An experiment in colour (saturation)

[rodeo_joe1]

Preamble (skip if easily bored):

I must admit to always being puzzled by the CIE 'horseshoe' as to what its perimeter boundary actually means. I always understood it to delineate the limit of human perception of colour - our ability to differentiate different hues or frequencies of colour and their saturation. 

Now, colour-saturation can be defined in several ways. As a starting point; a monochromatic colour (single frequency - i.e. Laser light) must have 100% saturation, and as the bandwidth of that hue is widened, the colour must become less saturated. Agreed? 

So what parameter defines the human eye's ability to differentiate colours and colour saturation and hence the 'horseshoe' perimeter? Is it a simple frequency difference? For example; can we differentiate a 1nm wavelength change? 5nm? 10nm? Or is it the 'dilution' of colour purity by contamination with white light or other frequencies of colour? And what units do we put on that ability to differentiate colours? Because those objective units appear to be lacking in the perimeter of the CIE 'horseshoe' 

Anyway The Experiment - 

Below are rectangles of 'pure' RGB colours, with words hidden in them. The words are in a desaturated and slightly different hue. The eyedropper tool in Photoshop easily picks up the difference - my eyesight, not so much. 

So the experiment is simple. Who can see the words? Or even detect the subtle colour variation?

I admit to totally failing to detect a difference, but then I doctored the text for that very purpose. However the saturation does vary by 11 percent according to Photoshop's HSL tool. 

If, as I suspect, nobody can detect the desaturated areas, then that throws doubt on the veracity of that 'horseshoe', which after all was created over 90 years ago in 1931!

Because if nobody can detect a 10% variation in saturation within the sRGB triangle, then what is the perimeter of the CIE horseshoe doing being well outside of it? 

Can anyone explain that apparent anomoly?

FWIW, I have no impairment of colour perception and consider myself to have an excellent ability to differentiate between colours (for a male person).

P. S. I hope the PNG files I've uploaded don't get mangled to JPEGs by PN's system. 

Edited March 4, 2023 by rodeo_joe1

[rodeo_joe1]

P. P. S

After uploading, I can just detect the lettering in 3 of the rectangles when viewed on my OLED screen phone. But then I know where to look. 

Edited March 4, 2023 by rodeo_joe1

[gary green]

I think I see the word "congratulation" in most of the images.  I can't really discern it in the purple image.

I'm viewing on my LG G6 phone and have to zoom in to detect the subtle color variation.

Edited March 4, 2023 by gary green

[rodeo_joe1]

There's also the word 'saturation' and a large exclamation mark. But I fear the experiment has been completely screwed up by PN's re-coding or compressing of my original PNG files. 

I downloaded the yellow file above, only to find that the brightness difference had been reduced to one pixel level and the lettering had zero difference in saturation. Hmmff! 

Anyway, try it for yourself in PS or another editor. Create a rectangle of R180+G180+B0, then select an area and alter only the saturation to 89 or 90% in the HSL boxes. It's difficult, if not impossible to detect that 10 or 11% saturation change. In any colour you care to mix. 

Edited March 4, 2023 by rodeo_joe1

[paddler4]

I know little enough about color science that I don't know whether this is correct (Dog, where are you?), but I believe that while the horseshoe is supposed to encompass all humanly perceivable colors, it is NOT the case that all differences among colors within the horseshoe are expected to be humanly discernable. The fact that we can see color where you added text is consistent with the standard definition of the horseshoe.

[Glenn McCreery]

If I copy the aqua colored image and paste it in Photoshop, then use curves adjustments to greatly increase the contrast across the narrow tonal range that the letters appear, I can just make out the letters. If I then make a duplicate image and use a high-pass filter on the top image in overlay mode, I get, with further contrast adjustment, the attached image. I was unable to see anything visually in the posting on my computer screen.

Edited March 5, 2023 by Glenn McCreery

[rodeo_joe1]

Thanks for the replies so far. 

My real interest is in how the perimeter (locus) of the CIE horseshoe was originally arrived at, and in what objective units it's measured or defined. It presumably defines a limit of saturation. How is that saturation defined, and in what units?

The hue scale in wavelength is obvious, but again with no objective indication of how refined human differentiation of hue difference is. 

For example: I suspect that no-one can distinguish a difference in colour between the spectral Sodium D-lines, which are less than 1 nanometer different in wavelength. So how far apart in wavelength do two similar hues have to be before the 'average' human eye can distinguish them?

It's this lack of absolute parametric definition that I find curious about the horseshoe. Because without objective units it's just a doodle made up in 1931 - before Lasers, before LEDs, before cheap diffraction gratings, before dichroic filters. Even before the perfection of the Photo-multiplier tube. Basically before any of the tools we'd expect to currently use in colorimetry were available. 

So does it really define the limits of human vision, or is it just a doodled illustration of it?

P.S. I posted this here deliberately to avoid an obnoxious, patronising and insulting intrusion from a certain self-styled expert. Who often fails to provide clear answers.

[rodeo_joe1]

On 3/5/2023 at 3:03 PM, paddler4 said:

it is NOT the case that all differences among colors within the horseshoe are expected to be humanly discernable

Then what use is it? 

Because colour is a human construct. It does not exist outside of our perception (obviously the electromagnetic radiation that stimulates that perception is real, but concepts like 'red', 'orange', 'purple', etc. are a human construct placed on a very small region of the EM spectrum). Therefore any objective mapping of colour must be limited to what the human eye/brain can perceive. Otherwise it's not 'colour', but only something we can detect artificially through instrumentation. 

 

[Bill C]

On 3/4/2023 at 4:56 AM, rodeo_joe1 said:

I must admit to always being puzzled by the CIE 'horseshoe' as to what its perimeter boundary actually means. I always understood it to delineate the limit of human perception of colour - our ability to differentiate different hues or frequencies of colour and their saturation.

Hmm... where to start? I was in a similar place, 1990s, with something of a "need" to become knowledgeable about digital color management. Which I spent considerable time at, and can describe how to generate the chromaticity diagram (aka horseshoe or shark fin). Essentially you multiply the CIE color-matching functions times the spectral makeup of each sample (your laser's wavelength, for example) to get the 3 CIEXYZ values (X, Y, and Z). Which are then "normalized" somehow. Then the x value for the chart is a proportion of CIE X to the sum of CIE X, Y, and Z. And similar for the y value. You don't have to include the z value cuz you know it will all add up to 1 (x plus y plus z equal 1). But in case it's not obvious, by using only the proportional values, the luminance aspect is lost. In essence this is a way to put what oughta be 3-D onto a 2D graph.

That said, I don't think it's very useful for what I think your purposes might be. The ability of humans to distinguish between "colors" on the chromaticity diagram is highest near the lower left of the chart, turning into larger and larger ellipses moving away from it. So it is not "perceptually uniform." 

The CIE sort of took care of this issue in the 1970s with the invention of both CIELAB and CIELUV, which ARE (roughly) perceptually uniform. (These can be calculated from from CIEXYZ after being "normalized" and adapted to the proper light source.)

9 hours ago, rodeo_joe1 said:

So how far apart in wavelength do two similar hues have to be before the 'average' human eye can distinguish them?

If someone wants to determine if two colors can be distinguished by a "standard observer," meaning a more or less average person without color deficiencies, I would suggest evaluating them in CIELAB. This gives three numbers, called L*, a*, and b* (pronounce the '*' as "star"). L* is seen as a lightness measurement; a* is red vs green, and b* is yellow vs blue. If you see this as a 3D shape it is possible to calculate the distance between any two specific "colors." (In the industry lingo any calculated "distance" between "colors" is referred to as delta E.) The general rule is that a "just noticeable" difference in colors is about 1 unit. Now, I don't think this is a precise rule, rather just a rule of thumb to estimate how significant a color difference is. In the real world, average color "errors" of 2 or 3 delta E units would typically not be a problem even in very high quality color prints. 

Fwiw my use for these sort of things has been with respect to color photography, not in measuring the limits of human perception, etc. These are things that most photographers won't care about but if one spends much time making and troubleshooting color profiles, both for printers and digital cameras, a basic understanding of these things can be pretty helpful. 

 

[rodeo_joe1]

1 hour ago, Bill C said:

Essentially you multiply the CIE color-matching functions times the spectral makeup of each sample (your laser's wavelength, for example) to get the 3 CIEXYZ values (X, Y, and Z). Which are then "normalized" somehow.

That's all very well, but the normalisation and XYZ values are geared toward reproducing the mysterious horseshoe. A self-fulfilling prophesy. 

There are still no absolute units nor definition for colour saturation, except that which plonks it somewhere in the horseshoe. Clearly a pointless exercise if you're trying to determine the meaning of the horseshoe itself in some absolute terms. 

Luminosity is slightly irrelevant, except that it obviously affects the ability of the human eye to perceive and discriminate colour. 

It's also clear that a monochromatic Laser light must be 100% saturated, but still sits within that horseshoe? Illogical Captain!

[JDMvW]

People should know about this one too

Margaret Livingstone's Vision and Art: The Biology of Seeing. (LINK)

 

There is some cross-cultural evidence that color is not exactly a "human construct" so much as it is a by-product of the human eye/brain physical structure.

[Bill C]

4 hours ago, rodeo_joe1 said:

There are still no absolute units nor definition for colour saturation,

Well how COULD there be absolute units for something that is essentially a human perception?

15 hours ago, rodeo_joe1 said:

My real interest is in how the perimeter (locus) of the CIE horseshoe was originally arrived at, and in what objective units it's measured or defined. It presumably defines a limit of saturation. How is that saturation defined, and in what units?

Well, I just described, roughly, how the "horseshoe" was arrived at. It seems to me that you're trying to make it be something that it's not, by requiring it to have units of saturation or whatever. 

[mikemorrellNL]

On 3/4/2023 at 11:56 AM, rodeo_joe1 said:

...

Anyway The Experiment - 

Below are rectangles of 'pure' RGB colours, with words hidden in them. The words are in a desaturated and slightly different hue. The eyedropper tool in Photoshop easily picks up the difference - my eyesight, not so much. 

So the experiment is simple. Who can see the words? Or even detect the subtle colour variation?

I admit to totally failing to detect a difference, but then I doctored the text for that very purpose. However the saturation does vary by 11 percent according to Photoshop's HSL tool. 

...

I fail to detect any difference too. Both on my Laptop and on my phone.

[paddler4]

20 hours ago, rodeo_joe1 said:

Then what use is it? 

Because colour is a human construct. It does not exist outside of our perception (obviously the electromagnetic radiation that stimulates that perception is real, but concepts like 'red', 'orange', 'purple', etc. are a human construct placed on a very small region of the EM spectrum). Therefore any objective mapping of colour must be limited to what the human eye/brain can perceive. Otherwise it's not 'colour', but only something we can detect artificially through instrumentation. 

 

You can perceive the red color of a stoplight, if you aren't color blind. You many not be able to distinguish between two red stoplights that are very slightly different in hue (or, more likely, in mix of hues). Does that make your perception of red useless?