Why does the colour wheel exist?

[sophiecentaur]

I can't think of a good evolutionary reason for my hypothesized red-bump-in-the-blue which supposedly supplies color discrimination into the ultra-violet. If it really exists it might be a chemical accident in the red cone. But a nice accidental color none-the-less...

I believe we are most sensitive to shades of green, supposedly to distinguish food sources. Any red gradation sensitivity for facial recognition would have to be a much later development since a) most faces were not "pink-white" to start with; and, b) the social necessity of mood recognition is also a recent thing. Chimps for instance are all about body-language, not subtle facial expression.

But this is all off topic, no? Can we hijack the thread?

Did you not read the title of the graph with that bump on it? It is not a sensitivity curve. It shows the combinations of primaries that can be used to give a MATCH for a given colour. Have you sussed out the difference?
IMHO that bump is there because we just haven't the ability to distinguish well on the fringe of short wave sensitivity.

 

[schip666!]

Did you not read the title of the graph with that bump on it? It is not a sensitivity curve. It shows the combinations of primaries that can be used to give a MATCH for a given colour. Have you sussed out the difference?
IMHO that bump is there because we just haven't the ability to distinguish well on the fringe of short wave sensitivity.

I don't understand why we are looping on this. I admitted that I figured out the difference and the reasoning behind the CIE graph bump in a long-ago post. However, if you look at the tabular sensitivity data in the previously ref'ed macaque paper you may find that there _is_ a slight increase in sensitivity in the high-blue region for the red cone. In macaques. I don't know for humans. _ALSO_ I believe the two graphs upon which we are cycling, the wiki color-vision and the CIE-color-matching, have different vertical scales where the CIE is log. The log scale would accentuate small differences at the bottom of the range.

The only slightly reasonable way to settle this is to find data for human red cone SENSITIVITY and see if it has a high-blue feature similar to the macaque data. And the only reference I've seen posted here which -- might -- have such data is behind Nature's subscription wall. So...I'm going to try to prevail upon some sciency-friends to hijack that paper and have a look for myself.

In other news, you could be right about the chrominance vs luminance issue in human skin, but I would still argue that the ability to distinguish mood via skin color is newer than could be explained by biological evolution. It might be the other way around...

 

[sophiecentaur]

OK, we do seem to be looping on this but I really couldn't see how the colour sensitivity of some monkey eye cells should have any detailed relevance to the "colour wheel" and human perception - when we already have data both on human perception and matching. Colour TV and film were studied in great detail and a lot of real money was spent in the commercial enterprise of colour reproduction. With respect, the monkey data represents a very small additional contribution (with the greatest respect of course; academic work is always potentially valuable) and would only be of really significant if there were to be an equivalent dertailed set of data on colour matching for monkeys. Bearing in mind that our perception is limited to around 8bits and that a 'just noticeable difference' on the chromaticity chart is around 2%, we don't really need more accurate data in order to put this to bed. As long as we agree that the 'red hump' on Wiki doesn't conflict with the 'non-red hump' data then I can let it rest. IF some monkeys / bees / pit vipers happen to have sensitivity peaks outside or inside the human range, I don't think it needs to interfere with the colour wheel thing. Unfortunate that such a naive term like colour wheel should be at the top of this thread - it should be on an Art / Painting forum - but, fair do's, it got the topic going.

I could rest easy if I thought that people were starting to realize that the colours of monochromatic light are a tiny fraction of what we actually see and that a true, saturated, (monochromatic) source is totally outside the experience of all of us who do not work in a spectrometry Lab. These "pure reds" that people say we see every day are really very impure because they originate from pigments. TV screens or interference filters - all extremely crude sources, with either a broad band spectrum or loads of white light diluting them.

Hanging on to the words "red orange yellow green blue indigo violet" in a discussion about colourimetry is the equivalent to saying "nature abhors a vacuum" when discussing the gas laws.

 

[schip666!]

OK, we do seem to be looping on this but I really couldn't see how the colour sensitivity of some monkey eye cells should have any detailed relevance...

You are absolutely right. I don't care about monkeys either...but it's the only data I've been able to access.

My point, which may be minor or moot, is this: to distinguish spectral-violet as a different "color" from blue we need to get some kind of additional sensory information. This could be just the declining ratios of red and green -- but that implies to me that we could "fake" a sensation of spectral violet by adding a bit of "pure" red and green to a mostly blue signal.

Since we can get a sensation of magenta, which is kinda like violet but not really (at least to my color memory since I haven't seen any good violets lately), by mixing only red with blue it would seem ockhams-razorish that the sensation of violet also has just a bit of red-signal in the mix. Thus my hypothesized and slightly proven-in-monkeys red-bump-in-the-blue argument.

Looking at the Photoshop magentas pallet -- which don't seem to match a nice violet -- somewhat convinces me that there is something else going on, but I would need a good spectral-violet signal for comparison.

Also the easily thrown number of colors distinguishable by the human eye is around 1 million, so the sensors have less than 8 bits of resolution and are probably logarithmic in response...Which is another thing that has always puzzled me: aren't "CRT" displays by nature linear, i.e., twice as much signal makes twice as much light? So how do they get matched to the eye? Are the 1 million stretched over the "extra" 15M colors in an 8+8+8 bit representation by mapping somehow?

Ah well...

 

[sophiecentaur]

You are absolutely right. I don't care about monkeys either...but it's the only data I've been able to access.

My point, which may be minor or moot, is this: to distinguish spectral-violet as a different "color" from blue we need to get some kind of additional sensory information. This could be just the declining ratios of red and green -- but that implies to me that we could "fake" a sensation of spectral violet by adding a bit of "pure" red and green to a mostly blue signal.

R,Gand B signals are not the outputs from your sensors. You need to get your terms defined properly if you want to get this straight. RGB are weightings of three standard primaries for a match with the colour you want to produce. There is loads of data showing the sensitivity of the eye receptors to the spectral colours - wiki shows some of these HML figures on that graph.

Since we can get a sensation of magenta, which is kinda like violet but not really (at least to my color memory since I haven't seen any good violets lately), by mixing only red with blue it would seem ockhams-razorish that the sensation of violet also has just a bit of red-signal in the mix. Thus my hypothesized and slightly proven-in-monkeys red-bump-in-the-blue argument.

The violet that we can actually produce with RGB signals is, indeed, inside the primary triangle and does involve red. Without some red and green, all you would get would be the primary blue (0,0,255). With equal R and B (255,0,255), you get saturated Magenta and adding some G and a bit less than equal R, you get violet(ish) colours, say (100,50 ,255).
All this is not helped by the fact that the primaries are not monochromatic, afaik, so the 'colour' of the blue primary could not actually lie exactly on the curved portion of the CIE chart - it would have to lie inside.

Looking at the Photoshop magentas pallet -- which don't seem to match a nice violet -- somewhat convinces me that there is something else going on, but I would need a good spectral-violet signal for comparison.

How many more times do I have to make this point? The Photoshop pallette consists colours that already have been synethesised with primaries. How could you expect it to have any real validity for colours outside the triangle?

Also the easily thrown number of colors distinguishable by the human eye is around 1 million, so the sensors have less than 8 bits of resolution and are probably logarithmic in response...Which is another thing that has always puzzled me: aren't "CRT" displays by nature linear, i.e., twice as much signal makes twice as much light? So how do they get matched to the eye? Are the 1 million stretched over the "extra" 15M colors in an 8+8+8 bit representation by mapping somehow?

Ah well...

The linearity is, indeed a problem and limits the exposure range of digital cameras. TV pictures and prints have a very limited contrast ratio. A contrasty scene can reveal all sorts of detail to the eye of the person there but the low level stuff is way down in the quantising noise because you are only dealing with a few bits of resolution.

 

[I like Serena]

R, G and B signals are not the outputs from your sensors.

I'd say the perceptual output of our sensors is spectrum red, spectrum violet, and something undefined in the middle.

That is, if only the sensor for red is stimulated, we'd perceive it as red near infrared.
And if only the sensor for blue is stimulated, we'd perceive it as spectrum violet (unless there is a red bump doing something).
But if only the sensor for green is stimulated, we do not know what we would see, since it cannot happen (not without some serious surgery ).

Any combination would be processed by our neural network to be perceived as some color modeled by the CIE diagram.

 

[sophiecentaur]

Aren't you assuming three narrow band analysis curves here? In the extreme, if there were none of that essential overlap, you'd only be aware of three possible colours.