L-(L)RGB Combines

[mmirot]

 I usually make LRGB images. My images are all binned 1x.
How do I use the Luminance created by the RGB data to combine with my native luminance ( L images)?
I have been told that this " L-LRGB " Image has higher Luminance S/N.

Max

[Juan Conejero]

Hi Max,

How do I use the Luminance created by the RGB data to combine with my native luminance ( L images)?

You can do that very easily with LRGBCombination. In the Channel Weights section, select a luminance ratio (L slider) smaller than one. This is a source:target ratio, where source=your_L_image and target=the_RGB_image. For example, if you set L=0.75, then the L image will contribute by a 75% to the final LRGB combination, and the other 25% will come from the RGB image.

The general workflow with LRGBCombination is as follows:

- Select the L, R, G and B source images. If you have combined your RGB data before (the recommended way), then you just have to select the L image; uncheck the rest of channels and they will be automatically taken from the target RGB image.

- Forget the "Target" combo box. This parameter is obsolete and will probably be removed (hopefully in version 1.5).

- If necessary, adjust the channel weights, as I have explained above.

- Tweak the luminance and saturation transfer functions. Note that these are midtones balance adjustments, so by decreasing them you are increasing brightness and color saturation, respectively. Note also that LRGBCombination can work on previews without problems, so you can try out ad-infinitum before actually applying to your image. Of course, define previews on your RGB image.

- You normally will want to use chrominance noise reduction. You must apply more noise reduction when you increase saturation.

- When done with parameters, apply the process to your RGB combined image, which will then become your LRGB combined image.

That's it!

[Jack Harvey]

Maybe I am confused but think you are asking how to get the Luminance from a RGB combine tothe re combine with the native L as well as the R,G,B to get a LLRGB?  If so then I would think you use channel extract to extract the Luminance from the RGB image by enabling only the L channel on the extract tool.  Now you have the Luminance from the RGB.  You can then make your native LRGB from using the LRGB tool and then use Pixel Math to add the extracted Luminance? 

I am probably making this harder than it needs to be.

[mmirot]

Juan,

I thought since PI keeps the RGB as color data and L seperate the LRGB combo would not work for this.
I was expecting an answer similar to Jacks.
Extract the L data from the RGB and combine it with the L in Pix math.


Max

[Nocturnal]

I don't think PI keeps L separate. A color image has RGB and an optional Alpha layer. L is derived from RGB when it is 'extracted'. I think the 'extract' name is a bit of a misnomer as nothing is removed from the source image. 'Derive' is more accurate.

[mmirot]

Then PI is combining L and L(rgb) data automatically?

Max

Ps : Juan will 1.5 soon be here? 

[Juan Conejero]

Hi Jack, Max and Sander,

Indeed PI doesn't keep a luminance component. The internal representation of images only supports grayscale and RGB images with nominal channels (gray or R, G and B) and optionally an unlimited number of alpha channels (from which the first one, which we call active alpha channel, defines opacity).

The CIE L* component (the nonlinear luminance), as well as a large number of different color space components for luminance and chrominance, are always generated on-the-fly and on-demand. This is one of the best reasons to use as many processors as possible with PI

Then PI is combining L and L(rgb) data automatically?

The LRGBCombination tool does that automagically, of course.

Juan will 1.5 soon be here?

Definitely it will hit the street this week

[mmirot]

Now I am really confused.

I would think it only generates a normal LRGB with Lum data only from the L images and color ( CIE a + b) data from RGB channels. Otherwise, the LRGB combine is mixing RGB derived luminance. Perhaps I don't understand this module.

Don't you have to combine a RGB image by turning into a Grayscale image then combine the product with the L  images to get L + L(extract) ? 

This way you get increased S/N  provided RGB exposure has good S/N too.  This has been suggest by several sources.

However, I worry that you can still degrade the end product with out some form of scaling, weighting or normaliztion.

Max

.
Max

[Juan Conejero]

I would think it only generates a normal LRGB with Lum data only from the L images and color ( CIE a + b) data from RGB channels. Otherwise, the LRGB combine is mixing RGB derived luminance.

With the LRGBCombination tool, this is completely up to the user. If you set the luminance ratio (the L value in the Channel Weights section) to 1, then all the luminance in the final combined image comes from the specified L source image. If you set the L ratio to < 1, then you'll get a proportional mix of source and RGB luminances.

Don't you have to combine a RGB image by turning into a Grayscale image then combine the product with the L  images to get L + L(extract) ?

You can do that if you want, but LRGBCombination will do it for you and is more accurate. Anyway, instead of a grayscale conversion I'd recommend extracting the luminance (Image > Extract > Luminance).

This way you get increased S/N  provided RGB exposure has good S/N too.

And provided also that you've used the same binning for RGB and L.

As I see it, mixing RGB luminance with L is a contradiction in terms. IMO, if you feel your SNR improves by keeping some luminance from the RGB data, then you should rethink whether a LRGB combination is actually of any good, or you'd do better by accumulating all of the exposure time as RGB. Just a thought; of course I can be wrong (remember: I'm just the guy that breaks keyboards )

However, I worry that you can still degrade the end product with out some form of scaling, weighting or normaliztion.

Of course. That's why LRGBCombination allows you to fine-tune the final luminance with a midtones balance adjustment (the luminance transfer function). The purpose of this adjustment is to perform a good adaptation between the imported luminance and the RGB chrominance. The basic idea is that all chrominance structures should have sufficient support in the luminance, and vice-versa.

The bottom line is (just IMO, of course): shoot RGB

[Jack Harvey]

We have gotten away from shooting Lum frames and just use unbinned R,G,B

[mmirot]

Actually, the efficiency of the LRGB strategy is what I hope to test. I have not done L 1x Rgb2x in a while.
 I shoot unbinned LRGB.
 I think that shooting the extra L frame may not be that helpful either. 
However, some tests show L+ (L) RGB as helpful. 

see

http://www.astro.uni-bonn.de/~mischa/ps/syntheticlum.html

Max

[mmirot]

BTW Jack,

Did you publish your AIC 2008 talk?

Max

[Jack Harvey]

No I did not as it was a workshop where we started with raw master frames and went through processing on 3-4 different images trying to showcase some particular aspect with each image.  Sorry but it really just did not lend itself to slides etc as it was live and on the fly so to speak.

[Astrocava]

Actually, the efficiency of the LRGB strategy is what I hope to test. I have not done L 1x Rgb2x in a while.
 I shoot unbinned LRGB.
 I think that shooting the extra L frame may not be that helpful either. 
However, some tests show L+ (L) RGB as helpful. 

see

http://www.astro.uni-bonn.de/~mischa/ps/syntheticlum.html

Max

I'm a totally novice in monochrome cameras and shotting through color filters, but I think the comparative should be done with the same time of combined exposure, between:

L     12x600
R    6x1200
G    6x1200
B    6x1200

and

R    8x1200
G    8x1200
B    8x1200

I don't understand why [RGB] RGB is better than RGB? Arent' they the same? If not, why don't use [[RGB]RGB]RGB and make it an iterative method?

Sergio
 

[mmirot]

I guess we should define the combinations as the article does.  [ ] defines luminance with the source of the luminance defined by the filter letters. 

" RGB: a classic image constructed solely of R, G and B exposures
[L]-RGB: an image where an clear- or luminance-filtered exposure is used as the [ luminance ] channel, and the colour information comes from RGB. 
[RGB]-RGB: the luminance channel is synthetically created from the RGB images
[LRGB]-RGB: the luminance image is comprised of an unfiltered exposure and the RGB exposures "



I was under the impression that you could also combine L +[RGB] to create a stronger luminance. So it would be [ [L] + [RGB] ] - RGB  image. A few well know imagers were saying thats one of the reasons why they now shoot LRGB bin 1x.

Quite franky,  I am with Jack. I think acquiring an extra L filter is problematic. It is not that effecient from an acquistion prospective. Perhaps the time is better spent going longer in RGB.
Also, You have to make sure the s/n quality of the color data matches the luminance.

This makes processing tricky. It is very easy to under cut one or more of the color channels leading to chrom. noise in the low S/N regions. 

I think binning RGB 2x doe not usually help if the exposures are already read noise limited.



  I don't why a RGB image would be better than [ RGB] - RGB. 

I would think PI or most software automatically be displaying the [ RGB] as Luminance.


Max