WB improve for Light Pollution Filters

[hixx]

Hi Ivo,
to recreate the lost yellows when shooting with LP filters, would it be possible to include a "yellow" creation filter which would increase red&green levels simultaneously by the same absolute amount (=yellow), in case pixels already show high correlation between R&G levels? This should work ok for plank curve sources like stars & galaxies. This would drag both orange & light green tones more towards yellow, widening the spectrum for more "natural" tones. Some Filters like IDAS LPS only cut only few yellow energy which could be recreated, so users suffering from light pollution certainly would benefit. It would be a simpler approach over shooting different luminance & color stacks. regards, hixx

[admin]

Hi Ivo,
to recreate the lost yellows when shooting with LP filters, would it be possible to include a "yellow" creation filter which would increase red&green levels simultaneously by the same absolute amount (=yellow), in case pixels already show high correlation between R&G levels? This should work ok for plank curve sources like stars & galaxies. This would drag both orange & light green tones more towards yellow, widening the spectrum for more "natural" tones. Some Filters like IDAS LPS only cut only few yellow energy which could be recreated, so users suffering from light pollution certainly would benefit. It would be a simpler approach over shooting different luminance & color stacks. regards, hixx

Hi,

This is essentially what color balancing in the Color module already does (e.g. it allows you to establish the best possible middle ground by boosting R&G, either by eyeballing or sampling objects or areas such as stars or galaxies). As you know and point out, small parts of the spectrum are removed. However, only large parts are recorded at once (in the red, green and blue channels). It is fundamentally impossible to target just a small part of the spectrum (e.g. use somehow use R&G correlation) afterwards; it is a little bit like trying to unscramble a scrambled egg.

Does that make sense?

[hixx]

Hi Ivo,
I was referring to something different. The Color Module applies a R/G bias correction applied to the whole image. I am talking about the software checking the pixel's R/G levels first and in case the correleation was high (i.e. similar R/G values), the filter would add the same absolut amount of R&G. If the correlation is low, it would leave R/G levels alone. Same below a certain threshold.
Example 1: a yellow star without LP filter may have R=90% & G=90%. With a LP filter on, yellow is reduced, but because stars are black body radiators, we would still see quite equal R/G values let's say R=70% & G=70%. All the Tool would do is detect this correlation (R/G=1) and boost R & G by the same level amount (not percentage!) which effectively adds yellow and restores the initial brightness of the yellow tone.
Example 2: an orange star without LP filter may have R=90% & G=60%. With a LP filter on, the yellow is reduced, but we would still see similar R/G values let say R=70% & G=35%. The star has now become very orange. This is what we see on LP shots, the yellow stars turn into either green or orange! All the Tool would do is detect the correlation (R/G=0,5) and boost R & G by the same level amount (not percentage!), yet with a smaller amount. This effectively adds yellow to the orange star only and turns its hue back towards a more yellow orange. Same is for "greenish" stars that are actuallly yellow-white.
Example 3: An O-III region shows 90% G but only 10% G. The tool would not see correlation between R&G channels hence leave the pixels alone and NOT add yellows here. Similar is for H-alpha with high R and low G.
Example 4: H-II regions with both H-alpha and H-beta. This is would make the tool work "in error". One may work on a Star Mask or increase the threshold, so it wold work on light yellow stars only.
Example5: Dark background may have a high correlation, but will be under a threshold below the Filter would not work, so black will not be changed.
Example 6: A white star may have R=G=B=70%. It would have sight yellow added, due to high correlation between R&G. This helps turning the rather hollow "LED" white tone more into a natural white as yellows become present again.

And that's the differences to the Color Module:
- it will add a yellow amount (same amount of R&G) instead of applying a constant factor to the R resp. G.
- it would do so depending on the actual pixel content, instead of applying a constant "yellow filter" to all selected pixels
- by doing so it would shift light orange and light green hues towards yellow, effectively widening the spectral coverage - in other words: recreate the lost yellows of black body radiators per pixel.
- It would leave all pixels of different hues alone, i.e. work selectively

I agree recreating colors is fundamentally not possible. This approach will surely fail shooting a natural landscape with LP filters. Yet, I think it still would improve stars & galaxies, because luckily these are black body radiators which preserve some color information even if small portions got cut.The only other objects in Deep Sky would be Gas clouds with low correlation (except H-beta, example 4). Hence I think it would work.
And yes, I agree this is not a scientific approach like shooting an extra color stack but the again LP filters aren't! Yet this feature would help improve shots of all the folks that need LPs and may not have the time for extensive nightshifts . But the math for that would be rather simple - probably a small set of wide and narrow gauss curve responses for the R&G correlation would do, along with a threshold plus a total "strength" parameter to control the "yellow" amount.

regards,
hixx

[admin]

Gotcha. StarTools' Tracking information definitely would make things easier. I have been thinking along similar lines, but could never decide on an approach to experiment with. I guess the biggest thing that is holding me back is having "ground truths" to work with, so we can validate/test the solution for various filters. Do you own a filter?

[hixx]

Hi Ivo,
yes I use an IDAS LPS-D1. For DSLR lens I use an LPS-P2. Both are very similar with few cutouts so would be a good start.
Both allow yellow greens (570nm) and orange (620nm) hues pass in a small 10 nm band and cut 580-610nm (plus other bands). It is not so much the lost energy in the yellow tones, that makes color balancing hard, but the fact that 5nm does make a huge difference in hue at the R/G crossover. This is different behaviour compared to the G/B crossover. However, as StarTools is squeezing the least bit of information out of the data, I can for example see two distinct zones in M31 core shots: one in orange and one in light green. Operating the color module, you may see the hue "switch" between the 570/620 nm colors. This is why I think a "yellow creator" would help with adding hues to get a nicer yellow gradient for galaxy cores and stars. In a "well" color balanced shot white stars look still greenish. Trying to fix with R/G bias usually brings up a reddish/purple tint over the whole picture. Using the Cap Green works OK, but leaves a purple tint on the rest.

I am not sure how tracking actually would help here, as the "yellow" information is lost in the original stack as it enters StarTools. However, it is still "coded" in the physical law of black body radiation. This approach should even improve the purple tint usually present in typical CLS shots. These usually cut between 530nm and 650 nm, and would probably need a wider "correlation anlaysis" curve in order to work on a wider spektrum. You are right, it would need some trial to fine tune the correlation analysis to create presets for different filters.
regards,
hixx

[admin]

Hi Ivo,
yes I use an IDAS LPS-D1. For DSLR lens I use an LPS-P2. Both are very similar with few cutouts so would be a good start.
Both allow yellow greens (570nm) and orange (620nm) hues pass in a small 10 nm band and cut 580-610nm (plus other bands). It is not so much the lost energy in the yellow tones, that makes color balancing hard, but the fact that 5nm does make a huge difference in hue at the R/G crossover. This is different behaviour compared to the G/B crossover. However, as StarTools is squeezing the least bit of information out of the data, I can for example see two distinct zones in M31 core shots: one in orange and one in light green. Operating the color module, you may see the hue "switch" between the 570/620 nm colors. This is why I think a "yellow creator" would help with adding hues to get a nicer yellow gradient for galaxy cores and stars. In a "well" color balanced shot white stars look still greenish. Trying to fix with R/G bias usually brings up a reddish/purple tint over the whole picture. Using the Cap Green works OK, but leaves a purple tint on the rest.

I am not sure how tracking actually would help here, as the "yellow" information is lost in the original stack as it enters StarTools. However, it is still "coded" in the physical law of black body radiation. This approach should even improve the purple tint usually present in typical CLS shots. These usually cut between 530nm and 650 nm, and would probably need a wider "correlation anlaysis" curve in order to work on a wider spektrum. You are right, it would need some trial to fine tune the correlation analysis to create presets for different filters.
regards,
hixx

Likely, the most optimal solution would start with a custom DLSR matrix, which addresses the r, g and b response (and channel overlap) after filtering with the LP filter. An image of a test chart under appropriate lighting conditions (e.g. overcast daylight) shot with and without filter should be a good starting point. I would then create a matrix optimizer to achieve the best possible psychovisual color skew minimization, weighted specifically for sRGB-corrected highlights (stars) and midtones (galaxies, stellar profiles) as opposed to the full dynamic range.

StarTools Tracking helps in a few ways; it makes it possible to recover coloring in the linear domain throughout the processing, as well as keep track of "true" highlights and midtones (rather than "final result" highlights and midtones that may/will be skewed due to local optimisation strategies), adding SNR Tracking into the mix allows more accurate coloring in lower SNR areas by noise reducing chrominance information prior to feeding it through the matrix (to avoid garbage in - garbage out effects).

Do you have a test chart at your disposal? Even just creating a test scene with some flat shaded colorful objects could be useful for initial testing...

[hixx]

I do not own a test chart, so I compiled a PDF for print out. I hope this approach would work for initial testing. I shared it with you on the dropbox. It contains various pads, please let me know which would work best for you. You need to wait for the next weekend for the shots however. We have tons of overcast days here now, but daylight will be gone upon my return from work.
I think I got the tracking thing now... It would be able to detect "yellow" in the linear data based upon the R/G correlation, so no processing would skew the original colours. Actually it wouldn't work without.

[admin]

I do not own a test chart, so I compiled a PDF for print out. I hope this approach would work for initial testing. I shared it with you on the dropbox. It contains various pads, please let me know which would work best for you.

The second chart would be probably be most helpful, but perhaps you can do both?

You need to wait for the next weekend for the shots however. We have tons of overcast days here now, but daylight will be gone upon my return from work.
I think I got the tracking thing now... It would be able to detect "yellow" in the linear data based upon the R/G correlation, so no processing would skew the original colours. Actually it wouldn't work without.

Correct! It would then - as well - go on to evaluate how local SNR per pixel and local+global stretching should impact spectrum reconstruction in the final image.

[hixx]

Hi Ivo,
I put test shots on the dropbox. These are Sony RAW files, shot with an unmodified Sony A7II. The TestChart is a combined version of the 2 drafts. Includes standard color checker fields as well as my own "astro color" design. Hope it is useful. - https://www.dropbox.com/s/5oh850bvv2j31 ... t.pdf?dl=0

The 3 shots include:
- clear: without filter - https://www.dropbox.com/s/fg4b98fk67zvs ... r.ARW?dl=0

- Haida Clear Night: frequently used filter for wideangle night fotography, since no interference coating involved. Has a weird frequency response, effectively enhancing red and blue, suppressing green & yellow. Gives a blue-magenta tint. Note the saturated reds. Similar type as Baader Neodymium, Hoya Red Enhancer etc. - https://www.dropbox.com/s/xk5kvtheq9ex0 ... t.ARW?dl=0

- IDAS LPS-P2: Interference Filter. Very small cut outs for Mercury and Sodium lamps. Other than most LP filters it has 2 pass bands at 570 and 620 nm. Gives the usual teal tint. Same as LPS-D1, except higher red cutoff around 690nm (compared to 670 nm for D1 type)
https://www.dropbox.com/s/xglnqtqipa3t4 ... 2.ARW?dl=0

The raw shots reveal the tint, but please try a white balance on the white area for all shots. You`ll see great white tone for all, but note the darkened "dirty" yellow tones in the top bar (yellow response test). Also the Star Color test fields reveal less distinctible white tones du to the missing yellows.
These have high R/G correlation (>0.8). The higher the correlation value, the more dirty the yellow gets in a 1:1 comparison to the clear shot.

Hope this helps
hixx

[admin]

Much appreciated!

Between almcl and your datasets I should, at the very least, have something to play with. No promises obviously, but let's see what we can do with this...

Thanks again!