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

"True color" again

Astrophotography
  • Please log in to reply
438 replies to this topic

#126 sharkmelley

sharkmelley

    Fly Me to the Moon

  • *****
  • Posts: 7,363
  • Joined: 19 Feb 2013
  • Loc: UK

Posted 05 March 2023 - 03:59 AM

Stars or other approximately black body radiators can also be accurately assigned their true color. Spectrometer data can be used to check or even artificially generate images of such objects.

 

https://clarkvision....ium.true.color/

 

Don't assume this "true colour" result is necessarily correct.  The methodology used was to convolve CMFs (colour matching functions) with the measured SPD (spectral power distribution) of the M42 emissions.  In principle this is an excellent way of doing things since it bypasses the known limitations of the colour responses of a camera.  However, in this particular case there were two problems:

  • The Stiles and Birch (stet) RGB CMFs were used instead of the standard CIE CMFs.  This means the resulting RGB values are in a non-standard colour space.  This is not necessarily a problem as long as the necessary transformation is done from this "Stiles and Burch" colour space to the standard colour space being used for the image. Unfortunately the RGB values were used to populate an sRGB image file with no attempt to perform the required transformation to sRGB.
  • The second problem is that the RGB values resulting from the convolution are linear but they've been used to populate an sRGB image file without the appropriate sRGB tone response curve (gamma) being applied.

This is all basic colour science and with a little extra effort it could have been done correctly.  Someone over on DPReview did perform the correct colorimetric steps on that M42 SPD.  If I remember correctly, there wasn't much difference in the case of M42 Trapezium region but that was by luck.   The same approach used for the SPD of stars leads to much bigger errors.

 

Yes, spectrometer data can certainly be used to determine the colour of stars and other night sky objects but only if the correct colorimetric processing workflow is used.

 

Mark 


Edited by sharkmelley, 05 March 2023 - 04:34 AM.


#127 BQ Octantis

BQ Octantis

    Cosmos

  • *****
  • Posts: 8,805
  • Joined: 29 Apr 2017
  • Loc: Path of Totality, Texas, USA

Posted 05 March 2023 - 07:01 AM

So now that we may be done throwing food at each other. let's look at some serious attempts to find the true (or realistic or natural) color of the Trapezium. This shows what can be done and how objective it can be, especially when most of the light comes from various narrow-band sources whose wavelengths (and human responses to those wavelengths) are known. Stars or other approximately black body radiators can also be accurately assigned their true color. Spectrometer data can be used to check or even artificially generate images of such objects.

 

https://clarkvision....ium.true.color/

https://www.dpreview.../thread/3969219

 

And since BQ mentioned Lodigruss, here is his take on the same subject:

 

http://www.astropix....color-m42-core/

 

The captured images and the image synthesized from spectrometer data agree reasonably well, for many tested digital cameras.

 

This is one way to test our equipment to see if it produces true-color (realistic color) astro imagery.

And it turns out that there are "sufficiently large  telescopes" to allow our eyes to begin to see the colors of the brightest nebulae:

 

https://www.cloudyni...ographed/page-2

 

though there is a fair bit of individual variation.

And now we've come full circle—which is another square on the CN Controversial Topic Bingo card. We're still missing a few, mostly what I would consider negative

 

  • TOMDEY Facts (these are always super-fun and informative!)
  • Deep pontifications of disproven dogma with certitude by demi-experts (these are painful and counter-productive)
  • Thread hijacked (these can be fun, particularly when friends plan or discuss day-to-day activities)
  • Ad hominem attacks between experts (these are the most painful to watch—I witnessed Lodriguss's meltdown in near-realtime)
  • Moderator warning
  • Thread locked

 

But this is a choose-your-own-adventure bingo, mate! You only have 63 posts, so I'll assume you don't know it works. As the OP posing the controversial question, you have to lay down the ground rules for exchanges. For instance, going off your two posts above:

 

1. Arguments only with information and charts from Roger Clark's site (or just links to entire threads that summarize our position):

 

Random quote #1:

 

"Color Space describes the different colors, called a gamut, that a device like a computer monitor or print can show, as well as the range of colors that an imaging device, like a digital camera can record. Color space models are simplified standards for devices and not a precise description of human color perception. The color displayed by different devices can appear different even when calibrated to the same color space. The black level also impacts the gamut and color saturation of images." [Source, emphasis mine]

 

 

Random Chart #1:

plot.2.5-thresh-contrast-10angles-photop

[Source]

 

Random Links:

 

#1: https://clarkvision.....the.night.sky/

#2: https://clarkvision....s/color-spaces/

#3: https://clarkvision....or-and-critics/

 

 

2. Arguments only with links to CN

 

2a. Only threads with TOMDEY posts

2b. Only threads with Lodriguss posts

2c. Only threads with Mark posts (that square already got played early in this thread)

 

(We could probably make an entire bingo card out of these…for each forum, at that!)

 

I much prefer the Road Less Traveled square: actually measuring and comparing what we're trying to prove or disprove. After all, this is an imaging forum! So that is the adventure path I would choose. That would look like this:

 

You've picked Trapezium, among the largest and brightest nebula in the sky (waaay above the noise and light pollution and so easy to obtain data). Your calibrated unmodded camera is your spectrometer. Did you calibrate your sensor (per Clark)? Have you taken a picture of Trapezium? At what RA/DEC point or points are you measuring chromaticity (xy, see Clark)? At what pixel scale? What are your measurements? Which of the reference "true color images" are you comparing to? Did you correctly convert the reference image color data from sRGB to xy? And what stretch do you plan to use for your stretched image? This last one is critical because the only way to view linear data "truly" is to do a gamma stretch with the gamma of your calibrated BT.709 monitor (the sRGB stretch is more contrasty so it looks better, but is not "true").

 

So what path do you choose?

 

BQ

 

P.S. You don't have to prove you can see DSO color at the eyepiece to me. I have literally seen DSO Ha at the eyepiece:

 

https://www.cloudyni...h-mak-cass-180/

 

And I have many friends who see the core of the Milky Way at zenith as green.


Edited by BQ Octantis, 05 March 2023 - 08:52 AM.


#128 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 232
  • Joined: 02 Oct 2017

Posted 05 March 2023 - 08:57 AM

BQ, I'd add to your bingo card several other items:

 

- People ignoring what the OP actually said and replying to their favorite straw man related to the topic.

- People arguing against an OP even when they fundamentally agree with him/her.

- People who point out that something can't be done perfectly, and therefore it is silly to try.

 

But maybe we are done with the games, the ranters have ranted, and we can begin to generate constructive advice.

 

We now agree that, in principle, we can indeed see colors (at least for the brightest objects) with a sufficiently large telescope.

We can also measure the actual spectral distribution at each pixel, for a few objects.

We have models that can convolve these with eye response functions to get pretty good results.

We have models and equipment that can display these results with reasonable accuracy.

 

In other words, there is such a thing as realistic color in the sense I described at the outset. So now we can discuss how an amateur /beginner like myself might achieve this.

 

I would like to ask people who don't think this is a worthwhile goal to please not bother to comment. Also please do not comment if you think the goal of realistic color is unreachable because it can't be done perfectly. I am not after perfection, I am after something close to the standard we accept in general photography.

 

Here is a useful comment by Jack Hogan in a long thread involving Roger and Mark on DPReview (checks off multiple bingo card lines!):

 

"So what's an (astro)photographer to do? The good news is that there are no substantive unanswered questions out of these threads (other than the academic subject of dark adaptation on which I would welcome ideas): those 'true color' pages are full of massive, career limiting mistakes on the theory and practice or rendering color, resulting in wildly wrong tones being produced from spectrum: don't use them as reference when rendering your own captures. But feel comfortable in knowing that if an astrophotographer sets his camera up for D50 throughout the raw->XYZ chain with linear processing they should be ok as far as color is concerned. Other Dxx whites will work as well, the key is consistency (WB, matrix, etc.), somewhere in one of these threads I explain why."

 

Original here.


Edited by loujost, 05 March 2023 - 09:02 AM.


#129 BQ Octantis

BQ Octantis

    Cosmos

  • *****
  • Posts: 8,805
  • Joined: 29 Apr 2017
  • Loc: Path of Totality, Texas, USA

Posted 05 March 2023 - 09:53 AM

BQ, I'd add to your bingo card several other items:

 

- People ignoring what the OP actually said and replying to their favorite straw man related to the topic.

- People arguing against an OP even when they fundamentally agree with him/her.

- People who point out that something can't be done perfectly, and therefore it is silly to try.

 

I'm clearly a 2.5. But I like to argue with, not against. And I don't think it's silly to try—I think we just need to be prepared to acknowledge the technical shortcomings of our claims.

 

 

In other words, there is such a thing as realistic color in the sense I described at the outset. So now we can discuss how an amateur /beginner like myself might achieve this.

 

I would like to ask people who don't think this is a worthwhile goal to please not bother to comment. Also please do not comment if you think the goal of realistic color is unreachable because it can't be done perfectly. I am not after perfection, I am after something close to the standard we accept in general photography.

 

All stock daylight cameras are designed and built to capture the spectrum of human vision to some reasonable degree of metameric matching. Get one, make sure it's got a raw mode. Calibrate it with the lens you plan to use on a set of reference colors (I've used paint color swatches from HomeDepot, Lowes, Walmart, Target, and finally a Datacolor Colorcheckr24) to obtain a CCM for the setup. Now you have a calibrated spectrometer. With the same setup, shoot a picture of your target and apply your CCM to the white-balanced, background-extracted data with pixel math. From the linear, color-corrected data, you can now make some chromaticity claims about the target. Note that CA makes star color claims problematic.

 

When you go to display your image when you make your claims, stretching is also important. The CCM yields linear data that requires stretching with your monitor's gamma to see "truly". It's not that a gamma stretch is color bleaching—it's that the raw camera primaries yield very unsaturated chroma (with hue errors) in linear space that depend on the CCM to correct (hence why it's called a color correction matrix). Asinh and GHT boost the chroma while maintaining hue as they stretch, so applying them to the linear CCM data will oversaturate pretty quickly. And you'll need to make sure your image processor knows what to do with the sRGB slope+gamma curve—or you'll have to apply it manually with pixel math.

 

Lastly, much like the Oracle at Delphi, you should consider the purpose of the forum where you pose your query. Color matching AP to human vision isn't remotely the objective of this forum. For that goal, you'll get less dogma in the DSLR forum—and you'll find some experts there who even have the tools to calculate the E of your CCM.

 

Cheers,

 

BQ


Edited by BQ Octantis, 05 March 2023 - 02:03 PM.


#130 BQ Octantis

BQ Octantis

    Cosmos

  • *****
  • Posts: 8,805
  • Joined: 29 Apr 2017
  • Loc: Path of Totality, Texas, USA

Posted 05 March 2023 - 06:00 PM

With the same setup, shoot a picture of your target and apply your CCM to the white-balanced, background-extracted data with pixel math. From the linear, color-corrected data, you can now make some chromaticity claims about the target. Note that CA makes star color claims problematic.

Here is an example—Trapezium in the truest of true colors based on a calibrated, color-managed workflow from data from a camera with a Metamerism Index of 73 (with no DSO hue/chroma manipulation beyond white balance, white & black points, and application of the daylight CCM):

 

Trapezium's Truer Colors (Click for full size @ 100% scale)

med_gallery_273658_21104_965298.jpg

 

Give it a go!

 

BQ

 

P.S. I question the accuracy of the "ground truth" reference image Roger created from filter data. And tint controls in Photoshop for color matching…really?


Edited by BQ Octantis, 05 March 2023 - 08:59 PM.


#131 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 232
  • Joined: 02 Oct 2017

Posted 05 March 2023 - 06:57 PM

BQ, thanks very much for this useful input. The sample picture you post is exactly what I am looking for, and it is nice to know that you also sometimes  find this goal of "true (or truer anyway) color" interesting!

 

Do you have links to other "true color" images that meet your criteria for accuracy?

 

I prefer to use monitors with wider gamuts than sRGB  but I expect the monitor calibration will take care of that difference, if the image is viewed in a color-manged program.

 

I didn't think to put this in the DSLR forum because it is more general than that. I use Nikon, Sony, Olympus, Pentax, Panasonic, and QHY cameras, and only the Nikon and Pentax are  DSLRs. I'd especially like to find filters that overlap enough to be able to obtain "truer colors" from cooled monochrome astro cameras. Do you know if any filter sets are available that mimic the behavior of Bayer microfilters?


Edited by loujost, 05 March 2023 - 06:58 PM.


#132 BQ Octantis

BQ Octantis

    Cosmos

  • *****
  • Posts: 8,805
  • Joined: 29 Apr 2017
  • Loc: Path of Totality, Texas, USA

Posted 05 March 2023 - 08:07 PM

For me, the effort is science in the service of art. The daylight CCM suffers metameric failure on bright stars (and it severely overstates the CA from my fast lenses), so I also have a set of Planckian CCMs to improve my star colors. The one for my Mak 180 renders Trapezium a shade much closer to what Roger Claims is "true" (which again, I don't believe exists):

 

Trapezium's Planckian Personality (Click for full size at 100% scale)

med_gallery_273658_21104_1002577.jpg

 

I'd have to back out the xy values for all the images to compare the chromaticities with Roger's "true colors", but Roger's use of Photoshop tint controls for color matching undermines the credibility of the comparison—so it's hardly worth the exercise. And I have no reason to doubt the accuracy of the DXOMark D50 CCM (which compares well to the 5 other CCMs I derived from color targets). But I tend esthetically to favor a blue-to-cyan Trapezium over the greenish cyan I get from the CCM.

 

Looking at the specs for color filters, the spectra tend to be very flat band-pass windows. And as I stated before, I was unsuccessful at getting Planckian colors from another OP's L-Enhance-on-OSC data. The metameric performance of AP-specific hardware just isn't a thing.

 

BQ

 

P.S. I got a TOMDEY bingo square today! cool.gif


Edited by BQ Octantis, 05 March 2023 - 08:56 PM.


#133 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 232
  • Joined: 02 Oct 2017

Posted 05 March 2023 - 09:43 PM


P.S. I got a TOMDEY bingo square today! cool.gif

I've been wondering what that means since you used the term yesterday. I see he is a user on CN but what does it mean?



#134 BQ Octantis

BQ Octantis

    Cosmos

  • *****
  • Posts: 8,805
  • Joined: 29 Apr 2017
  • Loc: Path of Totality, Texas, USA

Posted 05 March 2023 - 10:27 PM

I've been wondering what that means since you used the term yesterday. I see he is a user on CN but what does it mean?

 

flowerred.gif  Sorry, he's an amateur APer from the days of film who's also ostensibly retired from a career in remote sensing. On controversial or puzzling technical subjects, he'll occasionally chime in with an off-the wall but relevant example from years back from one of those two domains—usually with hand-annotated photographs or documents relevant to the subject at hand. It was from his posts that learned what a Lambertian target was, as well as étendue. So to me, those bingo squares are among the most coveted given their educational value and rarity.


Edited by BQ Octantis, 05 March 2023 - 10:29 PM.


#135 sharkmelley

sharkmelley

    Fly Me to the Moon

  • *****
  • Posts: 7,363
  • Joined: 19 Feb 2013
  • Loc: UK

Posted 06 March 2023 - 02:10 AM

Here's my (undersampled) version from an unmodified Canon EOS R:

 

EOSR_Trapezium.jpg

 

I also managed to find Jack Hogan's calculation in the DPReview conversation where we were discussing this with Roger Clark:

https://www.dpreview...s/post/60867429

 

Note how close the EOS R colour comes to Jack's calculation

 

Performing these calculations yourself might seem a bit daunting but Bruce Lindbloom has helpfully made available some spectral calculator spreadsheets:

http://www.brucelind...readsheets.html

 

All you need is the SPD (spectral power distribution) of your chosen object.

 

One final point.  If you use a Sony sensor instead of Canon then the result comes out a lot bluer.  But that's a whole new conversation!

 

Mark


Edited by sharkmelley, 06 March 2023 - 02:11 AM.


#136 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 232
  • Joined: 02 Oct 2017

Posted 06 March 2023 - 11:15 AM

Thanks, this is really interesting.

 

I suppose the difference between the Sony and Canon results are due to slightly different blue/green cutoffs of their Bayer filters? In ordinary photography the accuracy of color rendition of the oxygen emission line is not very important...

 

I just found an old diffraction grating spectrometer that I had forgotten about. Now I can instantly check the transmission characteristics of my filters, and calibrate my camera sensors in absolute terms (I can see the major Fraunhofer lines when looking at daylight, so it is easy to calibrate)... Yippeee!



#137 sharkmelley

sharkmelley

    Fly Me to the Moon

  • *****
  • Posts: 7,363
  • Joined: 19 Feb 2013
  • Loc: UK

Posted 06 March 2023 - 06:01 PM

I suppose the difference between the Sony and Canon results are due to slightly different blue/green cutoffs of their Bayer filters?

That's right - the overlap between blue and green is quite different.  Take a look at this comparison:

 

SunlightSpectra.jpg

 

The Canon 200D (the plot erroneously says D200) is fairly typical of recent Canon cameras and the Nikon D5300 is fairly typical of Sony sensors.  Take a look at the big difference in the position of the boundary between blue and green.

 

Mark


  • badgie likes this

#138 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 232
  • Joined: 02 Oct 2017

Posted 06 March 2023 - 06:04 PM

That's right - the overlap between blue and green is quite different.  Take a look at this comparison:

 

attachicon.gifSunlightSpectra.jpg

 

The Canon 200D (the plot erroneously says D200) is fairly typical of recent Canon cameras and the Nikon D5300 is fairly typical of Sony sensors.  Take a look at the big difference in the position of the boundary between blue and green.

 

Mark

That's a surprisingly big difference.



#139 BQ Octantis

BQ Octantis

    Cosmos

  • *****
  • Posts: 8,805
  • Joined: 29 Apr 2017
  • Loc: Path of Totality, Texas, USA

Posted 06 March 2023 - 11:17 PM

Here's my take on the true colors of Trapezium, based on a PCC white balance followed by a daylight-calibrated CCM.

 

I measured the 10×10 average BT.709 linear values for 27 points randomly placed on the brightest part of the nebula:

 

(Click for full size)

sml_gallery_273658_12412_223502.jpg

 

I calculated the xy values and plotted them on the same chromaticity diagram I used to calibrate my camera:

 

(Click for full size)

sml_gallery_273658_12412_96650.jpg

 

I also calculated the OKLab values from the data to assess the hue and chroma in a perceptual color space:

 

(Click for full size)

sml_gallery_273658_12412_140143.jpg

 

The hues in Trapezium clearly do not fall on a perfectly straight line. The hues for the higher chroma data blob range from 205˚ to 220˚. So Trapezium is clearly not just a single color.

 

Lastly, I recalculated the sRGB values for the (hue,chroma) pairs for a constant OKLab lightness value equal to the highest lightness of all the points (L=0.670). Properly sorted, this made for a very nice, perceptually uniform spectrum of the colors of Trapezium in sRGB color:

 

(Click for full size)

sml_gallery_273658_12412_60704.png

 

Comparing the measured data against the calibration points and the sRGB gamut, the measured chromaticities fell well within the calibration range of the setup and the ability of a BT.709 monitor to reproduce them. From a ground truth perspective, the DPReview reference color Mark provided had a (hue,chroma) pair of (204˚,0.079). Normalized to the same lightness I used for the table, the perceptually uniform sRGB values for the reference were (83,164,171). If I were to add it to the table and resort it, the reference would fall third from the bottom—well within family with the highest chroma data.

 

Cheers,

 

BQ


Edited by BQ Octantis, 06 March 2023 - 11:53 PM.


#140 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 232
  • Joined: 02 Oct 2017

Posted 07 March 2023 - 07:00 AM

BQ, is the smaller triangle in your xy graph sRGB and the larger one Adobe or ProPhoto RGB?



#141 BQ Octantis

BQ Octantis

    Cosmos

  • *****
  • Posts: 8,805
  • Joined: 29 Apr 2017
  • Loc: Path of Totality, Texas, USA

Posted 07 March 2023 - 02:13 PM

BQ, is the smaller triangle in your xy graph sRGB and the larger one Adobe or ProPhoto RGB?

The smaller triangle is the gamut for both sRGB and BT.709 (a.k.a. Rec.709). I quit using sRGB to refer to the primaries because of the constant confusion between its stretch curve (slope=12.92, gamma=2.4), its white point, and its primaries.

 

The larger triangle is the gamut for Adobe RGB, which I added from another source long after I built the spreadsheet and plot.

 

I might one day fix the legend…

 

BQ



#142 sharkmelley

sharkmelley

    Fly Me to the Moon

  • *****
  • Posts: 7,363
  • Joined: 19 Feb 2013
  • Loc: UK

Posted 07 March 2023 - 05:16 PM

The smaller triangle is the gamut for both sRGB and BT.709 (a.k.a. Rec.709). I quit using sRGB to refer to the primaries because of the constant confusion between its stretch curve (slope=12.92, gamma=2.4), its white point, and its primaries.

BT.709 also has a "stretch curve", a white point and primaries.  So how is it you think that sRGB is confusing but BT.709 is not?



#143 BQ Octantis

BQ Octantis

    Cosmos

  • *****
  • Posts: 8,805
  • Joined: 29 Apr 2017
  • Loc: Path of Totality, Texas, USA

Posted 07 March 2023 - 06:15 PM

BT.709 also has a "stretch curve", a white point and primaries.  So how is it you think that sRGB is confusing but BT.709 is not?

 

The sRGB primaries are literally defined as the BT.709 primaries. For instance, read the w3 sRGB spec for yourself:

 

"The CIE chromaticities for the red, green, and blue ITU-R BT.709 reference primaries, and for CIE Standard Illuminant D65, are given in Table 0.2."

Screen Shot 2023-03-07 at 6.14.24 PM.png

 

If I say "sRGB primaries", I've frequently gotten "oh, sRGB isn't linear".

 

If I say "BT.709 primaries", no one assumes I mean something other than that.

 

BQ


Edited by BQ Octantis, 07 March 2023 - 06:51 PM.


#144 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 232
  • Joined: 02 Oct 2017

Posted 07 March 2023 - 10:20 PM

I have my little spectrometer set up and working with my camera. With this I can see what the Ha line and any other lines really look like to my eyes. The red end of the spectrum seems compressed relative to photos of spectra online, making it hard to identify the Ha line, which is the Fraunhofer C line. Does anyone have any thoughts on the location of the Ha line on this spectrum? The other Fraunhofer lines (D, E, F, G) show that my photo is linear with respect to wavelength in those regions (0.085 nm/px) but I am not sure about the red part, nor do I know how sensitive the Olympus sensor is to Ha.

 

I have been comparing my photos with this one on the internet:

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

Attached Thumbnails

  • Oly Pen F spectrum.jpg

Edited by loujost, 07 March 2023 - 10:22 PM.


#145 sharkmelley

sharkmelley

    Fly Me to the Moon

  • *****
  • Posts: 7,363
  • Joined: 19 Feb 2013
  • Loc: UK

Posted 08 March 2023 - 02:21 AM

Does anyone have any thoughts on the location of the Ha line on this spectrum? 

The C line (H-alpha) is exactly where you marked it!

 

An image of a spectrum is very difficult to make natural looking because almost every wavelength is either out of the gamut of the colour space (typically sRGB) or out of the gamut of the screen displaying it, or both!  Typically the negative values of the out-of-gamut hues are simply clipped to zero but this causes annoying changes of hue.  The website here discusses various schemes for dealing with the out-of-gamut hues.  

 

I'm quite familiar with the synthetic spectrum at the bottom of the image you linked because in my opinion it's the best rendition I have come across.

 

Mark



#146 loujost

loujost

    Mariner 2

  • -----
  • topic starter
  • Posts: 232
  • Joined: 02 Oct 2017

Posted 08 March 2023 - 07:05 AM

The C line (H-alpha) is exactly where you marked it!

 

An image of a spectrum is very difficult to make natural looking because almost every wavelength is either out of the gamut of the colour space (typically sRGB) or out of the gamut of the screen displaying it, or both!  Typically the negative values of the out-of-gamut hues are simply clipped to zero but this causes annoying changes of hue.  The website here discusses various schemes for dealing with the out-of-gamut hues.  

 

I'm quite familiar with the synthetic spectrum at the bottom of the image you linked because in my opinion it's the best rendition I have come across.

 

Mark

Thanks Mark, That's a very interesting link you included. My main monitor has a very wide gamut so I can do a little better, but it is surprising how poor the yellows are represented by the photo of the spectrum, relative to what we see with our eyes.



#147 wongataa

wongataa

    Sputnik

  • -----
  • Posts: 27
  • Joined: 29 Jan 2023
  • Loc: UK

Posted 08 March 2023 - 12:59 PM

All stock daylight cameras are designed and built to capture the spectrum of human vision to some reasonable degree of metameric matching. Get one, make sure it's got a raw mode. Calibrate it with the lens you plan to use on a set of reference colors (I've used paint color swatches from HomeDepot, Lowes, Walmart, Target, and finally a Datacolor Colorcheckr24) to obtain a CCM for the setup. Now you have a calibrated spectrometer. With the same setup, shoot a picture of your target and apply your CCM to the white-balanced, background-extracted data with pixel math. From the linear, color-corrected data, you can now make some chromaticity claims about the target. Note that CA makes star color claims problematic.

 

When you go to display your image when you make your claims, stretching is also important. The CCM yields linear data that requires stretching with your monitor's gamma to see "truly". It's not that a gamma stretch is color bleaching—it's that the raw camera primaries yield very unsaturated chroma (with hue errors) in linear space that depend on the CCM to correct (hence why it's called a color correction matrix). Asinh and GHT boost the chroma while maintaining hue as they stretch, so applying them to the linear CCM data will oversaturate pretty quickly. And you'll need to make sure your image processor knows what to do with the sRGB slope+gamma curve—or you'll have to apply it manually with pixel math.

You really should be creating CCM to convert to wider gamut colour spaces than sRGB.  Ideally you would convert to ProPhotoRGB.  Using colour managed software together with calibrated monitors will then help you see all the colours your monitor can show.  If you did that it might be better to use a colour chart with a wider range of colours than the standard 24 patch one though.



#148 BQ Octantis

BQ Octantis

    Cosmos

  • *****
  • Posts: 8,805
  • Joined: 29 Apr 2017
  • Loc: Path of Totality, Texas, USA

Posted 08 March 2023 - 05:20 PM

You really should be creating CCM to convert to wider gamut colour spaces than sRGB.  Ideally you would convert to ProPhotoRGB.  Using colour managed software together with calibrated monitors will then help you see all the colours your monitor can show.  If you did that it might be better to use a colour chart with a wider range of colours than the standard 24 patch one though.

Welcome to CloudyNights, mate!

 

The truth is, I can't even get 24 colors (technically, 18 colors and 6 grays) to match with a single least-squares-fit CCM. Tossing in more colors will just create a one-size-fits less matrix. Only a lookup table can compensate for the metameric failure of the camera.

 

And to what end? If I'm shooting a particular object with a known measurable "truth", I don't need to optimize across the entire spectrum. I can just generate an optimized CCM for the chroma zone of interest. But even then, the exercise would be entirely academic: it wouldn't fix the metameric mismatch between even normal retinas:

 

post-273658-0-78144300-1678218006.jpg

"The images shown in Figure 3 illustrate the cone mosaics for five human subjects. In each image, the L cones are indicated in red, the M cones are in green, and the S cones are in blue. Each image was obtained from a different subject, and all subjects had normal trichromatic color vision according to standard tests. Several striking features of the mosaics are apparent."

[Source: D. Brainard, "Color and the Cone Mosaic", Annu. Rev. Vis. Sci. 2015. 1:519–46]

 

 

So I have no idea if what I see as "red" is the same as what you see as "red". Because individual perception (of any sense) is random, the 1931 CIE color space is based on the measurements of 10 (or 7) standard observers. It's much like the Scoville Scale, where 4 out of 5 people must agree that this chilli is red hot.

 

"A weakness of the Scoville organoleptic test is its imprecision due to human subjectivity, depending on the taster's palate and number of mouth heat receptors, which vary widely among subjects. Another shortcoming is sensory fatigue; the palate is quickly desensitized to capsaicinoids after tasting a few samples within a short time period. Results vary widely (up to ± 50%) between laboratories." [Source]

 

BQ



#149 BQ Octantis

BQ Octantis

    Cosmos

  • *****
  • Posts: 8,805
  • Joined: 29 Apr 2017
  • Loc: Path of Totality, Texas, USA

Posted 08 March 2023 - 05:28 PM

Typically the negative values of the out-of-gamut hues are simply clipped to zero but this causes annoying changes of hue.  The website here discusses various schemes for dealing with the out-of-gamut hues.

This is where I take exception to Roger's "modern" workflow with ACR or RawTherapee as somehow the gold standard. Only raw processors allow for the maintaining of negative values through the workflow instead of clipping to 0. Given the obvious hue change, when discussing "true" color I would expect such a credentialed expert—who's generated his own bloody xy diagrams by varying pixel sizes—to at least acknowledge the shortcomings and do some real chromaticity measurements vice doing a Warholesque comparison of daylight white balances with a generic CCM.


Edited by BQ Octantis, 08 March 2023 - 09:55 PM.


#150 BQ Octantis

BQ Octantis

    Cosmos

  • *****
  • Posts: 8,805
  • Joined: 29 Apr 2017
  • Loc: Path of Totality, Texas, USA

Posted 08 March 2023 - 07:26 PM

Here's my take on the true colors of Trapezium, based on a PCC white balance followed by a daylight-calibrated CCM.

I'm afraid I didn't disclose the whole truth on my prior take on Trapezium's true colors. My last remaining calibrated daylight CCM to shoot was my Mak 180—and yesterday was the first clear day since I calibrated my C5/750. And knowing approximately where the spectrum of the Trapezium data was on the xy plot, I optimized the new CCM for the set of points in the Trapezium Zone before applying it to the PCC white-balanced data:

 

Trapezium Tells the Truth (click for full size)

sml_gallery_273658_21104_1981097.jpg

 

There was a measureable shift in the hue and chroma—a quantification of the difference in the two truths:

 

(Click for full size toggle)

sml_gallery_273658_12412_1059557.png

 

Note how much closer the tip of the chroma vector gets to the DPReview D65 truth with the daylight-calibrated Trapezium-optimized CCM than the DXOMark D50 CCM. So I would call it the truthier of the two.

 

BQ


Edited by BQ Octantis, 08 March 2023 - 09:52 PM.



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





Also tagged with one or more of these keywords: Astrophotography



Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics