Possible New LRGB Technique?

[Jack Harvey]

After working with the New Ha Combine Technique described in another thread on this forum I wondered what would happen if I just applied this idea to a simple LRGB image.  So I took a processed RGB tiff image and extracted the luminance from it with the Channel Extract tool using CIE with only L checked.  One of the things that strikes me is the large amount of noise in this extracted Lum image.

Extracted Luminance from RGB.tiff


Then I took a Mean Luminance frame for the same object and stretched it to come closwe to the stretch in the extracted Lum.  Much less noise.

Processed New Luminance frame


I then extracted the R,G, & B channels from the original RGB.tiff (which no longer has its original luminance) and placed the New Luminace frame and the R,G, and B frames in the appropriate slots of the LRGB combine tool.  Here is the LRGB with only minor histogram tweaks.


This is just preliminary but it seems to give a pretty nice image with not as much processing in the end.  Obviously this image could be enhanced quite a bit, but I wanted to show the "raw" result.  It would seem to me that we are removing a fair amount of noise seen in the extracted Lum frame.

[Juan Conejero]

If I understand this, what you are doing is replacing the luminance of an already processed (i.e. stretched) RGB image with a new luminance image, after applying an equivalent stretch. Is this right?

This obviously works for the same reason as any LRGB: replace noisy L from RGB with high-SNR L.

What I don't understand is why is this different from a regular LRGB? After all, you have separate RGB and L and combine them to improve SNR.

[Jack Harvey]

I think it is different in that at the time I collect data I use L,R,G, & B filters all binned 1x1.  The Luminance I am extracting is the luminance that is obtained thru the R,G, and B filters and is present when assembling an RGB image.  Now I have a seperate data set of Luminance collected thru a Lum filter which i add after extracting the Lum from the RGB..

How is this different from a regular LRGB image - well I have thrown away the very noisy Luminance component from the R,G,B components???

Does this make a huge difference?  Too early for me to tell, but I like throwing away noise.  And it seems the Luminance filter collects better lum data than the individual R,G, & B filters?

[Juan Conejero]

I understand what you are doing. But... with PI's LRGBCombiantion tool you can do exactly that!  

You can throw away the luminance of RGB -which I call the implicit luminance- as the very first step of LRGB combination:

- Open the four linear raw images (R, G, B, L)

- Place all of them in the corresponding slots of LRGBCombination.

- Leave all parameters with default values (channel weights = 1, Uniform checked, functions = 0.5, no noise reduction).

- Apply the process globally (click the blue sphere button)

Now you have a LRGB image that has no implicit luminance at all. The luminance comes now exclusively from the L image.

Furthermore, you have a very important advantage: you can apply deconvolution to the luminance of this LRGB image, since it is still linear (with the help of ScreenTransferFunction of course). Thanks to the ability of PI to extract, process and reinsert the luminance on the fly, you can deconvolve the luminance of the LRGB image and see the true effect on the final color image. This is much better than deconvolving the isolated L image before LRGB combination (the traditional method). If you decide to do this, in these cases you should leave the Linear check box of Deconvolution unchecked, since there is no need to force extraction of a linear luminance, as it is already linear. ATrousWaveletTransform also benefits greatly from working with a linear image.

The LRGB image can be stretched in the usual way after deconvolution, wavelets, etc. I think this is just what you are pursuing.

It's great to get these topics rethought. Isn't this a really interesting forum? 8)

[Jack Harvey]

Well this is an aspect of LRGB tool that I was not aware of!  Thanks again for the education

[Juan Conejero]

Thank you Jack. These are really important topics and everybody can learn and improve their work when they are brought up.

It's a pleasure to be in touch with open-minded people that want to find new and better ways to do things.

[Jack Harvey]

If  what you just explained to me is true why then cannot we just use the LRGB tool to do the Ha Technique I have worked out in the other thread.  If we put Ha in the Lum and R,G,B in the appropriate slots, we should dispose of the implicit  Lum and substitute the Ha similar to what I have been doing with several steps?  Correct me if I am wrong<G>!

[Juan Conejero]

You're right; it works just the same way with Ha.

The only "problem" is that Ha is red. This means that if your chrominance includes objects that are not red (stars, OIII, reflection, other broadband), you need part of the implicit luminance to give them support. And this means that some noise from the implicit luminance will survive in the combined result.

But... if you mix Ha and a high-quality luminance before LRGB combination, and use that combination as the L component, things can be much better because you can again throw away the implicit L completely. This would be a LHaRGB combination.

The problem is to know how much L must be added to Ha to get an optimal result for each particular image. If you find a systematic way to do this, that would be a very good trick for sure

[Jack Harvey]

I think that is what Vincent is struggling with<G>

[Juan Conejero]

In some way, yes. However his method is more radical: he synthesizes an artificial broadband red component that has the same SNR as the narrowband Ha image. This artificial red replaces the original red of RGB completely. To achieve this he finds where continuous emission objects are present in the image, in order to reproduce them (the continuum map). This is conceptually simple, but very complex to implement. We still haven't found a solution applicable to all cases consistently, but Vicent will achieve this goal, be sure.

This doesn't invalidate more direct solutions like the LHaRGB combination that we were talking about. I think both are compatible solutions that can be applied successfully, depending on particular cases.