LRGB or RGB?

[twade]

To all,

I've been noticing more and more astrophotographers imaging LRGB unbinned.  To my understanding, LRGB came to life to save time.  The L-channel would be taken unbinned and would consume most of the imaging time, but since it is "unfiltered" you gather lots of photons in a shorter period of time.  The RGB channels would be binned 2x2 to save imaging time due to their lower transmission of light.  To my knowledge, I can't see any advantage in doing LRGB unbinned.  It seems you would get the same result with just RGB unbinned.  In fact, you are probably sacrificing some resolution when using LRGB unbinned since most lenses can not focus the entire spectrum of visible light that well.  In summary, if you are going to shoot unbinned, you should just go with RGB.  Does anybody else have any thoughts on this subject?

Thanks,

Wade

[Juan Conejero]

Hi Wade,

In my opinion (and I prevent you that I am much more a laboratory worker than a real imager), the answer depends on the types of objects in the image, and on the optical system used.

If the image includes small structures with high chrominance contents, spatial resolution can be nearly as important for the chrominance as it is for the luminance. For example, consider a prime focus image of a galaxy, M101 for example. There are many HII regions that are both very small and very strong in the chrominance. If you don't provide enough spatial resolution for the RGB components, these structures will tend to be desaturated, at least partially.

However, as you say the optical system must be able to yield the required resolution on the range of acquired wavelengths. Indeed this may not happen with many lenses.

if you are going to shoot unbinned, you should just go with RGB

I agree with this. If the individual RGB images are taken unbinned, a synthetic luminance computed from the RGB components is in general preferable.

A synthetic luminance has the important advantage that we can specify the contribution of each RGB component to the luminance, in terms of information contents. This can be done in PixInsight by defining a custom RGB working space (with the RGBWorkingSpace tool).

To generate a synthetic luminance, one can extract the L* component of CIE L*a*b*, or the Y component of CIE XYZ. The CIE XYZ space allows us to work separately on the luminance and the chrominance without altering the linearity of the data. This obviously works equally for one-shot color CCDs and DSLR cameras.

Now let's see what the expert imagers here have to say. I'm ready to stand corrected.

[Jack Harvey]

I am by no means a expert but here is my 2 cents worth.  I often shoot RGB d extract the Lum using the extract channels and the system you describe.  I either then manipulate the Lum (Histo, decon, wavelets etc) or substitute a master Ha frame in place of it.  http://pixinsight.com/forum/viewtopic.php?t=562

[Drew Sullivan]

Hello to all.

I am new to this forum but have had PixInsight for a while. I was directed here by my friend Rogelio Andreo whom you may know from his posts here.

Re binned or nonbinned RGB:

The _only_  advantage of binning 2x2 is that it improves that part of the Signal to Noise caused by camera read noise. (It does that by averaging the read noise of 4 pixels at a time.) Binning doesn’t “get you more signal” or anything like that.

If your subexposures binned 1x1 are long enough to reach sky limit i.e. if the adu in empty sky is maybe 1000-1500 phtons per pixel in a subexposure binned 1x1 then your sky noise is swamping any camera read noise and there is _no benefit_ to binning.

Reaching sky limit may not be possible at long Focal Length or in very dark sites so there binning. will help. Otherwise there is no benefit.

The disadvantage to binning is that you can’t then add the R + G + B to get “free” luminance data. If you add your R+G+B (with most RGB filters) you get the same photons you would have gotten with an L filter e.g. 1 hour each of R, G and B summed is the same as 1 hour of L. So you have “four hours of data collected (an hour each of L R G and B) in only three hours”. If you bin 2x2 you can’t do that (you would get L data binned 2x2).

You can take just RGB or LRGB. The LRGB gives you the Signal to Noise you would get with 2 hours Lum (add the L+R+G+B together or average them. Either will have the same Signal to Noise assuming the aveage does not have numbers so small as to cause rounding errors) and an hour each of R G and B. I.e. LRGB gives 5 hours of data in four hours.

Suppose you have 12 hours to image. You can take 4 hours each of RGB. That gives you (after making a synthetic Lum) 4 hours L and 4 hours each R G and B. Or you can take 3 hours each of LRGB. That will give you 6 hours of L and 3 hours each of RGB. The LRGB gives you a little more detail at the expense of _very_ slightly less color definition in dim areas (the Signal to Noise changes as the square root of the time so you gain more L than you lose in RGB). Without that better detail you probably can't show the dim areas anyway so I tend to use LRGB.

Regards,

Drew Sullivan
California

[mmirot]

I understand your point.

Somehow,  I don't think you will get double the Lum exposure out it.

Also, Your L filter and RGB_Lum extract are not quite the same as truely averaging L frames.
Also, If the S/N is poor in one of the RGB exposures it might actually degrade the Lum channel result.
In this case you might get a better result with just exposing each RGB exposures 33% longer.

I am starting to think that taking just RGB filter is also simpler.
 

I have not doen any S/N measurements so this is just  my thoughts at this point.
However, I know some top imagers that now do this routinely.
I think 2x bins will be history in a few years. It rarely helps if at all if you are sky limited as you point out.

Max Mirot

PS I never bin

[Warhen]

Adding a wrinkle new to me, CCDWare's John Smith stated at AIC 2012 that with some sensors' architecture, (16803 included), binning actually adds read noise! So, for me too- 1 x 1 on all images.

[jeffweiss9]

I heard John Smith's remark at AIC, but interpreted that to mean on some cameras, but definitely not mine (ST2000).  One thing that bothers me about taking R,G,B at 1x1 binning is that, say your read noise is at 5% or 10% of the sky noise so it is not contributing significantly at 2x2 binning and your 'normal' RGB exposure time, then you would have to increase that subexposure time by a factor of 2 ideally* to replicate the same condition.  Although that 2x increased exposure for RGB at 1x1 would just get me back to what I use for Luminosity and so is not a problem, that adds to the total collection time.  For L:RGB ratio of 1.0, 2 hours of total exposure would increase 50% to 3 hours.  But you would get better resolution for the color and the added synthetic luminosity for that price (steep for a field observer like myself).  So it seems that is the tradeoff.
    The other thing is L:RGB ratio, however. With 2x2 binning of the RGB, I agree with numbers elsewhere of about 1.25 or so ratio of the L to the RGB total exposure giving an 'optimum' result.  Taking LRGB all at 1x1 would give you 2.00 and tend to wash out the color, would it not?  It seems RGB at 1x1 and just using the synthetic luminosity would be better.   Or would you still get all the benefit of the increased luminosity?
..Just some thoughts (brewing for some time).
-Jeff

*Matt Thomas/SBIG points out in a parallel discussion at
http://www.cloudynights.com/ubbthreads/showflat.php/Cat/0/Number/5495336/Main/5494372
that read noise always increases with binning but for most cameras significantly less than the sqrt(4) for 2x2 software binning (assumed equivalent to no binning).  So read noise for those cameras is reduced by binning vice no binning (and read noise increases by only 25% for 2x2 binning with my ST2000).  As Drew pointed out, if your read out noise is negligible relative to the sky noise, this makes no difference and you get the resolution and luminance contribution from the RGB without binning.  But if you have very dark skies, are using NB or other where read noise is still a factor, then binning should should help with those cameras.

[RickS]

It might be a simplistic argument but it seems to me that collecting Luminance data is inherently more efficient than collecting filtered colour data.  You aren't filtering out any photons so there is more data and proportionally less shot noise.  I've managed to pull together LRGB images that looked OK with very few colour subs.  Of course, as Juan suggested, this may not work as well for targets with lots of small scale chrominance variation...

Cheers,
Rick.