• Members 976 posts
    March 28, 2023, 3:40 p.m.

    "Imagine you adjust two separate color-tunable LED fixtures to visually match in color appearance, but when you pull out your expensive, accurate spectroradiometer and measure the chromaticity of each, there is a significant difference that should be noticeable. How can that be?

    Alternatively, you tune two LED fixtures to be a perfect chromaticity match, according to that meter. Then, you show them to everyone in your office. Chaos ensues when the team realizes they are not a visual match for most of the group despite what the meter says."

    CIE Tristimulus functions

    www.ies.org/lda-magazine/featured-content/color-the-limits-of-a-standard-observer/

  • Members 621 posts
    March 28, 2023, 3:49 p.m.

    Iliah…awesome to see you here!

  • March 28, 2023, 3:54 p.m.

    Thanks Iliah for posting something I can stop and think about. I see the standard observer as a mathematical model, highly simplified, of the 'reality' of human colour perception. There have been papers on its failures. We know that it deviates from that ground truth simply in order to be mathematically manipulable - there is no guarantee that evolution produced something that matched the LIM condition. Still, it has worked reasonably well.

  • Members 976 posts
    March 28, 2023, 4:50 p.m.

    1931 observer works reasonably well, true. For what I'm doing for reproduction, 1964 observer worked better, but I was forced to convert to 1931 observer for RGB output (this is a general problem that prevented many from using 1964 observer). Luckily, most of my production output is CMYK. 2015 observer is much less prone to metameric error, especially blue <-> purple, but if the working RGB spaces don't support it, IMHO it is destined to very narrow use (textile / paint / pigment / light colour matching, mostly).

  • Members 976 posts
    March 28, 2023, 4:50 p.m.

    And back to you ;)

  • Members 78 posts
    March 29, 2023, 10:33 a.m.

    The reason why I tend to use the 1931 Observer is that most easily available standard color space data (e.g. white points, conversion matrices, etc.) are for it; and for compatibility with the vast majority of software in use today.

    Also anecdotally, when determining Forward Matrices by minimizing color differences with a standard target like a ColorChecker24, I get better mean dE2000 residual errors with the 1931 Observer than with 2015's. I haven't investigated why - but perhaps dE2000 expects 1931-referred data?

    Jack

  • Members 976 posts
    March 29, 2023, 10:46 a.m.

    Hi, Jack
    Lab is 1931-referenced, right. Also, how do you obtain reference XYZ values?

  • Members 78 posts
    March 29, 2023, 12:40 p.m.

    Hey Iliah, in those cases I try to do everything starting from spectral data. Here is the relative Matlab routine, hope it is self explanatory.

    function o = computeRef(o)
          %compute illuminant white point and target xyz/lab reference values referred to it.  All energy units.
    
          o.xyzn = o.illSpectrum * o.cmf;                           %white point of chosen illuminant
    
          o.cc24irradiance = o.cc24reflectance .* o.illSpectrum;
          o.xyzRef = o.cc24irradiance * o.cmf;                  %cc24 xyz reference values
          o.labRef = xyz2labn(o.xyzRef,o.xyzn);               %cc24 lab reference values
    
          o.xyzRef = o.xyzRef / o.xyzn(2);                         %normalize to Y
          o.xyzn = o.xyzn / o.xyzn(2);                                %normalize white point
    
          [o.cctRef,o.tintRef] = cctDuv(o.xyzn);                 %CCT and Tint of reference white
    end
    

    I suspect dE2000 internal parameters to be fine tuned to the 1931 Observer.

  • Members 976 posts
    March 29, 2023, 12:46 p.m.

    Right on both accounts, start from spectra and dE00 is for 1931 observer; also LCH from Lab not referenced to 1931 is a nightmare, Duv too.
    Good ideas in 2015, but very limited workflow, IMO not ready for practical use for general colour management, but at least they tell what CMFs to use.
    Maybe this will be of some help:
    files.cie.co.at/LpR%2060%20CIE%20RESEARCH%20Special%20-%20Cone-Fundamental-Based%20CIE%20Colorimetry.pdf
    colour.readthedocs.io/en/develop/_modules/colour/colorimetry/transformations.html

  • Members 878 posts
    March 29, 2023, 3:41 p.m.

    Is that something they imagined or something they actually experienced?

    BTW, this forum is hiding in an obscure corner of this site, hard to find.

    ++++
    The poster formerly known as J A C S.

  • Members 976 posts
    March 29, 2023, 5:41 p.m.
  • Members 1737 posts
    March 29, 2023, 11:44 p.m.

    ChatGPT misses the point.:

    chatgpt and the 1931 experiment.png

    chatgpt and the 1931 experiment.png

    PNG, 320.6 KB, uploaded by JimKasson on March 29, 2023.

  • Members 878 posts
    March 30, 2023, 11:17 a.m.

    That was a rhetorical question, and the answer seems to be that they imagined it, as they said. A more serious article, coauthored by three specialists, would present some evidence that those theoretical differences are actually observable in the light fixture industry to the extent they cause chaos.

  • March 30, 2023, 12:18 p.m.

    Here is an interesting thing. At present I'm spending some of my time refurbishing a house. I've been refitting the bathroom, which at present is finished with yellow tiles. The tiles aren't available any more, but some need replacement. I don't want to replace them all, because it's hard work and because they were put there by my father in his favourite colour. My plan has been to redistribute and decorate the spaces in a matching colour. I went to the paint shop with a piece of tile to get a match. He first took out the paint vendors' colour matching machine, measured it and got out the sample for the recipe suggested. Comparing them he thought it didn't quite match. So he got out his CA-410 and used that, did some calculations and arrived at a different spot, compared again and still wasn't satisfied. He moved three patches to the left, looked at it again, took them to the next room which had a different light source, took them outside into the daylight, came back and said 'I think that will do' I approved, he mixed it. I painted. After the paint had dried it was a perfect match. Yet the pigment mix must be completely different from the original tiles.

  • Members 976 posts
    March 30, 2023, 1:29 p.m.

    They didn't imagine it, because I and many others experienced it a lot.

  • Members 976 posts
    March 30, 2023, 1:36 p.m.

    Did he use one of 2015-referred CA-P427C / CA-P410C probes?

  • Members 878 posts
    March 30, 2023, 2:19 p.m.

    They forgot to quote you and the others.

  • Members 1737 posts
    March 30, 2023, 2:42 p.m.

    Illuminant metameric error is tricky to deal with in the absence of the prospective illuminant spectrum. Your paint provider made a good, and presumably educated, guess.

  • March 30, 2023, 8:29 p.m.

    I would say 'skilled' rather than educated. I'm not sure he had beed educated, but years of experience had been absorbed and allowed him to predict when his equipment wasn't giving the right answer. The match looks perfect in all of the illuminations I've seen it in so far.

  • Members 78 posts
    March 31, 2023, 9:54 a.m.

    Snapper-JACS, that' one creepy avatar ;-)

  • Members 878 posts
    March 31, 2023, 11:56 a.m.

    Brought to you by Bing AI or whatever they call it… That’s the future, l am afraid…

    I know, I will change when I have the time.