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
292 replies to this topic

#51 TxStars

TxStars

    Gemini

  • *****
  • Posts: 3,338
  • Joined: 01 Oct 2005
  • Loc: Lost In Space

Posted 14 March 2023 - 01:25 AM

I guess each model of phone has some variations.

Here is what my iPhone SE shows with a cheap grating.

 

Attached Thumbnails

  • Spectra.jpg


#52 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 243
  • Joined: 02 Oct 2017

Posted 14 March 2023 - 02:03 PM

I guess each model of phone has some variations.

Here is what my iPhone SE shows with a cheap grating.

Interesting, nice yellows and blues/violets but poor greens. There might be engineering or cost tradeoffs that prevent perfection over the whole range.



#53 KBHornblower

KBHornblower

    Vanguard

  • -----
  • Posts: 2,302
  • Joined: 01 Jul 2020
  • Loc: Falls Church, VA (Washington DC suburb)

Posted 14 March 2023 - 08:32 PM

That looks like the blending in a rainbow, from a source that is not a narrow line of light.  What sort of light source is this?



#54 badgie

badgie

    Viking 1

  • *****
  • Posts: 637
  • Joined: 13 Jan 2021
  • Loc: SF Area, California

Posted 14 March 2023 - 10:52 PM

Lou 100% on board with your points.   I may rethink my choice of square shouldered filters going forward.  There is also a similar inverse problem; you can't illuminate a scene with narrowband light and get a good color rendering (CRI) Hence the need for a CRI for lighting. 



#55 TxStars

TxStars

    Gemini

  • *****
  • Posts: 3,338
  • Joined: 01 Oct 2005
  • Loc: Lost In Space

Posted 15 March 2023 - 01:08 AM

The light source was a LED flashlight.

That image was done in a hotel room..

I used a trash plastic prism with the light held between my fingers.

Will reshoot with a proper prism and razor slit when I get home at the end of the month.



#56 columbidae

columbidae

    Ranger 4

  • -----
  • Posts: 380
  • Joined: 08 Sep 2022
  • Loc: Texas

Posted 15 March 2023 - 01:20 AM

Keep in mind that an LED isn't going to give a flat spectrum - an incandescent or the sun would be a better choice.
  • jdupton and sharkmelley like this

#57 KBHornblower

KBHornblower

    Vanguard

  • -----
  • Posts: 2,302
  • Joined: 01 Jul 2020
  • Loc: Falls Church, VA (Washington DC suburb)

Posted 15 March 2023 - 11:13 AM

Here are some new spectrum images I made just now.  I created a narrow vertical slit by leaning some scrap lumber in front of a window.  The light source is the side of a white house across the street in full sunshine.

 

SpectrumC1 iPhone.jpg

iPhone

 

 

SpectrumC1 fuji.JPG

Old Fuji camera

 

In this torture test I could not get the old camera much brighter without a lot of burning out.  The iPhone appears to be much more advanced, not surprising after 17 years.  Nevertheless the old camera still makes excellent photos of ordinary scenes and faces.



#58 KBHornblower

KBHornblower

    Vanguard

  • -----
  • Posts: 2,302
  • Joined: 01 Jul 2020
  • Loc: Falls Church, VA (Washington DC suburb)

Posted 16 March 2023 - 07:19 PM

Here is a new one with the Fuji camera in a room that is fully blacked out except for a slit in a cover over a south-facing window.  The noon Sun is shining on Scotch Magic Tape that is covering the slit.  The camera is on a tripod, with an exposure of about 5 seconds at f/4, ISO 400.

 

Spectrum F Fuji.jpg

 

In this one there appears to be little or no overlap between the red and the green, but enough tapered overlap between the green and the blue to give some gradation of aqua.  The light and dark bands within the green must be characteristics of the filter.  I identified the Fraunhofer lines from this image.

 

https://storage.goog...oferIII-02a.jpg

 

This shows where the C line (H alpha) would be if the filter did not cut it off.  Once again, the D lines would be rendered too red, while the F would be fairly close.

 

I tried the same thing with the iPhone, but it overexposed and blocked up everything.  I don't know how it got a better exposure in yesterday's image.  I have not yet learned how to adjust its exposure.


  • badgie likes this

#59 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 26,034
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 17 March 2023 - 08:45 AM

It is very important to consider the light source when doing these kinds of tests. LED or even some CFL, should be considered poor quality sources for such tests, as they don't emit broad spectrum light, or if they do, much of the spectrum emits at very low values while others are emitted at high values.

 

Sunlight is a good source, as it is decidedly broad spectrum, and has a very well known spectral power distribution. 


  • jdupton and sharkmelley like this

#60 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 243
  • Joined: 02 Oct 2017

Posted 17 March 2023 - 09:05 AM

It is very important to consider the light source when doing these kinds of tests. LED or even some CFL, should be considered poor quality sources for such tests, as they don't emit broad spectrum light, or if they do, much of the spectrum emits at very low values while others are emitted at high values.

 

Sunlight is a good source, as it is decidedly broad spectrum, and has a very well known spectral power distribution. 

Also, sunlight has those wonderful Fraunhofer lines, so it allows us to assign exact wavelengths to each part of the spectrum.


  • jdupton, sharkmelley and Jon Rista like this

#61 xploeris

xploeris

    Mariner 2

  • -----
  • Posts: 229
  • Joined: 18 Dec 2021

Posted 10 June 2023 - 10:39 PM

If you had a set of truly non-overlapping RGB filters, then loujost's claim would be correct: it would be impossible for any photon to pass more than one filter, so all light would be forced to red, green, or blue, and it would be impossible to produce yellow. (Any light that registers as >0 with the red filter MUST have green=0, and vice versa; there is NO way to mix red and green to produce yellow as one component will always be 0.) And, unless the cutoffs are perfectly "vertical", going from full transmission to none, then you'll lose a lot of the reddish-kinda-almost-yellowish light as well.

However, not only do those RGB filters not have perfectly sharp cutoffs in the real world, they actually do overlap a bit. loujost, you can see this for yourself: open that RGB filtered spectrum in Photoshop, or your image editor of choice, and crank up the curve like you were stretching a DSO. Hey, what's that in between red and green? It looks like a pale, ugly, quantized band of yellow! Which of course it is, because down at the bottom of the shoulders of your filters' frequency bands, there is a little bit of overlap where yellow photons can get through both filters. The yellow you're getting there is highly attenuated, but it's not 0, and after enough stretching it's hard to tell that it's even attenuated. (If you have a raw version of this photo, you can probably get cleaner colors out of it than I can out of this lossy-compressed 8-bit image.)

Someone asked how RGB filters can give them yellow galaxies if yellow is being lost, and there's two answers. One is that there's still a little pure yellow left in your data. The other is that blackbody radiation is multispectral, so the light from a star (and galaxies are mostly star-colored) can get through both filters. It might be more accurate to say that your galaxies are reddish green, not yellow! - but of course human vision can't tell the difference.



#62 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 243
  • Joined: 02 Oct 2017

Posted 10 June 2023 - 10:59 PM

If you had a set of truly non-overlapping RGB filters, then loujost's claim would be correct: it would be impossible for any photon to pass more than one filter, so all light would be forced to red, green, or blue, and it would be impossible to produce yellow. (Any light that registers as >0 with the red filter MUST have green=0, and vice versa; there is NO way to mix red and green to produce yellow as one component will always be 0.) And, unless the cutoffs are perfectly "vertical", going from full transmission to none, then you'll lose a lot of the reddish-kinda-almost-yellowish light as well.

However, not only do those RGB filters not have perfectly sharp cutoffs in the real world, they actually do overlap a bit. loujost, you can see this for yourself: open that RGB filtered spectrum in Photoshop, or your image editor of choice, and crank up the curve like you were stretching a DSO. Hey, what's that in between red and green? It looks like a pale, ugly, quantized band of yellow! Which of course it is, because down at the bottom of the shoulders of your filters' frequency bands, there is a little bit of overlap where yellow photons can get through both filters. The yellow you're getting there is highly attenuated, but it's not 0, and after enough stretching it's hard to tell that it's even attenuated. (If you have a raw version of this photo, you can probably get cleaner colors out of it than I can out of this lossy-compressed 8-bit image.)

Someone asked how RGB filters can give them yellow galaxies if yellow is being lost, and there's two answers. One is that there's still a little pure yellow left in your data. The other is that blackbody radiation is multispectral, so the light from a star (and galaxies are mostly star-colored) can get through both filters. It might be more accurate to say that your galaxies are reddish green, not yellow! - but of course human vision can't tell the difference.

 

There is some deliberate overlap of green and blue in many astro RBG filter sets, to capture the teal of oxygen. But the red filter does not overlap at all with the others. Neither spectrally pure yellow nor magenta can be captured by these filters. And all the different red emission lines get mapped into the same shade of red. Most of the green lines are mapped into a singe shade of green, and most of the blue and violet lines are mapped into a singe shade of blue. Color diversity is being lost. The tiny leaks you refer to are so dim as to be virtually undetectable without blowing out all the other colors in an image.

 

Of course, as I mentioned above, the colors that are "simplified" in this way are mostly the colors made by emission lines. Broadband light sources can be pretty accurately imaged with non-overlapping filters because they capture the right relative amounts of red, blue, and green.



#63 xploeris

xploeris

    Mariner 2

  • -----
  • Posts: 229
  • Joined: 18 Dec 2021

Posted 11 June 2023 - 05:25 AM

But the red filter does not overlap at all with the others. Neither spectrally pure yellow nor magenta can be captured by these filters. And all the different red emission lines get mapped into the same shade of red. Most of the green lines are mapped into a singe shade of green, and most of the blue and violet lines are mapped into a singe shade of blue. Color diversity is being lost. The tiny leaks you refer to are so dim as to be virtually undetectable without blowing out all the other colors in an image.

And yet the yellow is there. You don't have to take my word for it. You have the image. Try the experiment yourself. Actually, I can see the yellow in your filtered image, now that I know what to look for. It is quite subtle. I think there's more overlap than you realize.

As for the tiny leaks being dim - yes, but nonlinear stretching (which you would use on practically any DSO, except maybe the Milky Way) will pull those dim yellows out of the shadows. Especially if you're doing something in processing to enhance the color...



#64 Robert7980

Robert7980

    Soyuz

  • -----
  • Posts: 3,556
  • Joined: 20 Nov 2022
  • Loc: Western North Carolina

Posted 11 June 2023 - 06:32 AM

If RGB weren’t enough then there’s been no pictures in the digital era that are accurate, this is something that I’ve not heard outside this context, so it seems to be rather overstated… The example has no yellow, this is probably by design to block sodium street lighting and done on purpose. I would think just from a software perspective you’d ideally want absolutely zero overlap with infinitely steep pass-bands that when added are continuous.

 

Anything that overlaps will surely result in an odd color because the software has no way to know the color is at the edge of the band, it just knows how to mix relative luminosity contributions of RGB. Oddly enough nobody has ever seen colors beyond RGB just due to the quantum-electrodynamic physics of the eye… So to use overlapping bands you’d need software that either doesn’t exist or is very specialized… 


Edited by Robert7980, 11 June 2023 - 06:33 AM.


#65 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 243
  • Joined: 02 Oct 2017

Posted 11 June 2023 - 09:48 AM

If RGB weren’t enough then there’s been no pictures in the digital era that are accurate, this is something that I’ve not heard outside this context, so it seems to be rather overstated… The example has no yellow, this is probably by design to block sodium street lighting and done on purpose. I would think just from a software perspective you’d ideally want absolutely zero overlap with infinitely steep pass-bands that when added are continuous.

 

Anything that overlaps will surely result in an odd color because the software has no way to know the color is at the edge of the band, it just knows how to mix relative luminosity contributions of RGB. Oddly enough nobody has ever seen colors beyond RGB just due to the quantum-electrodynamic physics of the eye… So to use overlapping bands you’d need software that either doesn’t exist or is very specialized… 

Robert, you are mixing up how to display colors (RGB does fine) and how to capture colors (this is where the problem lies; and you can't display what is not captured). Ordinary digital photography usually works well because (1) most light is not spectrally pure and (2) the Bayer filter array more or less matches the sensitivity curves of our cones (by design).

 

You recognized that non-overlapping filters reduce color diversity when you suggested that RGB filters eliminate spectrally pure yellow on purpose. Whether on purpose or  not, you recognize if there is no overlap between RGB filters, you can never capture spectrally pure yellow.

 

The fact that Bayer filters, with their strongly overlapping colors, do work pretty well shows that there is nothing inherently wrong or impractical about using overlapping filters. 


  • jdupton and Jon Rista like this

#66 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 26,034
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 11 June 2023 - 12:52 PM

If you had a set of truly non-overlapping RGB filters, then loujost's claim would be correct: it would be impossible for any photon to pass more than one filter, so all light would be forced to red, green, or blue, and it would be impossible to produce yellow. (Any light that registers as >0 with the red filter MUST have green=0, and vice versa; there is NO way to mix red and green to produce yellow as one component will always be 0.) And, unless the cutoffs are perfectly "vertical", going from full transmission to none, then you'll lose a lot of the reddish-kinda-almost-yellowish light as well.

However, not only do those RGB filters not have perfectly sharp cutoffs in the real world, they actually do overlap a bit. loujost, you can see this for yourself: open that RGB filtered spectrum in Photoshop, or your image editor of choice, and crank up the curve like you were stretching a DSO. Hey, what's that in between red and green? It looks like a pale, ugly, quantized band of yellow! Which of course it is, because down at the bottom of the shoulders of your filters' frequency bands, there is a little bit of overlap where yellow photons can get through both filters. The yellow you're getting there is highly attenuated, but it's not 0, and after enough stretching it's hard to tell that it's even attenuated. (If you have a raw version of this photo, you can probably get cleaner colors out of it than I can out of this lossy-compressed 8-bit image.)

Someone asked how RGB filters can give them yellow galaxies if yellow is being lost, and there's two answers. One is that there's still a little pure yellow left in your data. The other is that blackbody radiation is multispectral, so the light from a star (and galaxies are mostly star-colored) can get through both filters. It might be more accurate to say that your galaxies are reddish green, not yellow! - but of course human vision can't tell the difference.

It depends entirely on the filter set. Many LRGB filter sets have a little bit of overlap between blue and green, while at the same time having a purposeful gap between green and red to eliminate LP (not that that is of any real use anymore these days, given the prevalence of LED-born LP.)

 

If you have an LP gap in your LRGB filters, you won't be capturing the yellows (or really much of orange, or the yellow-greens). If you have filters that meet but do not overlap, then you might get some oranges, but capturing pure yellow is going to be a problem. Yellow...a particularly common star color! 

 

You can reproduce yellowish colors...but they are usually very washed out and weak. So much so, that the blues in galaxies often completely dominate, and that's the color I think most astrophotographers come to expect in galaxy images these days: BLUE!! Blue giants, dominating all other colors in the galaxy, except maybe the orangish cores. I strongly believe that this is actually not an accurate depiction of galaxies...I think we capture very weak orange to yellow/white color, because of the nature of the filters we use. I think that overall, galaxies should generally be more white to slightly yellowish, with warmer (more orange) cores,  with SPECKLES of blue in the outer regions where blue giant clusters form (which also tend to be where you find nebula). I don't think that the bluer color of the blue giant clusters should dominate the color of galaxies like it generally does in most images these days. Further, if you really account for the particularly dominant form of stars in most galaxies, which as we now know is not even the white to yellow/orange F-, G- and K-type main sequence stars, but the red and brown dwarfs, which generally seem to outnumber all the other star types combined (we used to think the Milky Way had 125-200 billion stars...modern estimates are now over 400 billion, with the additional 200+ billion being red and brown dwarf stars, many of which I'm pleased to say are discovered by amateur imagers such as ourselves! ;)) When accounting for the aggregate light from these smaller stars, which emit a lot of IR and reds, the underlying color of galaxies is decidedly "warmer" on the color temperature scale than most frequently depicted.



#67 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 243
  • Joined: 02 Oct 2017

Posted 11 June 2023 - 01:43 PM

"If you have an LP gap in your LRGB filters, you won't be capturing the yellows (or really much of orange, or the yellow-greens). If you have filters that meet but do not overlap, then you might get some oranges, but capturing pure yellow is going to be a problem. Yellow...a particularly common star color!"

 

Jon, there is an important distinction between spectrally pure colors (like emission lines) and black-body radiators (like stars). For spectrally pure colors it doesn't matter if there is a gap or not.  Spectrally pure yellow will be absent in the final image in either case. It will either be recorded entirely as red or green  (if filter curves meet), or it will not be recorded at all, but in no case will you get orange or yellow or orange. See the photo that opens this thread.

 

On the other hand, for continuous-spectrum black-body radiators, the ratio of red/green/blue will still be more-or-less captured correctly by non-overlapping filter sets, though maybe if there is a gap, there might be inaccuracies.  I'm not sure how inaccurate the result will be for stars.



#68 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 26,034
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 11 June 2023 - 03:52 PM

"If you have an LP gap in your LRGB filters, you won't be capturing the yellows (or really much of orange, or the yellow-greens). If you have filters that meet but do not overlap, then you might get some oranges, but capturing pure yellow is going to be a problem. Yellow...a particularly common star color!"

 

Jon, there is an important distinction between spectrally pure colors (like emission lines) and black-body radiators (like stars). For spectrally pure colors it doesn't matter if there is a gap or not.  Spectrally pure yellow will be absent in the final image in either case. It will either be recorded entirely as red or green  (if filter curves meet), or it will not be recorded at all, but in no case will you get orange or yellow or orange. See the photo that opens this thread.

 

On the other hand, for continuous-spectrum black-body radiators, the ratio of red/green/blue will still be more-or-less captured correctly by non-overlapping filter sets, though maybe if there is a gap, there might be inaccuracies.  I'm not sure how inaccurate the result will be for stars.

I'm talking about inaccuracies. Black body radiators emit across the spectrum...so, if there is a gap in the R and G filters, then there is going to be a part of the spectrum, black body emitters included, that you simply do not capture. You would certainly capture some photons, but lacking any in that gap, in my experience those colors end up inaccurate indeed. I think this is a key reason why so many galaxy images, or globs, or other images of primarily stars, lack signal strength in that region (the yellows, oranges). Contrast with OSC images of primarily stars, and there are often notable color differences. You can go from a soft yellowish with mono+lrgb, to stronger orange with OSC. This is often quite evident in milky way images. It is also evident in many of the more prominent galaxy images, where the "warmer" end of the spectrum is very weak, washed out. 

 

Now, my evidence is certainly anecdotal, but it could be corroborated by more explicit testing. Between mono + rgb, mono + type-2c rgb, and OSC. Boy, it would be nice if we had an RGB filter set that overlapped similar to the cones of the human eye... I don't think such a filter set exists. Wonder if Chroma or Astronomik could be convinced to create a set though... Anyway, I think the loss of parts of the spectrum (anywhere, really, not necessarily just between greens and reds) and the consequences thereof, could be demonstrated, if we had the right data. 



#69 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 243
  • Joined: 02 Oct 2017

Posted 11 June 2023 - 03:59 PM

"Boy, it would be nice if we had an RGB filter set that overlapped similar to the cones of the human eye... I don't think such a filter set exists."

 

Yes, that's exactly what we need. I've been using my spectroscope to examine different filters that I have collected over the years, and most of them don't have the broad peak and smooth fall-off like our cones have. The closest we have are the Bayer filter arrays on our non-astro cameras.

 

Do cooled OSC astro cameras use the same Bayer filters as ordinary cameras?



#70 badgie

badgie

    Viking 1

  • *****
  • Posts: 637
  • Joined: 13 Jan 2021
  • Loc: SF Area, California

Posted 12 June 2023 - 08:06 AM

I think nearly all use the off the shelf CFAs as normal cameras.

Jon, why not do some simulation using Lou's data if he'll share and blackbody spectra? Also here are a few example whole galaxy spectra, however these dont illuminate you core/cluster distinction. http://astronomy.nms...26/slide01.html

A quick look didn't find a useful imaging spectrometry paper not focused on velocity distributions but I bet something is out there!

#71 Oort Cloud

Oort Cloud

    Fly Me to the Moon

  • -----
  • Posts: 5,430
  • Joined: 19 Nov 2020
  • Loc: New Jersey, USA

Posted 12 June 2023 - 09:18 AM

Boy, it would be nice if we had an RGB filter set that overlapped similar to the cones of the human eye... I don't think such a filter set exists.


Astronomik actually already makes a set:

https://www.astronom...filtersatz.html

#72 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 243
  • Joined: 02 Oct 2017

Posted 12 June 2023 - 09:30 AM

Astronomik actually already makes a set:

https://www.astronom...filtersatz.html

No they don't. Those are better than non-overlapping ones for maintaining color diversity, but they are not designed to mimic cone response. In particular there is no red-blue overlap, so spectrally pure violet is lost, and the peaks of green and red transmission are too far apart. 


Edited by loujost, 12 June 2023 - 09:31 AM.


#73 Oort Cloud

Oort Cloud

    Fly Me to the Moon

  • -----
  • Posts: 5,430
  • Joined: 19 Nov 2020
  • Loc: New Jersey, USA

Posted 12 June 2023 - 10:04 AM

No they don't. Those are better than non-overlapping ones for maintaining color diversity, but they are not designed to mimic cone response. In particular there is no red-blue overlap, so spectrally pure violet is lost, and the peaks of green and red transmission are too far apart.


It's right there in the description: "The separation corresponds to the colour sensitivity of the human eye according to DIN 5032".

#74 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 26,034
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 12 June 2023 - 12:04 PM

Astronomik actually already makes a set:

https://www.astronom...filtersatz.html

So this is the Type-2c filters. They do overlap, but, they don't correspond to the response of the human eye. Red and green in natural human sensitivity overlap a lot more than the Type-2c filter set. The Type-2c is certainly better, and will be my primary filter set for RGB imaging (and I'll be dropping L filters at my dark site), but, I don't think it would quite reproduce how the human eye might actually see things. As Lou mentioned, where is no red-blue overlap, as there is with human sight, which would limit proper reproduction of certain violets and magenta. 


Edited by Jon Rista, 12 June 2023 - 12:12 PM.


#75 italic

italic

    Apollo

  • *****
  • Posts: 1,067
  • Joined: 05 Apr 2022
  • Loc: San Francisco, CA

Posted 12 June 2023 - 02:12 PM

What about these UBVRI photometric filters from Baader? They overlap quite a bit between red/green, but not as much between green/blue. Hb is the only line that seems suppressed. Need violet? Mix in the ultraviolet filter, although I guess the mix may be up to one's taste. Would you end up mixing those by hand or could you make a calibrated mixing ratio based on transmission losses? I would assume you wouldn't want to use an L filter for this kind of filter set.




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