Comparison between LRGB and RGB images at the same integration time

Noah Tingey:
Howdy y'all -- I wanted to demonstrate how much of a difference luminance can make to a broadband image. Most comparisons that I've seen will use X hours of RGB compared against (X + Y) hours of LRGB (where Y is the number of additional hours in L). I don't think that makes for a very fair comparison. This comparison uses the same total integration time for each image: 5.5 hours.

In the RGB image, this is 5.5 hours split roughly evenly between R, G, and B.
In the LRGB image, 2.5 hours is L and 3 hours is split roughly evenly between R, G, and B.

Both images were stacked and processed using the same parameters and techniques. The light frames used in the LRGB integration were selected at random -- I didn't pick the best frames out of the bunch. Stars have been removed to allow for easier comparison.

Data acquired during a new moon in bortle 4. Details about the gear are on my astrobin post: https://app.astrobin.com/u/ThisIsntRealWakeUp?i=3i5tls

LRGB (click to open the full size):


RGB (click to open the full size):


If you ask me, the LRGB image is clearly superior to the RGB image. This demonstrates the benefits of a mono camera vs OSC, even for broadband imaging. (Though do note that an OSC camera also has the downside of bayer interpolation, which reduces sharpness. This is not represented here because both images were taken with a monochromatic camera).

i agree the lrgb looks more detailed. by quite a bit.

Nice comparison. Regardless of the endless debates, this easily becomes apparent to those who have done both.  In the same way, L/mono has  just a big of difference in sharpness&contrast

You can also combine L + RGB as a synthetic L to enhance snr even more. It doesn't work well on every target but for dust or galaxies it helps.
Here's an exemple on how to do it with Pixinsight: https://www.youtube.com/watch?v=Q2PLUI2hBvQ

Thanks for the comparison! Really good to see as a confirmation!

Is there also a consensus on how much RGB is optimal, before only adding more Lum? E.g.: I have so far 8h in Lum and 3h in each channel of RGB. How should I go forward if I want to capture 12 more hours? Is it more optimal to just shoot 12h Lum, or 6h Lum / 6h RGB?

My guess would be: It depends on the target / goal as gathering more integrarion time will mostly benefit revealing more, faint details. If you go for colorless details (IFN, ..), only Lum is sufficient. If you want e.g. more background stars, more RGB is better. If you want more faint details that also have color (also IFN as an example), you will need more Lum and more RGB.

Is that assumption - more or less - correct?

The go-to ratio that I see is 4:1:1:1 L:R:G:B. But I've never seen someone experiment with this nor do I know where this rule of thumb comes from.

In the future, I plan to experiment with ~25 each of R, G, and B (or however many is needed to ensure I have no walking noise or dry pixels after a drizzle integration) and then the rest will be only L. This is because you can aggressively denoise your RGB data before applying your luminance with very, very little noticeable degradation from denoising your RGB data. To demonstrate this, here is a comparison where I applied L to a blurred RGB image versus applying the same L to the unblurred RGB image:

Blurred RGB, to give you a sense of how much I Gaussian blurred my RGB before applying L:


L + Blurred RGB:


L + RGB that was never blurred:


Really the only difference that I can easily spot is that there's no chrominance noise in the L + blurred RGB image and that smaller stars have lost their color. So I think that you can safely denoise your RGB stack pretty aggressively before applying L. Which means you don't need to spend a lot of time gathering RGB to get a high SNR in the first place. But I'll play around with this some more and see how it turns out in my future images.

Though like you said, I bet there are some targets where it's important to not skimp on your RGB integration. Like if you want fine color detail in small galaxies, perhaps.

Editing to add: To be clear, I don't think you should denoise your images by using a Gaussian blur. This is just to demonstrate how sloppily you can denoise an image while still being able to pull back a lot of the detail with luminance.

Noah Tingey:
My apologies if I've misunderstood what you're saying, but just to be clear: both the images in my comparison were taken on a mono camera -- neither one had a CFA. It's just that the RGB-only image did not use L whereas the LRGB image did.

No I'm the one that has it wrong.  Thanks.. Didn't read your original post properly

Noah Tingey:
...

Though like you said, I bet there are some targets where it's important to not skimp on your RGB integration. Like if you want fine color detail in small galaxies, perhaps.

Editing to add: To be clear, I don't think you should denoise your images by using a Gaussian blur. This is just to demonstrate how sloppily you can denoise an image while still being able to pull back a lot of the detail with luminance.

Like you said, for most galaxies, I tend not to denoise RGB as much since you obliterate fine color details in the galaxy.     I tend to denoise and lower the saturation of the sky background areas more in most of my images, and minimally denoise the brighter, higher S/N areas to preserve fine color details there.

Also, the latest Noise XTerminator allows selective color noise reduction, which is great for RGB components of an image.

Ani Shastry:
@Noah Tingey, did you try creating a synthetic luminance (using the ImageIntegration tool in PixInsight) from the RGB-only dataset and doing an LRGB combination with that synthetic luminance similar to how you combined "real" luminance with RGB?

To cut down on chrominance noise without losing much luminance detail in the RGB image, I extracted the luminance of the RGB image after channel-combining and doing an initial stretch. I then processed it the same way I processed the real Luminance for the LRGB image.

I did it this way for two reasons:
1. I wanted to give the RGB-only image its best chance, editing it as I would if I were an OSC camera user or someone without real luminance data to work with.
2. Because I knew that I would be doing some processes specific to the luminance of the LRGB image, and I wanted to keep things as apples-to-apples as possible. So I didn't want to do some edits to the real luminance without doing the same edits to the extracted luminance.

So not quite the same as making a synthetic luminance with the ImageIntegration tool. But I did extract the luminance from the RGB image and process it as if it were real luminance data.

hi

sometime i recommend a super luminance , better SNR 
just using image integration tool ,add LRGB masters only( never narrowband data) and select SNR from weights window

the result will be a LUM with better SNR ,mean more details and less noise 

 and maybe in some case you will need  to control your lum with Linear fit
CS
Brian

Noah Tingey:
To cut down on chrominance noise without losing much luminance detail in the RGB image, I extracted the luminance of the RGB image after channel-combining and doing an initial stretch. I then processed it the same way I processed the real Luminance for the LRGB image.

Instead of extracting luminance from RGB, what does the comparison look like if, as suggested above, you create a synthetic luminance by integrating RGB channels and then create a new LRGB with the synthetic luminance and the original RGB. The synthetic luminance should have higher SN than the luminance extracted from the RGB channels, and in my opinion would be a better comparison of what can be done with only RGB data as opposed to LRGB data.

andrea tasselli:
It seems like demonstrating the obvious.

Certainly in some sense it is, but there are many debates in places like CloudyNights and possibly elsewhere, people love to be contrarian.

Brian Puhl:

andrea tasselli:
It seems like demonstrating the obvious.

You'd be surprised how many people are preaching that luminance is pointless, and the subsequent cult following they're developing.

The simplest way I can think of this question is "do I want to throw away 2/3s of my photons on all of my subs or some of my subs?"