Why do we bother with R, G and also B ?

[bitli]

I had a problem with the Blue channel on an exposure and was not able to complete it for a long time due to weather. So I created a synthetic blue as follow: resize the L to the binned G and R, linear fit of the resized L, G and R to a common base (I think I used R), PixelMath for B = 3*L-R-G.  Then usual L RGB with original L naturally.

The result was surprisingly usable with the help of PI BN and color calibration.

Thinking of it, given a fixed telescope time, would it not make sense to spend the color time on just two colors (may be R + B) to lower their noise (50 % more frames) and have better rejection and synthesize the third color as above? Did somebody try this in some systematic way ?
-- bitli

[Dimitris Platis]

A brief test shows serious problems in saturated areas..such as stars...
It could work but only in some cases.....I changed the pixelmath equation to 3*Lum-0.5*Green-1.5*Red to be able to get accurate colors....even after BN/CC

[rockyraccoon]

Good question!
I have not tried this systematically, but on a previous occasion, I ran out of imaging time and ended up synthetizing the G channel out of the R and B.
The result was good enough for me , and I ended up wondering the exact same thing.
Image in question: https://www.flickr.com/photos/dmal/15167688064/in/set-72157649207688112

I'm wondering if anyone has proven one way or the other whether it's advantageous to just shoot two channels rather than three.

[bitli]

Good shot rockyraccoon,
Dimitris, indeed I had almost no saturated star in my test - I guess that the images must be reasonably homogenous for this system to work.
-- bitli

[mschuster]

Hi bitli,

One reason is noise. When you synthesize you pick up the quadrature sum of noise from the inputs. So the synthetic always has lower SNR than the inputs. Spending more time on the inputs will not change this fact. The loss will be noticeable in faint structures.

Mike

[Niall Saunders]

I approached the problem this way . . . . .

IF _location = ("NewMexico" OR "Namibia" OR "AltoPlana_Chile"
'you've got it easy - it hasn't rained here for 13.6 billion years

   THEN use(mono_imaging, R, G, B, L, Ha)

ELSE
'accept that you will never complete a monochrome imaging run using filters

   SO put(mono_imaging_kit, on_eBay)
         buy(decent_OSC_imager(QHY10))
         collect(easily, exposures per night(30, subs(300s)))

   msgbox("Be Happy")

END IF

I've tried this in PJSR, and it works.

All I have to do now is re-learn how to use PixInsight all over again - it's incredible how much you can forget in just a couple of years (Harry, your dulcet tones keep me amused throughout the cold dark hours of poking captured photons onto multi-TB hard drives).

Cheers,
Niall
Clinterty Observatories
Aberdeen, UK

[rockyraccoon]

ok - so let's say the noise on R, G, and B is the same and has a value of  "1".
Synth G will have a noise of sqrt(2) = 1.41
So SNR for green only will be 41% higher than it could be, but with the time save I have 33% more signal on two channels.
If I'm using a Lum filter to tease out details on fainter structures, does it really matter that the green channel is marginally noisier?

Hi bitli,

One reason is noise. When you synthesize you pick up the quadrature sum of noise from the inputs. So the synthetic always has lower SNR than the inputs. Spending more time on the inputs will not change this fact. The loss will be noticeable in faint structures.

Mike

[bitli]

It would be interesting to have a detailed math calculation about the noise. It may not be obvious (at least for me), as the luminance may be significantly less noisy when we do the software binning (in theory this should double the signal to noise ration, if I make no mistake).
 
But just noise figures is not the only consideration.
For example we can get better rejection during integration if we have say 7 + 8 images than 5 + 5 + 5. For low image count, this may make a difference in some situation. 
Not everybody as the chance of imaging long nights in sunny Scotland :-)

I am not promoting giving up one channel (the title was a little bit provocative), but it seems it is a viable alternative in case of problems and I would like to understand the rationals of the 3 channels better. I looked at TV using L + 2 colors for many years without problem.

BTW I do remote imaging and an image on a large telescope cost me about the price of a PI license, so the # of images matters very much.  But if somebody wants to sponsor me ...
-- bitli

[jkmorse]

Bitli,

First off thanks for starting this great post.  I stumbled across it when trying to figure out how to save a weekend's worth of imaging when I used my older STF8300 CCD to shoot RGBs (I now only use synthetic lums) and forgot that the Lum filter was in the first slot on this one rather than in the fourth slot like my main CCD.  As a result, I ended up with a series of LRGs instead of RGBs.  I am now in the process of trying out your fix.

Note, in response to the issue with zero values showing up in saturated stars when you run the fix, Carlos had a nice solution on a different problem that should work here.  He stated as a workaround, you may try PixelMath to change all the zeros to ones (with something like "iif($T==0.0,1.0,$T)" ).  Sounds elegant and I will be trying it this weekend since I am having this problem with a couple of my synthetic blue images. 

In any event, just wanted to say thanks for a really helpful post.

Best,

Jim