What's the importance of a Luminance layer?

[pipnina]

In my mind, RGB *should* be enough. But then I don't know exactly what a luminance layer does for an image or how it / the RGB laters are used in a processed image.

Hoping for someone to enlighten me

    ~pip

[Owmuchonomy]

The simple way I got my head around this was to take a normal image, I used my M42, put it into Photoshop and choose LAB mode. Then if you look at each channel in this mode it is possible to see how much information is contained in the luminance channel compared with the others, most of it to be honest. An expert imager may come along soon and explain it.

[brantuk]

I just had a read of this link and I think it should answer your query:

http://www.robgendlerastropics.com/LRGB.html

Hth

[Stub Mandrel]

Here are a couple of charts that may help. Try opening them and splitting them into LAB RGB or CMYK and see how different the splits are!

Desaturating them before you do the split may also help understand the role of the different channels.

RGB:

CMYK

It's also interesting how (on my monitor at least) the CMYK (not the RGB) version seems to have the colours more as I would expect to see them.

There's also a LAB - where L is lightness, A is green (-) to red (+) and B is yellow to blue, but tyeh system won't let me upload a TIFF version (you can't save a LAB jpg)

(I'm only pretending to understand this, you know...)

Edited August 6, 2015 by Stub Mandrel

[cfpendock]

I like the reference by Brantuk of the article by Robert Gendler.  The explanations for LRGB and HaRGB are clear but I'm not sure about the RRGB.  Does it mean that one should obtain high resolution images using the red filter, and then use this instead of a clear IR filter for the luminance channel?

Chris

[Uranium235]

I like the reference by Brantuk of the article by Robert Gendler.  The explanations for LRGB and HaRGB are clear but I'm not sure about the RRGB.  Does it mean that one should obtain high resolution images using the red filter, and then use this instead of a clear IR filter for the luminance channel?

Chris

Nope, becuase if you use an R filter for lum then you are losing 2/3rds of your data. The point of a luminance channel is to capture all of the available (visible) wavelengths at the same time and enable you to concentrate your efforts of noise reduction (via stacking), sharpness and detail into one layer. Basically, the L layer is the most important part of your image - so best not to sell it short.

Lum = Carries your detail, contrast and cleanliness

RGB = Carries the colour, thats all (thats why you can afford to mash it to death with noise reduction in Ps)

[Uranium235]

The above can also apply to DSLR imaging using the LLRGB technique - where you split the workflow into two streams:

1) A monoised version of you image, that enables you to get a better grip on the noise and concentrate on getting plenty of contrast/sharpness.

2) The RGB layer, where you toss aside the fine details in favour of getting the best colour scheme possible.

Then merge the two for your final product.

[cfpendock]

Thanks Rob.  Exactly what I always thought. In fact I have often regarded the RGB part of the image as merely providing a colour "wash", rather like the tinting of black and white photographs before the days of colour.  

But it doesn't explain the RRGB method described by Robert Gendler, particularly his comparative images of the horsehead in LRGB and RRGB.  Except that his RRGB image took nearly twice as much time, presumably because the extra time was just using the red filter? And no clear filter used for this?

chris

[pipnina]

It seems kinda weird to me that LRGB is just colour pasted on top of a detailed greyscale image... But I guess that's how it is.

Thanks for the info guys!

[RikM]

LRGB is a great timesaver. When you are collecting the L, you are getting RG & B all at the same time, in full resolution rather than a few pixel for each as with a OSC.

[Stub Mandrel]

When I split an RGB of the dumbell nebula, the G & B layers showed much better detail than the red layer, but boosting them just turned the red bits grey :-(

It's hard stuff this image processing!

[Scorpius]

Seems like the trick is to get enough RGB to let you adjust the color layer transparent enough for the L to show through without washing out the color.

Which reminds me of something I've been meaning to ask - is there a rule of thumb for how transparent the color layer should end up in PS if you've got the right amount of RGB? Are we talking 25, 50, or 75% - just wondering if there's a ball park number or is it simply dependent on too many other factors?

[Uranium235]

Seems like the trick is to get enough RGB to let you adjust the color layer transparent enough for the L to show through without washing out the color.

Which reminds me of something I've been meaning to ask - is there a rule of thumb for how transparent the color layer should end up in PS if you've got the right amount of RGB? Are we talking 25, 50, or 75% - just wondering if there's a ball park number or is it simply dependent on too many other factors?

More Ps tricks required to stop colour wash-out:

Instead of pasting the RGB into the Lum layer and blending as colour. Do it the other way round.... paste the lum into the colour at an opacity of 30% (blend as luminance), merge, boost saturation, then paste it in again at 50%, boost saturation, merge, paste again at 75%, boost saturation.... you can see where that is going    Do it in little bits.

[Uranium235]

But it doesn't explain the RRGB method described by Robert Gendler, particularly his comparative images of the horsehead in LRGB and RRGB.  Except that his RRGB image took nearly twice as much time, presumably because the extra time was just using the red filter? And no clear filter used for this?

That method would considerably increase the amount of exposure you have to do in RGB because you would need to collect a larger stack R,G & B to get the noise down - pretty inefficient.... or should I say R, G & Cloud!!   Its something we cant afford to do in the UK.

[Scorpius]

More Ps tricks required to stop colour wash-out:

Instead of pasting the RGB into the Lum layer and blending as colour. Do it the other way round.... past the lum into the colour at an opacity of 30%, boost saturation, then paste it in again at 50%, boost saturation, paste again at 75%, boost saturation.... you can see where that is going    Do it in little bits.

No kidding - so I've been doing it a** backwards this whole time?

Think maybe it's time to have another go at some of my old subs and then compare the results...

Thanks for the advice and specific instructions

[ollypenrice]

More Ps tricks required to stop colour wash-out:

Instead of pasting the RGB into the Lum layer and blending as colour. Do it the other way round.... past the lum into the colour at an opacity of 30% (blend as luminace), boost saturation, then paste it in again at 50%, boost saturation, paste again at 75%, boost saturation.... you can see where that is going    Do it in little bits.

Yes, but there's more. At each partial iteration of luminance over colour you can slightly blur the luminance layer to reduce noise. Flatten it onto the RGB and re-apply the lum at higher opacity and, again boost RGB saturation and give L a slight blur. It is at the last application of Lum, ideally at 100% opacity, that you no longer apply the slight blur and so you restore the full resolution of the L. If you don't apply the slight blur between iterations I can't see what you gain from the iterative approach. That's not to say that you don't gain, it's just to say that I don't get it!

As has been said, the key point about lum is that it is ultra fast since it is R plus G plus B at the same time. RGB or OSC cannot compete with this speed. It works because you simply don't need as much colour information as you need luminance, making LRGB the most efficient system. If you use red as lum, or Ha as lum (as was often done a few years ago but has, thankfully, fallen from fashion) you are illuminating your image in the light of red (or narrowband red.) How can this be right? Try it on a daytime image. Red is red. Luminance is full visual spectrum. I want my natural colour images illuminated by the full visual spectrum. For me that's the whole point. Not that there are rules, but I'm stating my own imaging intentions here.

Olly

Edited August 6, 2015 by ollypenrice

[Uranium235]

Ooops yes, forgot to mention the blur/smoothing between iterations

From time to time, Ive also used the dust & scratches filter when the colour layer is particularly noisy.

[ollypenrice]

Ooops yes, forgot to mention the blur/smoothing between iterations

From time to time, Ive also used the dust & scratches filter when the colour layer is particularly noisy.

Indeed, and why not? If you go into Lab colour space and apply an insane blur to a and b channels then recombine you will see hardly any (or no) loss of resolution. Essentially it's the same thing, and also explains why you can bin colour.

Olly

[Magnus_e]

I found a great forum post that explaned this type of prosessing a couple of days ago.

The video explande how to do a DSLR LLRGB prosess, but the only difference would be that you use a actual luminnance layer, if you have a ccd.

It made my orion nebula from february a lot nicer to look at

The grayscale (luminocity)

The LLRGB image

My preveous best effort prosessing this image.

Link to Astronomers Do It In The Dark, prosessing video

Wont claim that i have got all the power of this method represented in my image, but the prosess is werry straight forward and a lot less confusing than what i have tried before

[ollypenrice]

It's certainly true that the processing priorities in L are entirely different from those in RGB so processing them separately is a good idea. In RGB you aim for low noise, especially low colour noise, and strong colour saturation, but there is no need for detail. In L you're looking for the faint stuff, the detail and the contrast. Obtrusive noise reduction won't do much damage to RGB but it will certainly harm L.

In LRGB you can extract a synthetic L layer from the RGB and blend it with the real lum layer, too. I find it's worth about 25% of real lum per unit exposure time, so 4 hours' synthetic lum equals one hour of real - on my setups. This won't work if your RGB was binned because it will lack the resolution to be useful.

Olly

[RikM]

At the moment, I just take the extra time and only gather RGB. I stack each separately for colour and then deselect filter separation and stack the whole lot together for an L. I process as for normal LRGB described above, adding the colour in stages, boosting the saturation and reducing the noise each time. Seems the best compromise for me at the moment and I am satisfied with how the images are coming out. (The stars are a mess though, sorry )

[Stub Mandrel]

I'm going to go back to some of my RAWs and give this a try.

[Stub Mandrel]

OK this is probably done in far too much of a hurry but here's my previous best M27:

And here's a quick try at LRGB (combining 20 30s pictures from three nights worth, having deleted more than that number of ones with star trailing). The L layer is just a monochrome conversion of the stacked but unprocessed 48-bit RGB, the RGB is processed by splitting into the three layers and adjusting for best contrast on each one. Recombined there was a strong magenta cast, corrected by shifting colour balance towards green.

Still not sure what I'm doing or how to highlight the red outer 'shell' without giving a red cast to the background. Somewhat hampered by lack of flats. Might have done better by upping yellow instead of shifting towards green, so I didn't kill the red?

[StuartT]

If there is a need to add a luminance channel when RGB imaging, why not when SHO imaging?

[teoria_del_big_bang]

9 minutes ago, StuartT said:

If there is a need to add a luminance channel when RGB imaging, why not when SHO imaging?

People do add Luminance to NB.
I might sometimes have even done so myself to at least one of my NB images.
I have seen people use a synthetic Luminance made from the SHO channels, maybe some even use just the dominant channel (usually Ha),  then make that channel really contrasty and then add as Luminance (sparingly) to the SHO (or whatever pallet used).
I think some say not to use Luminance as it can reduce the natural high contrast you already get between the NB channels.

I say all this from stuff I have read rather than any huge knowledge of doing so, and so this is more my assumption rather than experience.
So my guess is that Luminance may help to bring out some details on certain NB images but not others and to only use it IF it does anything for you.
I guess often it could be just a lot of extra processing for very little return and so often just not worth it.

Steve