An excellent method of blending RGB stars with a narrowband image [Deep Sky] Processing techniques · Shannon Foye · ... · 60 · 7301 · 48

This looks like the answer to all my prayers regarding this massive hurdle in my processing abilities!  I'll give this a shot once I get a new set of data collected. Thank you for sharing this!

I’ll have to give this a try Shannon. I do pretty much the same process you have described except not doing this NB Stars Lum layer. Thanks for posting this.

Dale

James Lamb just recently did a video on YouTube describing a method of combining RGB stars with a narrowband image. The basic combination is very similar pixel math. The interesting twist he uses is to add a cosine curve. Here's the pixel math for the red channel:

Rn = pixel($T,x(),y(),0);
Gn = pixel($T,x(),y(),1);
Bn = pixel($T,x(),y(),2);
Ln = max(Rn, Gn, Bn, 0.001);
Rs = pixel(Stars,x(),y(),0);
Gs = pixel(Stars,x(),y(),1);
Bs = pixel(Stars,x(),y(),2);
Ls = max(Rs, Gs, Bs, 0.001);
L = ~((~Ls)*(~Ln));
L*(Rn/Ln + 0.5*(Rs/Ls-Rn/Ln)*(~cos(pi()*Ls)))

You repeat that for each channel, modifying the last line to be Gn/Ln... Gs/Ls-... for green and Bn/Ln...Bs/Ls... for blue. That last multiplier (~cos(pi()*Ls)) helps to better preserve the star colors.

If you haven't yet watched his videos, he's got a lot of great, really deep dives into topics. You can view the video that produces the pixel math above right here: https://www.youtube.com/watch?v=hX3LKhwqwvY

First unscreen the (RGB) stars between an original (stretched image) and its starless version by using ~(~original/~starless) .
Then, blend these (RGB) stars into NB data with combine(NBimage,unscreened_stars,op_screen())  . 
The unscreened stars will not have a background (and if there is one, you can make it zero by raising the black point a little. 
THere will be very few artifacts since the unscreened stars image has no background.

-adam

Adam Block:
blend these (RGB) stars into NB data with combine(NBimage,unscreened_stars,op_screen())

Adam,

With an image that has intricate narrowband structures, StarXTerminator will pick up some of the fine, bright structures as stars and remove them with the stars. I experienced this with my Crescent Nebula image. Without applying the luminance of the unscreened narrowband stars to the RGB image, these fine narrowband structures will be lost.

Also, some of the small narrowband stars that are removed may not be present in the RGB image. This may happen when the RGB image consists of just an hour or two of data and the narrowband data consists of dozens of hours. These small stars will be also be lost if the luminance of the unscreened narrowband stars is not applied to the RGB stars before screen blending it with the starless narrowband image.

Artifacts may appear where these small stars and small narrowband structures are not restored to the narrowband/RGB blended image.

To remedy this problem, I tried applying the luminance of the original narrowband image (prior to star removal) to the final narrowband/RGB screen blended image. This corrected the vast majority of the tiny artifacts. I still had one remaining large artifact in the halo of the a largest star in the middle of the Crescent Nebula.

That one remaining large artifact bothered me, and so I tried applying the luminance of the screened narrowband stars to the RGB image before screen blending it with the starless narrowband image. To my eye, the result was perfect. All small stars and intricate narrowband structures were returned to the narrowband image.

When I blinked the before and after narrowband images, the only difference that I could see down to the level of individual pixels was that the star colors had changed from HOO to RGB. The nebula details were perfectly identical between the two images. The luminance of the two images were identical because I had applied the luminance of the unscreened narrowband stars to the RGB image before screen blending the RBG image with the starless narrowband image.

This specific workflow only worked for me with the Crescent nebula because I did not alter the narrowband image after removing the stars. I had 50 hours of narrowband data. The Ha and Oiii combined beautifully with no need to manipulate the starless image before screen blending the RGB stars into in.

Adam Block:
Russ has a new version (new training) that might address some of the artifacts you were dealing with. I have been beta testing it. 
-Adam

I tried the new version. I saw a slight improvement in the one large star that gave me problems. The problem of small stars and structures removed in the narrowband image not being replaced by the RGB image was still present.

As a side note, I used the original RGB image instead of the screened stars to blend with the narrowband starless image. I stretched the image enough to bring out the stars. Most of the nebula that was brought out in the stretch was subsequently removed when the luminance of the narrowband screened stars was applied to the RGB image. The luminance levels of any remaining nebula pixels in the RGB image were much lower than the corresponding pixels in the starless narrowband  image, and therefore didn't create noise or artifacts in the blended image.

Hi all

Very interesting when using Pi but how do do that with Astroart or Startools etc...
Thanks and CS

Patrick

Interesting approach with the cosine function.
I just tried it but found it to be not as good as the bare screening.

Dominik Weinbrenner:
Interesting approach with the cosine function.
I just tried it but found it to be not as good as the bare screening.

Me too - that´s why I posted my concern and results.... "combine...." is better - at least up to now for me...

The image on the left is the narrowband before applying StarXTerminator  (with the box for unscreened stars checked). The middle and right images show the results of applying the luminance of the extracted, unscreened narrowband stars to the RGB unscreened stars and alternatively, to the original RGB image. I found no meaningful difference between the two images.

Please note that I did not alter the starless narrowband image before screen blending the RGB stars in to it.

Shannon Foye:
L = 1-(1-Ls)*(1-Ln);
L*(Rn/Ln + 0.5*(Rs/Ls-Rn/Ln)*(1-cos(pi()*Ls)))

Your second to last line is different.

They are the same. Using the expression ~x is identical to 1-x. Thus, the last two lines of the expression can be written as:

L = ~((~Ls)*(~Ln));
L*(Rn/Ln + 0.5*(Rs/Ls-Rn/Ln)*(~cos(pi()*Ls)))

I explained the same in a comment on James' video.

Shannon Foye:
Also, James Lamb has updated the video description with a shorted version of his expression. It is a single RGB/K expression:

Ln = max($T[0], $T[1], $T[2], 0.001);
Ls = max(Stars[0], Stars[1], Stars[2], 0.001);
L = 1-(1-Ls)*(1-Ln);
F = (1-cos(pi()*Ls))/2;
L*( (1-F)*$T/Ln + F*Stars/Ls )

I hadn't seen his update... that's quite simplified. Thanks!

Hello again,

I played around with various methods and algorithms, which are presented in that string of feedbacks....strange results, I must say, which I got:

First set of two screenshots: Left a crop of the stars as extracted with StarXterminator inside PI - in the area of the red/blue M27 nebula, all stars are heavily colored in the color of the nebula, but in the RGB image of the nebula, they are almost white - see right screenshot

If I extract the stars in PS with the same algorithm and subtract the nebula with stars from the starless image, I get stars which are less and dfferently colored and better usable  - left the stary only out of PI extraction and right the stars out of PS with subtraction method


Now it is getting even stranger for me - I combine the stars with the described algorithm in PI - all stars inside the nebula are colored with the color from the RGB image and therefore unusable
Whereas on the right side you see the stars coming from the PS subtraction - not good at all, but certainly a bit better and their color less disturbing



However, when I use the well known method of screen blending with PM, I get the best results... see yourself



I have no explanation for this and I cannot understand it either ..... for me, the combine(1,2,op_screen()) works best..

Maybe someone of you can enlighten me.. ?