292 replies to this topic

[loujost]

Judging from some previous threads on CN, some astrophotographers think that capturing RGB information is sufficient to reproduce all colors in the scene, because (almost) any color can be represented as an RGB triplet. After all, that's how monitors generate the colors we see in our photos.

 

While this is (mostly) true on the display side pf things, it is not true on the capture side. If you only capture RGB with typical astro filters that do not overlap, you will be unable to capture all the colors present in the scene, if some of those colors are spectrally pure.

 

People have talked about this elsewhere on CN (with some pushback) but I have not seen a photographic demonstration. So here one is. Here is a spectrum photographed by a stock digital camera, whose RGB filters overlap (thus mimicking the overlapping color sensitivities of the three kinds of human cones).

 

And here is the same spectrum photographed with non-overlapping astro RGB filters on a monochrome camera.

 

The loss of color diversity is striking. Now in astro subjects, maybe some of these colors are rare, but not all of them.

 

I am not arguing about color accuracy here, but rather about an aesthetic issue.

Edited by loujost, 12 March 2023 - 04:38 PM.

[hyiger]

Judging from some previous threads on CN, some astrophotographers think that capturing RGB information is sufficient to reproduce all colors in the scene, because (almost) any color can be represented as an RGB triplet. After all, that's how monitors generate the colors we see in our photos.

 

While this is (mostly) true on the display side pf things, it is not true on the capture side. If you only capture RGB with typical astro filters that do not overlap, you will be unable to capture all the colors present in the scene, if some of those colors are spectrally pure.

 

People have talked about this elsewhere on CN (with some pushback) but I have not seen a photographic demonstration. So here one is. Here is a spectrum photographed by a stock digital camera, whose RGB filters overlap (thus mimicking the overlapping color sensitivities of the three kinds of human cones).

PEN F sm.jpg

 

And here is the same spectrum photographed with non-overlapping astro RGB filters on a monochrome camera.

Final RGB no overlap sm.jpg

 

The loss of color diversity is striking. Now in astro subjects, maybe some of these colors are rare, but not all of them.

 

I am not arguing about color accuracy here, but rather about an aesthetic issue.

My guess is that in the case of the color camera with a Bayer matrix (which has RGGB or GRBG etc filters) that this "overlap" occurs due how the debayer process is interpolating adjacent pixels? Also, when broadband imaging I'm not really concerned so much about accurate color as I am with SNR. On more colorful subjects like emission nebula I'm using narrow band filters assigned to the RGB channels so I'm getting false color to begin with.  

Edited by hyiger, 12 March 2023 - 05:00 PM.

[columbidae]

Judging from some previous threads on CN, some astrophotographers think that capturing RGB information is sufficient to reproduce all colors in the scene, because (almost) any color can be represented as an RGB triplet. After all, that's how monitors generate the colors we see in our photos.

 

While this is (mostly) true on the display side pf things, it is not true on the capture side. If you only capture RGB with typical astro filters that do not overlap, you will be unable to capture all the colors present in the scene, if some of those colors are spectrally pure.

 

People have talked about this elsewhere on CN (with some pushback) but I have not seen a photographic demonstration. So here one is. Here is a spectrum photographed by a stock digital camera, whose RGB filters overlap (thus mimicking the overlapping color sensitivities of the three kinds of human cones).

PEN F sm.jpg

 

And here is the same spectrum photographed with non-overlapping astro RGB filters on a monochrome camera.

Final RGB no overlap sm.jpg

 

The loss of color diversity is striking. Now in astro subjects, maybe some of these colors are rare, but not all of them.

 

I am not arguing about color accuracy here, but rather about an aesthetic issue.

Does the spectrum in the first photograph look like that to your eyes as displayed?  I ask because there's a rainbow outside right now has a lot more yellow and violet in comparison. (or perhaps it's apples and oranges)

[TxStars]

I have not seen many images that show what the eye can see.

With mono camera you would have to use 8-10 filters to come close.

And with a OSC you get a poor representation of what you see.

Attached Thumbnails

• 

Edited by TxStars, 12 March 2023 - 06:32 PM.

[loujost]

My guess is that in the case of the color camera with a Bayer matrix (which has RGGB or GRBG etc filters) that this "overlap" occurs due how the debayer process is interpolating adjacent pixels? Also, when broadband imaging I'm not really concerned so much about accurate color as I am with SNR. On more colorful subjects like emission nebula I'm using narrow band filters assigned to the RGB channels so I'm getting false color to begin with.  

Maybe that is part of it, but I think the Bayer filters have to respond more like eye cones; after all, they are trying to mimic eye response to colors.

 

If you are only using two or three narrowband filters per image, and your subject contains multiple spectral colors (more than three), then your images are suffering from the problem this demonstrates. You are assigning multiple spectral colors to the same display color.

 

I should emphasize that this issue is not exactly about accurate color (though that is part of it); in my humble opinion the loss of color complexity is detrimental to an image.

Edited by loujost, 12 March 2023 - 06:39 PM.

[loujost]

Does the spectrum in the first photograph look like that to your eyes as displayed?  I ask because there's a rainbow outside right now has a lot more yellow and violet in comparison. (or perhaps it's apples and oranges)

No. My eye shows a much brighter yellow area, and a somewhat more distinct violet area. I believe this is a limitation of today's display devices. So in fact the loss of color diversity when using non-overlapping RGB filters is even greater than what I show here.

Edited by loujost, 12 March 2023 - 06:35 PM.

[imtl]

Maybe that is part of it, but I think the Bayer filters have to respond more like eye cones; after all, they are trying to mimic eye response to colors.

If you are only using two or three narrowband filters per image, and your subject contains multiple spectral colors (more than three), then your images are suffering from the problem this demonstrates. You are assigning multiple spectral colors to the same display color.

''Trying to mimic'' does not mean they are successful.

[columbidae]

No. My eye shows a much brighter yellow area, and a somewhat more distinct violet area. I believe this is a limitation of today's display devices. So in fact the loss of color diversity when using non-overlapping RGB filters is even greater than what I show here.

As soon as I asked, I also started to wonder if it might have been my display as well. Sure isn't as simple as I wish it was.

[loujost]

''Trying to mimic'' does not mean they are successful.

Right, but here I'm not talking about accuracy, just color complexity/diversity.

[James Peirce]

It seems as though you are applying different criteria to the color camera data and mono camera data? Your OSC spectrum is accounting for signal captured across "filters" (i.e. on the color filter array) whereas your mono data representation is not accounting at all for signal captured across filters.

Edited by James Peirce, 12 March 2023 - 06:47 PM.

[loujost]

It seems as though you are applying different criteria to the color camera data and mono camera data? Your OSC spectrum is accounting for signal captured across "filters" (i.e. on the color filter array) whereas your mono data is not accounting at all for signal captured across filters.

That's right, but it's not my fault, the difference is unavoidable if the mono data is produced with three (or fewer) non-overlapping filters. That's my point.

Edited by loujost, 12 March 2023 - 06:51 PM.

[loujost]

I have not seen many images that show what the eye can see.

With mono camera you would have to use 8-10 filters to come close.

And with a OSC you get a poor representation of what you see.

Yes, you can avoid much of the loss of color diversity by using many narrowband filters, and assigning each one to a unique color (not straight R, G, or B).

Edited by loujost, 12 March 2023 - 06:54 PM.

[James Peirce]

That's right, but it's not my fault, the difference is unavoidable if the mono data is produced with non-overlapping filters. That's my point.

To be clear, "yellow" in the CFA is "yellow" because it is captured across multiple color filters (whether RGB on mono or the RGB cells on a CFA). We still capture "yellow" with non-overlapping mono filters. Your data representation are applying disparate criteria. You are also going to lose very little of "yellow" with non-overlapping filters provided coverage is still relatively comprehensive. And even if someone is using one of those odd filters that, more than lacking overlap, features actual material gaps in bandpasses (I assume light pollution mitigation?) you are still going to get representations of affected colors (albeit diminished) because those colors are the product of photons from a range of light sources.

I just don't understand what the visualization is supposed to represent because it is representing blended colors in the CFA example and isolated colors in the mono example. Which is not an equal way to represent color reproduction across the spectrum.

There may also be something of note to be said for overlap in stock color camera filters (as well as the green bias, which is another concern which is accounted for in those bandpass ranges) both having association with representing light spectrum as we perceive it (playing to the biases of the human eye) as opposed to balancing spectrum that is actually delivered in terms of photons.

Edited by James Peirce, 12 March 2023 - 07:05 PM.

[loujost]

To be clear, "yellow" in the CFA is "yellow" because it is captured across multiple color filters (whether RGB on mono or the RGB cells on a CFA). We still capture "yellow" with non-overlapping mono filters. Your data representation are applying disparate criteria. You are also going to lose very little of "yellow" with non-overlapping filters provided coverage is still relatively comprehensive. And even if someone is using one of those odd filters that, more than lacking overlap, features actual material gaps in bandpasses (I assume light pollution mitigation?) you are still going to get representations of affected colors (albeit diminished) because those colors are the product of photons from a range of light sources.

I just don't understand what the visualization is supposed to represent because it is representing blended colors in the CFA example and isolated colors in the mono example. Which is not an equal way to represent color reproduction across the spectrum.

There may also be something of note to be said for overlap in stock color camera filters (as well as the green bias, which is another concern which is accounted for in those bandpass ranges) both having association with representing light spectrum as we perceive it (playing to the biases of the human eye) as opposed to balancing spectrum that is actually delivered in terms of photons.

Yes, you can capture yellow in both set-ups, but in the case of non-overlapping RGB filters on a mono camera, you will map that yellow into either the red or green channel, not both. So when you combine the channels, you cannot ever get yellow, you will get only red or green.

 

Both images were taken of exactly the same subject.

Edited by loujost, 12 March 2023 - 07:49 PM.

[James Peirce]

Yes, you can capture yellow in both set-ups, but in the case of non-overlapping RGB filters on a mono camera, you will map that yellow into either the red or green channel, not both. So when you combine the channels, you cannot ever get yellow, you will get only red or green.

 

Both images were taken of exactly the same subject.

You're overlooking something important. For the sake of simplicity let's say we're imaging long enough that representative photons from our target have reached our sensor to be collected or not. Setting aside that our 'yellow' is going to be a product of photons emitted across a range of spectrum near and far from 600nm or so, say we are discussing photons reaching us at specific spectrum that falls outside the range of our red filter but inside the range of our nearby green filter. Those, given enough, are going to be captured representatively on the CFA array because the various filtered RGB pixels with their varied sensitivity are going to pick it up. We won't be picking up those photons when we have our red filter on. But when we put our green filter on, we will be picking up those photons. For the sake of argument, say we have contiguous coverage across this spectrum range via RGB mono filters and via the CFA filter. Because our RGB mono image is a product of combining those channels we still get our contiguous coverage, even though we were not picking up those specific photons when we had the red filter on.

 

If we imagine a CFA that is designed to pick up the visual spectrum with no bias (which is not how they are designed, but let's assume) that photon is going to be captured across green and red cells with a distribution of success (say, green at 35% and red at 65%, not accounting for other causes for signal loss). On the green filter we are capturing it at 0% and the red filter at 100%, still not accounting for a range of other factors). We still end up with equivalent representation once we combine the signal time captured across those RGB filters 1:1:1. (On the CFA camera this might look more like 50% on green with twice the sensor surface area and 85% on red as it falls in a range where signal is biased).

 

We didn't 'lose' that signal. It isn't missing. Because we are combining the signal captured across those filters for the final result. Much as we are combining signal captured from across the cells on a CFA array. We end up with deviation where the CFA is biased in favor of stronger signal (which it is by design across a range of spectrum) or where there is actually a gap in signal on the mono filters. But where the mono color filters are designed to be, collectively, contiguous, we are actually getting a good representation of color across said spectrum.

[whwang]

The above few replies talked about yellow.  Let's use yellow as an example.

 

An important issue here is whether it is:

A) monochromatic yellow (for example, a single wavelength 580 nm light), or

B) a continuum yellow (a spectrum distributed from 560 to 610 nm), or even

C) a double monochromatic light (for example, a single wavelength 560 nm line plus a single wavelength 610 nm line).

(I believe you get the point now.  You can imagine many more combinations of green and red lights that lead to "yellow.")

 

To our human eyes, all three will look yellow.  The "yellow" you see from the spectrum you show is case-A yellow.  It will be rendered yellow in your DSLR case, and either red or green in your non-overlaping filter case.  However, both B) and C) will be rendered yellow by non-overlaping filters as long as you don't further separate them with a spectrograph.

 

In real astronomical images, most of the subtle color come from continuum objects.  They are not monochromatic.  On such objects (like cases B and C), even non-overlaping filters can render their color variations.  They won't look purely red, green, or blue.  A potential problem is OIII.  It is cyanish and it is almost monochromatic.  If one uses truly non-overlaping filters, it will be either green or blue and never cyan.  Fortunately most astronomical filters purposely place an overlap between G and B around the wavelength of OIII.  So this is not really a problem.

[TxStars]

Mono imaging = fun with filters.

 

Attached Thumbnails

• 

[loujost]

You're overlooking something important. For the sake of simplicity let's say we're imaging long enough that representative photons from our target have reached our sensor to be collected or not. Setting aside that our 'yellow' is going to be a product of photons emitted across a range of spectrum near and far from 600nm or so, say we are discussing photons reaching us at specific spectrum that falls outside the range of our red filter but inside the range of our nearby green filter. Those, given enough, are going to be captured representatively on the CFA array because the various filtered RGB pixels with their varied sensitivity are going to pick it up. We won't be picking up those photons when we have our red filter on. But when we put our green filter on, we will be picking up those photons. For the sake of argument, say we have contiguous coverage across this spectrum range via RGB mono filters and via the CFA filter. Because our RGB mono image is a product of combining those channels we still get our contiguous coverage, even though we were not picking up those specific photons when we had the red filter on.

 

If we imagine a CFA that is designed to pick up the visual spectrum with no bias (which is not how they are designed, but let's assume) that photon is going to be captured across green and red cells with a distribution of success (say, green at 35% and red at 65%, not accounting for other causes for signal loss). On the green filter we are capturing it at 0% and the red filter at 100%, still not accounting for a range of other factors). We still end up with equivalent representation once we combine the signal time captured across those RGB filters 1:1:1. (On the CFA camera this might look more like 50% on green with twice the sensor surface area and 85% on red as it falls in a range where signal is biased).

 

We didn't 'lose' that signal. It isn't missing. Because we are combining the signal captured across those filters for the final result. Much as we are combining signal captured from across the cells on a CFA array. We end up with deviation where the CFA is biased in favor of stronger signal (which it is by design across a range of spectrum) or where there is actually a gap in signal on the mono filters. But where the mono color filters are designed to be, collectively, contiguous, we are actually getting a good representation of color across said spectrum.

As I said above, yes, you are capturing the yellow photon with the RGB setup. but it is always mapped to either red or green. If we are talking about spectrally pure yellow, no pixel will have both R and G values. More generally, for spectrally pure colors, every pixel will be either (R, 0.,0) ), (0, G. 0), or (0, 0, B). .That's what happens when spectrally pure colors are imaged by three non-overlapping R, G, and B filters on a mono camera.. And you can see this result in my image. So no yellow will appear in the final image. However if the colors are not spectrally pure colors, or if they are mixtures of spectrally pure colors, then the right colors might be produced.

Edited by loujost, 12 March 2023 - 09:12 PM.

[loujost]

The above few replies talked about yellow.  Let's use yellow as an example.

 

An important issue here is whether it is:

A) monochromatic yellow (for example, a single wavelength 580 nm light), or

B) a continuum yellow (a spectrum distributed from 560 to 610 nm), or even

C) a double monochromatic light (for example, a single wavelength 560 nm line plus a single wavelength 610 nm line).

(I believe you get the point now.  You can imagine many more combinations of green and red lights that lead to "yellow.")

 

To our human eyes, all three will look yellow.  The "yellow" you see from the spectrum you show is case-A yellow.  It will be rendered yellow in your DSLR case, and either red or green in your non-overlaping filter case.  However, both B) and C) will be rendered yellow by non-overlaping filters as long as you don't further separate them with a spectrograph.

 

In real astronomical images, most of the subtle color come from continuum objects.  They are not monochromatic.  On such objects (like cases B and C), even non-overlaping filters can render their color variations.  They won't look purely red, green, or blue.  A potential problem is OIII.  It is cyanish and it is almost monochromatic.  If one uses truly non-overlaping filters, it will be either green or blue and never cyan.  Fortunately most astronomical filters purposely place an overlap between G and B around the wavelength of OIII.  So this is not really a problem.

Yes, I was careful to specify that this happens only for spectrally pure colors, and only for non-overlapping filters. The prevalence of narrowband imaging shows that spectrally pure colors are not unusual in astrophotography. Ha, Hb, OIII SII are not uncommon. OIII might be captured correctly , but only when the filters are non-overlapping (hence my care in wording above).
 

[arbit]

As I said above, yes, you are capturing the yellow photon with the RGB setup. but it is always mapped to either red or green. If we are talking about spectrally pure yellow, no pixel will have both R and G values. More generally, for spectrally pure colors, every pixel will be either (R, 0.,0) ), (0, G. 0), or (0, 0, B). .That's what happens when spectrally pure colors are imaged by three non-overlapping R, G, and B filters on a mono camera.. And you can see this result in my image. So no yellow will appear in the final image. However if the colors are not spectrally pure colors, or if they are mixtures of spectrally pure colors, then the right colors might be produced.

Then where does the yellow in my RGB images come from? Clearly, I am missing some nuance here.

While on the topic, a question I haven't found an answer to.

Astronomik has a 2c series filters which overlap much.more than their "normal" RGB set. They say those follow the human eye better. Would that set make a difference then in a final image?

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



Sent from my SM-S908E using Tapatalk

[loujost]

Then where does the yellow in my RGB images come from? Clearly, I am missing some nuance here.

While on the topic, a question I haven't found an answer to.

Astronomik has a 2c series filters which overlap much.more than their "normal" RGB set. They say those follow the human eye better. Would that set make a difference then in a final image?

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

astronomik-lrgb-typ2c_trans.jpg

Sent from my SM-S908E using Tapatalk

I don't know where your yellow comes from, but if you read my comment carefully, maybe you can see whether my conditions apply to your situation.

 

Yes, those overlapping filters should greatly reduce the problem I described, retaining the color diversity of the object regardless of whether the color is spectrally pure (as in emission nebulae) or not.

 

However, that filter set does not really mimic human vision as they say. Human red and green cones have very large overlaps, while blue cones have much less overlap. The red cones also have a slight sensitivity to violet; this is missing in the Astronomik filter set so spectrally pure violet will be lost..

Edited by loujost, 12 March 2023 - 09:55 PM.

[James Peirce]

Yes, I was careful to specify that this happens only for spectrally pure colors, and only for non-overlapping filters. The prevalence of narrowband imaging shows that spectrally pure colors are not unusual in astrophotography. Ha, Hb, OIII SII are not uncommon. OIII might be captured correctly , but only when the filters are non-overlapping (hence my care in wording above).

If your spectrum is contiguous between the filters you are capturing proper representation of the "spectrally pure" colors as well, as I specifically attempted to explain (not necessarily successfully) in my previous reply. You will capture that bandpass regardless of whether 1) it is captured with high efficiency in red but not green (these scenarios occur frequently with mono RGB filters that feature little overlap), or 2) it is captured with 50% efficiency in both green and red or 3) it is captured with 30% efficiency in green and 70% efficiency in red (the later cases being more probable for overlapping filters like the type 2c example arbit shared or CFA).

 

We are only losing "spectrally pure" wavelengths if the filters specifically leave a gap at that bandpass range. And even then, it can be left for reasons worthy of consideration (e.g. biases taking into account quantum efficiency relative to wavelengths of modern sensors, done in color filter arrays to tailor exposures best for the human eye, or in astronomy filters to push toward more accurate capture of spectrum, or omitting a range deliberately such as to cut sodium vapor). An example where such factors are considered would be the Astronomik DeepSky RGB filters.

 

You actually have a lot at play with consumer sensor color filter arrays which works against balanced color reproduction for deep sky imaging, requiring fussier color calibration (or PixInsight’s crazy-good SPCC), and a lot which works in favor of color balance with astronomy filters. And the key difference here is the goal of balancing against limits of the human eyes for the world around us as we perceive it vs capturing the broad spectrum well and with balance. And there's also a lot of room, here, to get caught up in nuance which simply isn’t meaningful in an applied sense, such as trying to pin down a clearly represented deleterious consequence of cutting sodium vapor. "Spectrally pure" in yellow is a good example of getting caught up in the weeds, here, as that bandpass range is one which is strongly contributed to by a wide range of light sources, rounding back to the issue with the original post’s graphics where one represents blended spectrum (CFA) and the other (mono RGB) does not. It would be a bigger issue if we were creating blind spots for important deep space signals like Hydrogen-alpha.

Edited by James Peirce, 12 March 2023 - 09:50 PM.

[loujost]

We are only losing "spectrally pure" wavelengths if the filters specifically leave a gap at that bandpass range. And even then, it can be left for reasons worthy of consideration (e.g. biases taking into account quantum efficiency relative to wavelengths of modern sensors, done in color filter arrays to tailor exposures best for the human eye, or in astronomy filters to push toward more accurate capture of spectrum, or omitting a range deliberately such as to cut sodium vapor). An example where such factors are considered would be the Astronomik DeepSky RGB filters.

 

 

You seem to be misunderstanding something. Even so, you can see the effect in my pictures, and you can repeat these for yourself.  I am not saying that a spectrally pure yellow photon is lost, I am saying that it is mapped to green or red, so that no yellow will appear in the LRGB image. Take a picture of a spectrum with your own mono camera and nonoverlapping RGB filters. You won't get any yellow. You won't get any celeste or violet or orange either. It is physically impossible. You will just get plain red, plain green, and plain blue.

Edited by loujost, 12 March 2023 - 10:17 PM.

[James Peirce]

You seem to be misunderstanding something. Even so, you can see the effect in my pictures, and you can repeat these for yourself.  I am not saying that a spectrally pure yellow photon is lost, I am saying that it is mapped to green or red, so that no yellow will appear in the LRGB image. Take a picture of a spectrum with your own mono camera and nonoverlapping RGB filters. You won't get any yellow.

At this point I'd be re-hashing what has already been explained above.

[loujost]

At this point I'd be re-hashing what has already been explained above.

Check it yourself. Do the experiment.