Jump to content

  •  

CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.

Photo

Loss of color diversity in LRGB photography when the filters do not overlap

  • Please log in to reply
307 replies to this topic

#276 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 254
  • Joined: 02 Oct 2017

Posted 23 April 2024 - 05:28 PM

You can already do this with an OSC. We recently proved we could color match with two astrocams—the ASI224MC and the ASI533MC—to very low E. All you need is a UV/IR cut filter. We even got the True Colors of the planets using the lunar regolith to model and remove red shift from atmospheric absorption…

 

https://www.cloudyni...th-an-asi224mc/

 

Take your astrocam, plug in the UV/IR cut, an shoot a CC24 under D65 lighting, and Bob's your uncle.

 

BQ

Absolutely! Very nice, I missed that thread. I was only suggesting that if one wanted to also include more data for some reason, such as UV and IR, a monochrome camera would be able to do that. There is also a resolution benefit as mentioned by slippy



#277 slippy

slippy

    Explorer 1

  • *****
  • Posts: 68
  • Joined: 10 Sep 2021
  • Loc: Seattle

Posted 23 April 2024 - 05:31 PM

Sounds interesting, although I don’t understand that for the most part! It’s still a bit difficult for me to break out of the thought that color is anything other than RGB 0-255. All of the other encoding schemes, white balance, gamma, luminance, etc are lost on me in terms of how they relate to an integer on a colored pixel. But that’s ok, I can acknowledge the basic principle.

 

That just made me realize something though - it works with OSC because that takes a snapshot of all channels simultaneously, and that gives you the relative values of each at a point in time. But if you’re taking long exposures of varying length and in different conditions, hours or days apart, doesn’t that kind of go out the window? It seems like you’d need a common reference to normalize them. Or can something like SPCC provide that? You can’t sneak a color checker into the shot, and can’t really shoot it under the same conditions that you’ll be using for your color data, but if stars have known reference colors, does that fix it?


  • Shubham likes this

#278 badgie

badgie

    Viking 1

  • *****
  • Posts: 649
  • Joined: 13 Jan 2021
  • Loc: SF Area, California

Posted 23 April 2024 - 05:33 PM


Anyway, as for the advantage, OSC probably already does a decent job of this. However, it’s inherently lower resolution due to the bayer matrix, and since most people already have and prefer mono cams, it would be a way to use those similar to how they use them for RGB data today, just with a bit more spectrum coverage. But because they’re all curves, it would still be less efficient than RGB.

 

I’m still not sure I’d bother going the tristimulus route anytime soon, just exploring ideas.

RE resolution CFA drizzle handles this so not a major impact on resolution or integration time, except for 2x on green compared to red.  But as a a mono camera user I catch you point!



#279 BQ Octantis

BQ Octantis

    Voyager 1

  • *****
  • Posts: 10,623
  • Joined: 29 Apr 2017
  • Loc: Nova, USA

Posted 23 April 2024 - 06:25 PM

Sounds interesting, although I don’t understand that for the most part! It’s still a bit difficult for me to break out of the thought that color is anything other than RGB 0-255. All of the other encoding schemes, white balance, gamma, luminance, etc are lost on me in terms of how they relate to an integer on a colored pixel. But that’s ok, I can acknowledge the basic principle.

 

That just made me realize something though - it works with OSC because that takes a snapshot of all channels simultaneously, and that gives you the relative values of each at a point in time. But if you’re taking long exposures of varying length and in different conditions, hours or days apart, doesn’t that kind of go out the window? It seems like you’d need a common reference to normalize them. Or can something like SPCC provide that? You can’t sneak a color checker into the shot, and can’t really shoot it under the same conditions that you’ll be using for your color data, but if stars have known reference colors, does that fix it?

 

sRGB is indeed a set of color vectors with components on the interval [0,1]. But visual-matched color requires negative values of sRGB because the sRGB gamut is a crop of the visible color horseshoe. No amount of white balance will fix this—only a CCM on a set of filters that match the reasonably match the Luther-Ives condition will do so. Even then, there is no display that can display all the visible colors.

 

SPCC only provides the white balance scale factors for the three sRGB channels to achieve some white reference. But the white reference for AP is arbitrary—read Section 7.3 of the PI RTFM.

 

Even B-V is arbitrary—not to mention, white is in the eye of the beholder

 

BQ


Edited by BQ Octantis, 23 April 2024 - 06:31 PM.


#280 Borodog

Borodog

    Hubble

  • *****
  • Posts: 12,928
  • Joined: 26 Oct 2020
  • Loc: St. Augustine, FL

Posted 24 April 2024 - 02:53 PM

And another who completely misunderstands the issue…

Oh I understand the issue quite well. I've been saying for a long time that the loss in effective color bit depth from using non-overlapping square band pass RGB filters is almost as bad as shooting L, which is why LRGB images have a certain "only primary colors need apply" look to them that you can tell immediately. I'm just saying that people that get hung up on supposed "accuracy" in AP are barking up the wrong tree. There is virtually nothing remotely "accurate" about literally any part of AP.


  • Shubham likes this

#281 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 254
  • Joined: 02 Oct 2017

Posted 24 April 2024 - 03:28 PM

Oh I understand the issue quite well. I've been saying for a long time that the loss in effective color bit depth from using non-overlapping square band pass RGB filters is almost as bad as shooting L, which is why LRGB images have a certain "only primary colors need apply" look to them that you can tell immediately. I'm just saying that people that get hung up on supposed "accuracy" in AP are barking up the wrong tree. There is virtually nothing remotely "accurate" about literally any part of AP.

Well, that is a choice people make for themselves. One can choose accuracy sometimes.



#282 Borodog

Borodog

    Hubble

  • *****
  • Posts: 12,928
  • Joined: 26 Oct 2020
  • Loc: St. Augustine, FL

Posted 24 April 2024 - 03:45 PM

Well, that is a choice people make for themselves. One can choose accuracy sometimes.

They could. But they don't. Because nobody would like the results. Because what astronomical objects actually look like to the human eye doesn't actually look that good, if they're even visible at all.

 

As an example, the only people I've ever seen go to the trouble to process a planetary image the way it would look in the telescope are Tulloch and BQ Octanis, and not offense to them, but it looks like washed out hot garbage.

 

Long integration times? Right out. The human eye has an effective integration time of about 0.1s.

 

Super hard stretches? Right out. Better stick with gamma = 2.2 for accuracy.

 

All that pretty red Hα? Practically gone, its nearly invisible to humans (that's why DSLRs filter almost all of it out).

 

All those pretty colors? Gone. Almost all of these objects, if you can see them in the telescope at all, are monochromatic to the human eye.

 

The only exceptions that I know of are some kinds of lunar imaging and white light solar.


  • licho52 and Shubham like this

#283 BQ Octantis

BQ Octantis

    Voyager 1

  • *****
  • Posts: 10,623
  • Joined: 29 Apr 2017
  • Loc: Nova, USA

Posted 24 April 2024 - 04:02 PM

The only exceptions that I know of are some kinds of lunar imaging and white light solar.

Nope. If you output a linear (i.e., gamma compressed to accomodate the gamma correction of the display driver), color-correct (i.e., corrected for atmospheric yellowing for the given target altitude) image, they, too, are washed out hot garbage.

 

BQ


Edited by BQ Octantis, 24 April 2024 - 04:03 PM.


#284 slippy

slippy

    Explorer 1

  • *****
  • Posts: 68
  • Joined: 10 Sep 2021
  • Loc: Seattle

Posted 24 April 2024 - 04:03 PM

They could. But they don't. Because nobody would like the results. Because what astronomical objects actually look like to the human eye doesn't actually look that good, if they're even visible at all.

 

As an example, the only people I've ever seen go to the trouble to process a planetary image the way it would look in the telescope are Tulloch and BQ Octanis, and not offense to them, but it looks like washed out hot garbage.

 

Long integration times? Right out. The human eye has an effective integration time of about 0.1s.

 

Super hard stretches? Right out. Better stick with gamma = 2.2 for accuracy.

 

All that pretty red Hα? Practically gone, its nearly invisible to humans (that's why DSLRs filter almost all of it out).

 

All those pretty colors? Gone. Almost all of these objects, if you can see them in the telescope at all, are monochromatic to the human eye.

 

The only exceptions that I know of are some kinds of lunar imaging and white light solar.

Color is orthogonal to exposure, and you’re wrong about visible wavelengths, so, much of that is actually not true at all, but more importantly, it’s besides the point and off topic.



#285 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 254
  • Joined: 02 Oct 2017

Posted 24 April 2024 - 04:52 PM

Color is orthogonal to exposure, and you’re wrong about visible wavelengths, so, much of that is actually not true at all, but more importantly, it’s besides the point and off topic.

There is a long thread on natural color here:

 

https://www.cloudyni...ue-color-again/

 

That would be a more appropriate place for this subject. The things you have mentioned have been addressed there, for the most part.



#286 SeymoreStars

SeymoreStars

    Aurora

  • *****
  • Posts: 4,712
  • Joined: 08 May 2014
  • Loc: Pennsyltucky

Posted 24 April 2024 - 05:19 PM

I haven't had time to read the whole thread. Have the "true" color police showed up yet?

 

I have both Chroma RGB and "classic" RVB filters. Which would produce the best result on an RGB subject?

 

 

Attached Thumbnails

  • Screenshot 2024-04-24 181434.png
  • Screenshot 2024-04-24 181227.png

  • Shubham likes this

#287 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 254
  • Joined: 02 Oct 2017

Posted 24 April 2024 - 05:26 PM

I haven't had time to read the whole thread. Have the "true" color police showed up yet?

 

I have both Chroma RGB and "classic" RVB filters. Which would produce the best result on an RGB subject?

The second set would be the better of the two, to address the loss-of-color-diversity issue discussed in the present thread.

 

If you really care about the issue of true color, please read the other thread.



#288 SeymoreStars

SeymoreStars

    Aurora

  • *****
  • Posts: 4,712
  • Joined: 08 May 2014
  • Loc: Pennsyltucky

Posted 24 April 2024 - 05:31 PM

The second set would be the better of the two, to address the loss-of-color-diversity issue discussed in the present thread.

 

If you really care about the issue of true color, please read the other thread.

I have no desire to dicuss "true" color. LOL



#289 slippy

slippy

    Explorer 1

  • *****
  • Posts: 68
  • Joined: 10 Sep 2021
  • Loc: Seattle

Posted 24 April 2024 - 05:42 PM

The first set would not be able to take pictures of the sodium tail on some comets or mercury, the second would. In this sense it’s not even an accuracy issue, it simply can’t even capture it because of that hole between filter bands. The second set would have weak response to Ha, Nii, Sii, and some other less common lines, but it would still see it.



#290 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 254
  • Joined: 02 Oct 2017

Posted 24 April 2024 - 05:54 PM

I have no desire to dicuss "true" color. LOL

I figured that. I thought I would be polite anyway.


Edited by loujost, 24 April 2024 - 05:55 PM.


#291 Oort Cloud

Oort Cloud

    Fly Me to the Moon

  • -----
  • Posts: 6,944
  • Joined: 19 Nov 2020
  • Loc: New Jersey, USA

Posted 24 April 2024 - 06:46 PM

The first set would not be able to take pictures of the sodium tail on some comets or mercury, the second would. In this sense it’s not even an accuracy issue, it simply can’t even capture it because of that hole between filter bands. The second set would have weak response to Ha, Nii, Sii, and some other less common lines, but it would still see it.


If the intended purpose of these filters is astronomy, you'd think they would have extended the red coverage some to better "see" at a minimum, H-alpha.

I still want a set, but I'd probably go with Astronomik, as I'm sure they're a fraction of what Chroma charges.

#292 badgie

badgie

    Viking 1

  • *****
  • Posts: 649
  • Joined: 13 Jan 2021
  • Loc: SF Area, California

Posted 25 April 2024 - 08:29 AM

We can approach this problem directly as the challenge of mapping a unit interval into a set of x numbers using n basis functions.  Typically, those numbers are called R, G, B and n = 3.  I see two competing objectives:

1: Capturing as many photons as possible

2: Providing the maximum separation between (in information theory sense) between nearby points on the unit interval.

3??  The ability to capture information in the context of strong backgrounds (e.g. capture OIII with a strong Ha background)

 

Any other useful criteria?

 

This is similar to BQ Octanis's analysis except the hue analysis includes the response of the eye (If I understand correctly) as opposed to information content. 



#293 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 254
  • Joined: 02 Oct 2017

Posted 25 April 2024 - 08:49 AM

We can approach this problem directly as the challenge of mapping a unit interval into a set of x numbers using n basis functions.  Typically, those numbers are called R, G, B and n = 3.  I see two competing objectives:

1: Capturing as many photons as possible

2: Providing the maximum separation between (in information theory sense) between nearby points on the unit interval.

3??  The ability to capture information in the context of strong backgrounds (e.g. capture OIII with a strong Ha background)

 

Any other useful criteria?

 

This is similar to BQ Octanis's analysis except the hue analysis includes the response of the eye (If I understand correctly) as opposed to information content. 

In fact my title for this thread referred to diversity in the information-theoretic sense. This could even be quantified, comparing the diversity of the signal to the diversity of the output. That's the basis of my own scientific work, in biology. I explain the connection between information (Shannon entropy) and diversity (they are not exactly equivalent) in articles such as

https://www.research...diversity_Oikos

https://www.research...Beta_Components



#294 Shubham

Shubham

    Vostok 1

  • *****
  • Posts: 159
  • Joined: 04 Jan 2023
  • Loc: DFW, TX

Posted 09 May 2025 - 03:33 PM

Wow! This thread has been enlightening to say the least! I came with a desire to go with Chroma’s square cut-off no-overlap Bessell UBVRI filters (the ones that Frank said he has) being under the impression somehow that ALL colors, spectral or non-spectral could be faithfully captured with just RGB and thought of any overlap as “contamination.” My reasoning was somewhere along the lines of “having well defined, deterministic and consistent color across different objects and having good color information that is interpretable and informative” as Frank also suggested (although I’d certainly appreciate further elaboration, Frank). But after going through this thread, I’m certainly gained more knowledge and at the same time even more confusion but in a good way.

 

I think we can all agree that gaps are bad, whether you want to go with non-overlap or w/ overlap filters. You lose information if you have gaps.

 

I’m now more interested in comparing Chroma’s “Classic” Bessell filters(w/ overlaps) to the square cut-off Bessells that I was initially considering. As I understand it, one needs both varied (gausian-like) transmittance (aka no flat tops) AND overlap to achieve “better?” differentiation between wavelengths in a spectral range. Whether this outcome is actually scientifically desirable (my goal), is still something idk. I care more about scientific meaning and interpretability in an image more than what it looks like, and it just so happens that space still looks stunning from that perspective without having to optimize for a glitterbomb image (okay, I’m exaggerating with this term here, but you get the point), and Frank here is really like to have your perspective on what made you choose the specific square filters that you have, and their advantages.

 

One thing to keep in mind is that even for the flat-top 100% transmission filters the signal captured across different wavelengths is not even because you have to factor in the camera sensor’s QE response curve (and transmission through the optics, but let’s for now pretend that we are using a polychromat scope with more or less even transmission across the NUV-VIS-NIR range). I will be attaching the transmission curves for the Chroma filters and the measured response curve for the IMX455-type sensor below.

 

Would love to hear y’all’s thoughts!

 

Shubham

 

spectra-viewer-snapshot.png

 

spectra-viewer-snapshot-2.png

 

IMG_8874.png


Edited by Shubham, 10 May 2025 - 01:13 AM.


#295 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 254
  • Joined: 02 Oct 2017

Posted 09 May 2025 - 08:42 PM

I'm glad you found the thread interesting! I was surprised at some of the very hostile negative reactions it stirred up, so it is nice to hear from someone who found it useful. As you now know, you need those overlapping filters if you want to capture spectrally pure colors accurately. That first filter set graph looks great!


Edited by loujost, 09 May 2025 - 08:42 PM.

  • SeymoreStars likes this

#296 SeymoreStars

SeymoreStars

    Aurora

  • *****
  • Posts: 4,712
  • Joined: 08 May 2014
  • Loc: Pennsyltucky

Posted 09 May 2025 - 10:14 PM

I found the thread useful also. Some people's minds like concrete are all mixed up and permanently set and they resist any attempt to challange their view point (to the detriment of all).



#297 Shubham

Shubham

    Vostok 1

  • *****
  • Posts: 159
  • Joined: 04 Jan 2023
  • Loc: DFW, TX

Posted 10 May 2025 - 10:53 AM

So I was curious what the effective QE would look like for the Classic Bessell UBVRI vs square cut-off UBVRI filters with IMX455 sensor. For an interesting comparison, I also plotted Hubble Space Telescope’s WFC3/UVIS1 detector throughputs.

 

Here are the results:

 

output Large.jpeg

 

output-2 Large.jpeg

 

output-3.png

 

Interesting indeed.

 

I see that most of Hubble's filters have rather square-ish cut-offs but the V band does seem a bit different. In many ways that Hubble maps these passbands to different channels––for example BVI to blue, green & red and Ha+[NII] added to red along with I and U either added as violet/purple or mapped to blue along with B (in which case F275W (UV) alone is mapped to violet/purple)––the V band does seem to create a gentler transition from blue B to green V (or blue V to green Wide V when V-WideV-I mapping is done). I'm not sure if this overlap was intentional or just a happenstance. The [OIII], Hb & HeII lines all seem to get mixed to pure green since the overlap is even further to the left.

 

I gotta say, the Classic Bessell UBVRI filters w/ overlapping seem much better choice than the square cut-off ones to me now. Greater diversity of colors, AND you can have spectral violet too (by capturing in U band and mapping to violate/purple) in a smooth transition from the blue channel. Even from a purely scientific perspective, these seem to be superior.

 

I'd still love to hear any advantages of the square cut-off filters, and any insight available into Hubble's filter arsenal from more knowledgable people around here. This stuff really excites me!

 

Shubham


  • kgb likes this

#298 BQ Octantis

BQ Octantis

    Voyager 1

  • *****
  • Posts: 10,623
  • Joined: 29 Apr 2017
  • Loc: Nova, USA

Posted 11 May 2025 - 05:49 AM

If you're going to make scientific observations, you probably want to differentiate individual wavelengths via spectroscopy. Lumping bands together doesn't allow you to differentiate emission lines that would help you determine proportions of constituents. If you're going to go with spectrographs, in selecting your bands I think you need to consider what it is they're including and excluding. Filters that overlap will indeed have cross-contamination from other emissions, which illuminates the value of square filters. They have no bleedover, so you know the level you measure is only the intended bands, and the grayscale spectrograph you create is only the intended science.

 

Esthetics is an entirely different matter--and indeed, color vectors captured in a dense color space (with overlap for every wavelength) will better fill the sRGB display space. The only problem is determining how to go from a 5-dimensional space to a 3-dimensional space...

 

BQ



#299 Shubham

Shubham

    Vostok 1

  • *****
  • Posts: 159
  • Joined: 04 Jan 2023
  • Loc: DFW, TX

Posted 11 May 2025 - 07:50 AM

If you're going to make scientific observations, you probably want to differentiate individual wavelengths via spectroscopy. Lumping bands together doesn't allow you to differentiate emission lines that would help you determine proportions of constituents. If you're going to go with spectrographs, in selecting your bands I think you need to consider what it is they're including and excluding. Filters that overlap will indeed have cross-contamination from other emissions, which illuminates the value of square filters. They have no bleedover, so you know the level you measure is only the intended bands, and the grayscale spectrograph you create is only the intended science.

 

Esthetics is an entirely different matter--and indeed, color vectors captured in a dense color space (with overlap for every wavelength) will better fill the sRGB display space. The only problem is determining how to go from a 5-dimensional space to a 3-dimensional space...

 

BQ

Although one of my goals is actually to dabble in some citizen science, my primary goal would still be imaging. I know this is cringe, but I like the idea of optimizing for "meaning" and "interpretability" in an image, as Frank says, rather than for clicks on Social media. I want to image to quench my curiosity (which does lie in NUV & NIR as well) in how I could do my part in capturing the universe in as faithful and meaningful way as I can.

 

So getting back to filters and color diversity. Earlier I was thinking of going with the square cutoff UBVRI filters and mapping U+B, V, R+I to blue, green and red respectively (let's just assume for now that the optics are optimized for the 320-1000nm range we are interested in here). The logic behind that was to increase the range of EM spectrum observed and bring out the electric blues in hot O-B stars and deep reds in relatively cooler K-M stars kinda like what Hubble does. That to me seems like an informative image, even though it might not contain the spectrally pure violet, cyan, yellow or orange colors. You know that when you look at the blue channel that's the information from 320-488nm, green shows 491-574nm, and everything from 575-1005nm is red. Obviously, the camera sensor's QE response (and the optics) do naturally cause a slope in effective QE of the system in U & I bands, which does help in somewhat achieving some distinction between the different regions of the spectrum by creating varied shades of blue or red. That to me, is a meaningful image.

 

After reading this thread, the idea of being able to capture the spectrally pure emissions in their true colors (the color that we "see" that wavelength in, not necessarily the object in space) did seem interesting. If I understand this correctly, the goal is to have HeI & NaI emissions appear yellow as our eye would see that wavelength and not red as they would appear if mapped to a square cutoff filter, and then likewise for other emission lines. Sounds great. But how do you expand this to outside of the visible range? You can even map U to violet/purple, but what about I? Let's say we limit ourselves to the visual spectrum. Looking at the Classic overlapping filters it seems like the HeI & NaI emissions still wont be yellow but rather orange. Wasn't our goal to see spectral yellow as yellow and not red? Is it okay if it is still not yellow but rather orange instead? I do want to learn more here cuz at this point I think that sure, color diversity would be great, but I still would not know what those colors represent because everything would be arbitrarily skewed to slightly different colors. So now I have an image that's more "colorful" but the colors have been somewhat contaminated.

 

Now as I'm reading another great thread "True color again" by loujost, I'm realising that honestly for visual spectrum there's nothing better than a modern mirrorless camera in faithfully capturing colors in a way that we see the world in. The moment we even move to an OSC "astro" camera, the lack of LPF filters, etc. make it shift away from how we see the world from our own eyes. Given that information, I think if one wants "True Color" they should go no further than a modern mirrorless camera (and I know even that is not 100% accurate "True color" but oh well). Now that we have that covered how do we choose which filters to go with for monochrome imaging? It brings me back again to the square cutoff UBVRI filters. At least there I know what the blue, green and red channels in the final image would represent.

 

This is where I am right now. I'm honestly more confused than I was before coming accross these insightful threads, but that is a good thing. All it means is that I know that I need to learn more. And therefore I ask you (and others) to chime in, I'd like to see what you guys think about this topic.


  • SeymoreStars and Brain&Force like this

#300 SeymoreStars

SeymoreStars

    Aurora

  • *****
  • Posts: 4,712
  • Joined: 08 May 2014
  • Loc: Pennsyltucky

Posted 11 May 2025 - 09:45 AM

These are my filter sets, I have no desire to change them. Buy once cry once.

 

 

Attached Thumbnails

  • Screenshot 2025-05-11 103938.jpg
  • Screenshot 2025-05-11 103957.jpg
  • Screenshot 2025-05-11 104026.jpg



CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.


Recent Topics






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics