I believe so. 

[vlaiv]

10 minutes ago, Datalord said:

I uploaded them as fits in 32bit if you want to have a poke at them.

I'm missing something here?

You uploaded something different? It's been cropped differently and possibly already stretched?

Look at that star profile:

Not proper thing, is it?

[vlaiv]

Well, there is only so much a man can do with limited 16bit data

There is a severe issue with OIII and SII subs at the edges for some reason (left/right edge) which I could not get rid of - hence bluish / purplish left and right side of the image.

[Datalord]

1 hour ago, vlaiv said:

You uploaded something different? It's been cropped differently and possibly already stretched?

Ah, sorry, that was the wrong set. 

[Datalord]

Finally got around to finishing this up. Thanks for the help!