Quote:
Originally Posted by topheart
That's just stunning Lee!
I have a strong personal preference for RGB stars in NB images.
Tick!!
Well done!
Cheers,
Tim
|
Thanks very much Tim

I like the nice small NB stars so I always resisted doing RGB stars, but now I've figured out how to get the best of both worlds so now I'm more likely to use them.
Quote:
Originally Posted by vlazg
Absolutely awesome Lee, detail is great and the palette is a lot subtler than some we see but still very striking and much more pleasing to my eye.
Well done 
|
Thanks very much George :-) Yeah, this palette is very close to my first attempt at this image, which I think was either the first, or one of the first SHO images I ever did, so I guess that makes it a bit boring compared to some, but I like it.
Quote:
Originally Posted by gregbradley
Stunning detail. So sharp, amazing.
I see you used 30 seconds for the RGB for the stars and you got nice colour. So perhaps this is a good strategy on regular broadband imaging to get good star colours.
Greg.
|
Thanks very much Greg :-)
I might have gone longer on the RGB stars but for two things: I couldn't be bothered setting up per-event gain, so I shot these at the relatively high gain I use for NB and secondly, my dark library goes from 30s to 180s. Even at 30s with gain 111 I blew out quite a lot of stars.
I get lots of colour in my stars (often too much) by using Arcsinh in PixInsight to do the majority of the stretch.
Quote:
Originally Posted by Placidus
Lee, the RGB stars are astonishingly good.
How, algorithmically, did you get rid of the NB stars? Did you produce a "starless" version and then add the RGB stars?
I'm yet to find an algorithm (such as "use bipolynomial regression on the Winzorised background to interpolate what's under the stars, starting at 0.73 FWHM from centre of star, with a 1.5 FWHM feather", as opposed to "use Bloggs's Photoshop Action" , or "Use Smythe's Pixinsight Process", that does it at all well.
Will understand if it is a secret.
Best,
Mike
|
haha, thanks Mike :-)
No real magic here on my part, I'm standing on the shoulders of giants. I used Starnet++ to remove the stars... I used the PixInsight plugin version, but you can also get a CLI version of it if you're inclined..
Starnet++ uses AI to remove the stars. Unfortunately the model was trained using a refractor (a Tak I believe?) so my stars don't fit the model all that well. It does a pretty amazing job even still, but leaves quite a lot of artifacts that I have to clean up... I did that using clone stamp.
From here I applied my one "trick" which I thought of recently. Pretty obvious in retrospect, I'm sure others have been doing this for ages and I'm just late to the party. Sounds more involved than it is, but...
* I extracted the starless H from the star-filled H; this leaves me with an image containing just the H stars
* I stretched the RGB colour image containing the stars to roughly approximate the size of the H stars (very roughly, not really important)
* I separated the RGB stars into channels a and b from the lab colour space
* I separated the fully processed starless NB image into L, a and b channels
* Using the H stars as a mask, I blend the a and b channels from the stars into the a and b channels from the NB image
* I then add the H stars to the L channel from the NB image
* I now recombine the Lab channels and have a NB image with stars the size of the H ones, but the colour of the RGB ones
Hope that makes sense (and helps!)
Quote:
Originally Posted by Geoff45
That's one of the best pictures I've seen for quite a while. It has an ethereal dream-like quality. Nice varied colours and nicely muted.
|
Wow Geoff, thanks for that, I really appreciate the kind words