I was surprised a site search on this topic yielded 0 results. Has anyone here experimented with stiles and birch chromacity? If not, is it of interest?
Here was my introduction to it, also detailing some of the issues with cie xyz:
https://clarkvision.com/articles/color-cie-chromaticity-and-perception/
https://clarkvision.com/articles/color-spaces/
What do you mean by “experimented”? As photographers we have very limited access to change hardware (camera, monitor, prints). What kind of experiments do you suggest that can be done just by software?
This type of information is getting beyond my understanding, which is why I was interested if others here had a take. But as I understand, in editing, many processes go from rgb > cie xyz and back. I wonder if there might be virtue in going to stiles and birch instead. Otherwise, it appears an alternate color space has been presented with deeper reds and blues. I haven’t seen this available as working space in any piece of software, I wonder if anyone has tried implementing it?
Someone straighten me out if my understanding is off.
It’s not a working space, it’s simply a description of our eyes’ response, so I expect there’s no good way to use it performantly.
XYZ is a simplification of that with fully positive response curves, it seems, and that causes some nonuniformity.
So, according to Clark,
“The simplifications and approximations of the CIE standard date from the 1930s when people did not want to do hand calculations with negative numbers. Modern computers for at least the last 70+ years have no such limitations. The simplifications and approximations are now adding confusion and further approximations to a complex world as we move into better color display technology. These simplifications are holding back development and presentation of accurate perceived color. It is time to abandon the current CIE approximation models and get back to the fundamental underlying data on human perceived color.”
CIE 1931 XYZ is one of the two connection spaces used in ICC profiles. Every transform from one space to another goes through a connection space, and that is predominantly XYZ. I think there’s a lot of inertia behind the CIE reference, in tools and workflows, so changing it out would be onerous.
What’s interesting to me in the comparison is the degree of similarity across the years and methodologies. 17 subjects for Wright and Guild, 45 subjects for Stiles and Birch. Yes, there’s goofiness in the lower end of the high response, but the human sensitivities peak in about the same places. That starts to be statistically significant.
Color encoding in general purpose cameras is an approximation, based on three bandpass filters that roughly peak where the XYZ sensitivities peak. That’s the other consideration: for general purposes, would the difference be noticeable enough to warrant a switch?
Read (Wikipedia entries may have factual errors but they are good enough for general comprehension.)
For ICC to work, approximations and non-negative curves are necessary. There are newer models and spaces that address these issues. I suppose it would take ages for them to find traction. In the meantime, new colour spaces have emerged such as Jzazbz and Oklab Colorspace to make colour more palatable (measured by JND and other metrics).
If we are aiming for perfection, there is no such thing (I don’t even like Jzazbz ATM), but there are a dozen little things we could do to improve the situation such as LUT corrections and deriving camera profiles from sensor spectral profiles.
Not sure that this is the right interpretation. The claim Clark seems to make (and I have no idea how to test that) is that Stiles-Birch shows less ambiguity in the blue spectral region. I’m not even sure that he claims that the negative tristimulus values are the cause for less ambiguity, or if just the chromaticity matching functions are more precise. I’d be curious if Stiles-Birch still has ambiguity but (much?) less than CIE XYZ or if it has none for testing with that-2Gaussian-method.
With more spectral blue lightsources on every christmas tree and everybody wielding their smartphones with HDR10 displays…maybe.
Camera-SSFs are also agnostic in terms of CIE or Stiles-Birch, so it’s a very good idea to have them anyway. I think after reading about Clarks spectrometry of displays I think the concept and importance of monitor calibration (and monitor profiling) should be omnipresent.
That was what I meant.
In addition to what I wrote in the previous post, I want to note that there is a level of variability in the HVS among humans and different viewing conditions. Not only is approximation and non-negativity necessary for modelling and practical transformation between colour spaces (using 3x3 matrices) but smoothing general takes out the kinks in data. We do it all the time.
There is much to learn in the realm of colour science. Research methods have become more robust since. I am curious what we would find if we examined chromacity more closely and broadly in light of high dynamic range and various lighting conditions, not just the basic ones covered by CIECAM02/16.