StarTools

So there's been rather lengthy discussion over on CN-EDSI as to whether it is better to spend time taking L, or to devote that same time gathering more of RGB instead. I think this would interesting to those of us doing mono.

As the whole thread now spans almost 30 pages, I haven't been able to follow everything, but would break it down as follows:

• Over my head
• Alleged apples-to-apples comparisons that might not be realistic
• Not applicable to us due to ST compose and split luminance-chrominance reapplied in Color
• Gone off the rails

Even so, the basics as to what is the most efficient way to amass the cleanest resulting image is still applicable, I would think.

I also wonder (and will try some experiments) if massaging the relative weighting in Compose via the exposure sliders to be different than default would be of use. I have already done this at times for narrowband, where as a result of both camera sensor sensitivity and weak originating signal, the quality of (usually) OIII and SII is not relatively commensurate with the exposure times compared to the Ha. Is there any usefulness for that in broadband LRGB/RGB Compose, or is the sensor's QE curve already baked into signal data so that equal weighting for Synth L remains appropriate?

Thoughts?

First little experiment (partially) complete.

I have yet to dive deeper into my archives, but the data in front of me right now followed my usual OCD pattern of acquisition -- mostly equal total exposure each for L, RGB, and Ha. Ditto on the second night.

That makes it easy to run off some comparisons, at least if I didn't toss interim stacking files into the recycle bin. And for the Horsehead I did not.

So, I was able to load my night one LRGB into compose. I kept default weighting by exposure (total was 146m: 72, 25, 25, 24), cropped the edges, and saved out the linear TIFF, meaning my 16-bit L+ synthetic L in "grayscale" (each channel identical). Then I composed a synthetic L from my two-night stacked masters. Not identical but close enough I think (total 134m: 45, 45, 44), cropped similarly and saved.

Loading each into PI and running the noise estimator script gave me 0.2626 for the LRGB, and 0.3446 (i.e. higher noise) for the RGB only.

I think that makes sense. My filters are all one LRGB kit, and so the L should match the broken up R, G, and B pretty well for bandwidth. ST Compose was all weighted by exact exposure. So, more efficient photon collection per time, leading to a better and cleaner synth L for processing.

The question then becomes what happens after color composition, since for the (longer) RGB-only data a cleaner chrominance will have been stashed away for the Color module.

Unfortunately I have not yet found a way to add that chrominance data back in (wanting to use ST's default Scientific + CIELab Retention if I could) while linear and allowing a save-out, so that I could then run a noise estimator on the file. ST was not cooperating with me on my big plans.

Guess it isn't built for that. I have to think up some other way I can get an objective measure after luminance+chrominance.

...continuing my conversation with myself...

I was able to experiment more last night. Again maybe something I was doing wrong, thought-wise, but never could get ST to accept a non-linear L to combine with linear RGB to apply as per the Color module. The plan had been to use another program to stretch, where I knew it would be exactly identical in lieu of dynamic. Well, probably just not meant for mid-stage insertion that way.

Within ST, I did try to use FD for stretch, with only the digital development turned on, and then subsequently colored. Similar results with the LRGB objectively less noisy in the noise estimator script. But again, possibly not real world, since we would always process to the data in front of us (as does ST).

So then I just decided to do a full independent ST processing, matched as closely as possible, but letting ST do all it's dynamic adjustments in the places it wants to do that.

At first glance, the results of LRGB vs similar-total-exposure RGB are remarkably similar. The noise estimator on the final images though, again puts LRGB well ahead.


But with closer inspection, while noisier, the RGB-only rendition seems to have tighter stars, and obviously so. Perhaps a little cleaner on the structural detail (to the extent seen past the noise, of course). I suppose there could be several things behind that.

Seeing and thus HFR differences night to night, for one. I suppose I should run a separate LRGB stack only of the second night, to check on that. But of course, any time we are using different ingredients for stacks, there will be differences. WBPP will select different reference frames to stack against, and to generate LN modeling.

There could also be variance between the masterlights, so I guess I will have to explore that next. Using 20s for L and 60s for each RGB sub, my intent was to have them match up well. Meaning, the L wasn't so comparatively overexposed that it bloated. I'll have to see.

Another possibility is that I subconsciously, or ST dynamically, stretched the LRGB a wee bit more? Lastly perhaps there is something in the way that ST composites LRGB.

Despite being as similar as you see, and with the RGB stars "tighter," the LRGB had significantly better PSF profiles for SVD. Not that SVD didn't work well on both composites, because it seemed to despite smaller but less clean PSF samples, but the LRGB needed noticeably more deringing strength (and showed the remnant glitches in 1.9 on that).

So, will continue playing with the data and maybe try to dig up some more from the archives. I'm not sure I have anything deep though, likely just more short integration projects.

But as of right now, not seeing that I am ready to give up using the L filter.

just sharing to me on my monitor the LRGB stars are slightly stronger in colour seen most on amber stars where as the LGB are less colour intense, is that maybe why they appear slightly tighter

Mike in Rancho wrote: ↑Mon Feb 26, 2024 5:46 pm ...continuing my conversation with myself...

That's much better than talking to the little white rabbit on your shoulder, Mike!

Hi Mike,
An interesting topic
Gee this LRGB vs RGB thing has been thrashed around on so many forums over many years and I still don’t think there’s a definitive answer , too many variables involved and so subjective.
However thanks for raising it in the ST forum.
With my limited experience with Mono and dedicated filters and limited clear skies to test theories,in regard to Broadband imaging ,at this early stage in my Mono journey I have decided to image RGB from my Bortle 8 location and L RGB from my Bortle 3 location.
I tried LRGB on NGC 253 Sculptor Galaxy from my Bortle 8 location and the Noise introduced by the Luminance subs (30 sec) was significant with no further fine detail achieved ( integration was 6 hours and ratio was 3:1:1:1 ) RGB subs were 60 sec
Decided to process the same data set in just RGB using Synthetic Luminance and the results were marginally cleaner with better detail.
Maybe with 10 to 15 hours of data ( or more ) using LRGB the result would be better.
Luminance under Bortle 8 is a real problem but under Bortle 3 can only improve the image.
I don’t have any comparison images to show you
My limited experiences to date

Clear Skies
Martin

happy-kat wrote: ↑Mon Feb 26, 2024 7:02 pm just sharing to me on my monitor the LRGB stars are slightly stronger in colour seen most on amber stars where as the LGB are less colour intense, is that maybe why they appear slightly tighter

Yes, no, maybe?

For the star tightness, no it was actually spatial. I would have made a blinking GIF, but with the images already color I'd have to convert everything to grayscale (I can't seem to make a decent color GIF).

Interestingly, I actually was seeing the color strengths the other way. The RGB, being tighter, had coalesced more color towards the cores, thus appearing stronger in color. Though I suppose that's likely some kind of illusion. Also, the LRGB, for whatever reason, seemed to end up a bit more stretched by OptiDev. Which in my eye led to a wee bit paler coloring.

So what I did tonight was to process a third version, that being the whole nine yards of LRGB from both nights. Again I let ST do its dynamic thing where it wanted to, but otherwise mostly matched the processing.

First, the objective test using PI's noise estimator showed the best (less noise) numbers of all. Expected. Second, the stars indeed became tighter, even though it was LRGB. So, I am reasonably certain that night two had much better seeing. So, star spatial sizes on my 266m LRGB are very similar to my 134m RGB, meaning tighter than my first night 146m LRGB.

Anyway there could be other aspects at play as well. Trying to get perfect comparisons with differing underlying data turns out to be pretty hard, I am learning! And again especially with so much of ST being dynamic to the data rather than absolutes. Beyond OptiDev itself, even Wipe might have...wiped...things ever so differently.

Speaking of Wipe, when toggling between the courtesy stretches of Luminance and Color, I was noticing a concerning disconnect between the two as far as the stretches spatially matching for the stars (the luminance was bigger). Wondering if that's a potential problem or carries over, and if so why. Then again I'm not completely certain yet on the mechanics of how the linear chrominance gets combined into the non-linear processed luminance.

Finally, I guess I also have to re-think the SVD PSF sampling I got from the one-night LRGB, which I thought was good based on the white outlines and sampling colorization that was in front of me. But now I'm starting to think the bloated stars from the bad seeing were giving me false top candidates for stars to click on. Maybe kind of like the artificial blur discussed in the other recent thread.

decay wrote: ↑Mon Feb 26, 2024 7:21 pm That's much better than talking to the little white rabbit on your shoulder, Mike!

Dietmar you finally did it! A reference that left me, for a change, completely clueless. Some kind of Teutonic fairy tale, perhaps? I thought they were all about wolves and witches, not white wabbits.

Startrek wrote: ↑Tue Feb 27, 2024 1:07 am Gee this LRGB vs RGB thing has been thrashed around on so many forums over many years and I still don’t think there’s a definitive answer , too many variables involved and so subjective.

Luminance under Bortle 8 is a real problem but under Bortle 3 can only improve the image.

Hi Martin.

I guess I have to admit my ignorance. I mean I guess I realized one could just take RGB only, but never really gave it much thought. Didn't know it was an area of debate even, until it popped up recently.

I'm not seeing how taking L could be a problem though, as with my (limited so far I guess) experiments above the L was rather beneficial. This data is all B8 too, of course. 3 hours of L, meaning 6 total, should be noticeably better than 3 hours RGB only. Both subjectively and objectively.

That seems to call for some further exploration to see what happened.