Re: [APML] Synthetic Luminance Channel for Film Images

[Jerry Lodriguss <jml@astropix.com>]

>there was another question about this, answered by Stan Moore on the SBIG
>list, maybe it helps:


Hi Volker,

You left out the complete post by Paul Kane that Stan was replying to. I'm 
going to try to put these posts in sequence with comments. Mike Cook was 
kind enough to pass them on to me.

Right now, I am back to leaning towards thinking that a synthetic RGB 
doesn't really gain anything. Common sense and Occam's razor would indicate 
that you can't get something for nothing, and I think Paul explains it best.

I measured the s/n in a blank area of sky, but that seems to only be 
measuring the apparent graininess of the image. There isn't any real signal 
there that we would care about.

So if we can lessen the noise or graininess that's good, but if the process 
lowers the contrast of the real image data in the process, and we have to 
increase the contrast to get it back, then the s/n in the blank sky will go 
back up.

I may go back and include a gray scale step wedge in the tests, but 
non-linear curves are kind of a pain to duplicate by hand, so it may be 
tough to match them exactly.





 From Paul K:

>I really doubt that using LRGB with a synthetic L channel can
>increase the real S/N. I believe that the results demonstrated on
>Jerry's page are simply due to the effect of linear and non-linear
>transformations used to construct the L channel.
>
>The Grayscale method mixes R,G, and B by applying a predefined linear
>scale. I'm guessing that so does the Desaturate method, although the
>linear scale is different.
>
>The Lab L channel is a non-linear combination of R,G, and B that is
>meant to be more perceptually accurate.
>
>As the result of applying these different L channels in an LRGB
>composite, different emphasis is given to lighter or darker areas and
>midtones. Such an change in contrast can create an artificially
>higher local S/N, but it certainly does not generate "more" signal.
>What's more, the same effect can be achieved by applying the right
>non-linear (curves) or linear (levels) transforms to the original RGB
>image, bypassing the LRGB step altogether, IMHO.


Me:

I think this is really what's going on. I ran into something very similar 
when I explored different Photoshop methods of compositing multiple color 
originals.  Photoshop has a lot of different compositing modes, like 
screen, and softlight, that apply even more non-linear curves in the 
compositing process, and they can be duplicated in curves.

The true gain comes from compositing two different originals. With a 
synthetic luminance image we don't have separate original data.



 From Stan:

>But it does seem possible to transfer some of the chromatic S/N to
>the luminance, the penalty being lower chromatic spacial resolution
>(a familiar characteristic of LRGB <g>).


Me:

I thought the lower chromatic spatial resolution was because the color was 
shot originally at a lower resolution because it was binned CCD data.

However, the idea of transferring some of the S/N from the chrominance to 
the luminance seems possibly plausible, but it would seem to have to 
balance back out when re-combined with the color data. I will have to think 
about this some more.




 From Mike Cook:

>I think Stan is trying to rationalize how it might be possible to obtain
>higher S/N since you can't get something for nothing.  It is analogous to
>binning the color channels but instead making the trade by a processing
>technique.


 From Stan

>For a weakly colored object (like a galaxy), combining the RGB
>exposures into a greyscale produces S/N approx 1.7x higher than each
>filtered exp.


Me:

Yes, but a galaxy is a broadband source, so there is signal in all three 
channels. I wonder if the s/n increase of a grayscale composite of the 
individual RGB channels would yield as high a s/n increase for an emission 
object.

Also, it seems obvious that three times the exposure in a composite of the 
three individual channels would have better s/n than a single exposure of 
one of the channels.


That's it so far.


Jerry

Astronomical photography: http://www.astropix.com



--  APML Archives at <http://astro.umsystem.edu/apml/>  ---
             Unsubscribe at <majordomo@seds.org>