The How and Why of Superluminance

Thanks, but I’m looking to understand the reasons why one approach is better than the other. It isn’t in fact always done in the way you described as I discovered when I searched the topic and found people advocating variety of approaches. It is entirely possible that they way you suggest is the best, but I am trying to get to a theoretical foundation of why.

Just like Andrea, I am adding the L, R, G & B channels to create a super luminance.

For this I am using PI ImageIntegration process, with no rejection and and weighting being either PSFSignalWeigth or PSFSNR depending on the object processed (PSFSignalWeigth for galaxies to promote best FWHM and PSFSNR for nebula where theSNR is the most critical).

So, if I have acquired 2h of each filters for example (so 8h in total) and in the case images are identical in term of FWHM and sky brightness, the result will be very close to a 4h Lum exposure : 2h of true Lum + 2h of (R+G+B) …

Daniel

Mark,

There are a couple of reasons for both a separate L-channel and for creating a “super” L-channel.

1) For any given exposure time, the signal will be higher with the L filter than from any of the RGB filters simply because more light gets through. So, the pure Lum-data will go deeper than any of the RGB channels. Remember that SNR varies as the square root of the signal strength so the Lum-data will also be “cleaner” than your RGB data. In a LRGB image, it’s the Lum-data that drives the overall SNR so that’s the primary reason to take the Lum-data in the first place.

2) You are right that you can further improve the SNR by not tossing out the luminance portion of the RGB combination. The trick is to properly combine it with the “real” luminance signal, which requires proper statistical weighting. Simply tossing everything into the ImageIntegration tool does not properly combine the results and in some circumstances, that approach can actually make things worse. I’ve had success with the following method.

1) Extract the Lum signal from the RGB combination converted to LAB space.

2) Make a copy of the file.

3) Make a copy of your Lum data.

4) Load the originals plus the copies (4 files) into the ImageIntegration tool.

5) Average them using “Additive with scaling” normalization and SNR weighting.

You need to use four files simply because the ImageIntegration tool won’t allow a weighted average using only two files. SNR weighting will insure that the combination is statistically “correct” with respect to signal strength. I believe that you could also use the PSF Signal Strength weighting factor and get a similar result. Juan has stated that the later is more statistically “robust” but I’ve never done any experimenting to see if there’s a measurable difference.

Remember that there is a small difference between the bandpass of your Lum-filter and the Lum channel that you pull from the RGB data, but in my experience, that issue has no significant effect on the result.

Hopefully this answers your question.

John

To be clear, I am a believer in the benefits of a super luminance and I understand why it is better. I am really asking about the best approach for creating one.

Mark McComiskey:
John,

That makes sense. I have always felt that running integration using the individual R, G and B masters plus the Lum must be problematic given the massive mismatch in bandwidth. Cloning the Lum and RGB lightness to get over the 4 frame minimum makes a lot of sense.

andrea tasselli · Mar 7, 2026 at 07:03 PM

*Only if you scale the members by their relative SNR

That was my whole point and that’s why I recommended using weighted averaging by either SNR or PSF Signal Weight.

John

A follow-on question: should DBE be applied to the masters before running integration?

Not sure I follow. The unprocessed masters come out of preprocessing without having had gradients removed (apart from the impact of locaL normalization, if used). It’s pretty typical to then apply DBE or the equivalent to the masters to remove gradients that appear in the masters as an early step in the processing.

It occurred to me that removing these gradients from the Lum and the RGB before integrating the Lum and RGB lightness might help avoid any issues created by incompatible gradients and so few “subs” for the integration.

Got it. So you apply gradient removal before creating the Superluminance, just to confirm?

Every time I’ve tried to create a Super Luminance using RGB data, I’ve obviously noticed a fairly clear difference and a meaningful gain in SNR. However, in practice, that improvement is often too subtle to be noticeable to the eye once it’s integrated into my full processing workflow, so for me it rarely feels worth the time required to build it.

My impression is that it might only really be worthwhile if the RGB integration time is much longer than the luminance integration time. But since that’s not really a logical way to structure an acquisition plan, it ends up being a fairly rare situation.

I mainly use luminance data to reveal contrast and dynamic range. In the end, I feel that the most noticeable choices affecting the final image come from aesthetic decisions, how contrast and stretching tools are applied to the luminance alone.

So I tend not to get distracted by things that are objectively better from a data standpoint but that, in practice, don’t translate into a clearly visible improvement in the final image, at least for me.

Scott Badger · Mar 9, 2026 at 03:34 PM

John Hayes · Mar 7, 2026, 05:48 PM

I’ve had success with the following method.

1) Extract the Lum signal from the RGB combination converted to LAB space.

2) Make a copy of the file.

3) Make a copy of your Lum data.

4) Load the originals plus the copies (4 files) into the ImageIntegration tool.

5) Average them using “Additive with scaling” normalization and SNR weighting.

Hi John, I hadn’t heard of this method before and since I’m right at that point in my current processing, I’ll give it a try.

Before extracting the Lab lightness, are you doing anything with the RGB combo; SPCC, gradient correction, BlurX correct only, BlurX sharpen? And/or anything to either the Lum or lightness before integration?

Cheers,
Scott

Scott,

It is probably best to create the super-Lum combination with the image statistics that come right out of the raw stack. So, I’d recommend combining the raw, stacked Lum data with the Lum channel of the combined RGB data right after it’s stacked. Once you have everything combined, then do SPCC, gradient correction, BXT, NXT, or whatever you want on that result.

John