Using the color wheel to explain why purple is close to red is circular logic. The color wheel is an abstract human construction that places red near purple arbitrarily. The color wheel does not prove that red is fundamentally close to purple. If you’re stuck on color wheels and the kindergarten explanation, then I understand why my questions aren’t making sense to you, and I’m happy to leave it there.

I’m not sure why you’re jumping to erroneous conclusions about my background, or why it matters to you. But since you asked, and in the hopes that maybe it could get this conversation back on a productive course, I’ve studied and used color academically and professionally my entire life, including human vision and perception, and computer graphics. I’ve written a few papers with well known color and perception experts on the topic of novel perceptually uniform color spaces, for example. But I wouldn’t say I’m a color or perception expert myself, because the topic is surprisingly deep and tricky, it bears little resemblance to kindergarten knowledge of color wheels. I’ve written high dynamic range spectral renderers in the past, studied spatial and temporal tone mapping, written color conversion code from spectral to CIE XYZ, xyY, Lab, sRGB and other spaces, and like you I’ve written programs and articles about transparency. Premultiplied alpha is one of my favorite topics when it comes to alpha channels.

Your JavaScript auto-masking prototype looks really nice. The video didn’t really say much about color, but you may be onto something if you’re pulling your mattes fully automatically. Have you released it or sold it?

Since you asked, I’m very familiar with the Munsell space and RGB. RGB has been practical and useful for a long time, but it’s just about one of the worst color spaces there is. This is why people who care about color specify sRGB rather than RGB, and why people who work in CG production and people who make LCD displays don’t use “RGB” anymore. The three biggest problems with RGB are lack of standardization (“RGB” isn’t well defined), RGB is a very perceptually non-uniform color space, and RGB is a crappy and relatively unintuitive space for artists (HSV & HSB are better, but researchers and color UX people have designed much better spaces & pickers for artists).

Okay, so, with all that in mind, maybe you’d care to take a more careful stab at explaining why you think purple is “close” to red, and what - exactly - that means?

> Using the color wheel to explain why purple is close to red is circular logic. The color wheel is an abstract human construction that places red near purple arbitrarily. The color wheel does not prove that red is fundamentally close to purple. If you’re stuck on color wheels and the kindergarten explanation, then I understand why my questions aren’t making sense to you, and I’m happy to leave it there.

The specific claim is that either red is close to purple or that all these models are showing things wrong, which is not circular logic.

You were invited to claim the models are wrong if you want. Do you want to?

Which models are you asking about, do you mean RGB & the color wheel? I feel like I just did explain my position in depth on both, did you miss it? The color wheel is an abstraction and RGB is useful but flawed. It would be silly to call either one “wrong”, but neither one was invented to chart wavelength or human cone response. Rob’s invitation wasn’t particularly genuine, nor is yours, the question is an attempted trap and his was a failed attempt to appeal to authority.

In any case, to stand by your claim that either red is close to purple or that all these models are wrong, you must first define what red being close to purple means. Rob couldn’t do it, can you?

Rob has now defined the proximity between red and purple as what’s on the color wheel. It absolutely is circular logic, because it’s a fact that the color wheel is not a physically accurate representation of color, and Rob already knows that. Purple is the only named color on the color wheel that is not a color of the rainbow, the only one that is a mix of two mono-wavelength colors. The color wheel is not naturally a circle, purple’s a synthetic addition, used for the purpose of making the colors of the rainbow into a circle by tying the ends, red and blue, together. Rob knows all this because he’s made diagrams of it that call out this fact and posted them to Quora.

This doesn’t seem to be particularly helpful here. Do you have an opinion on the left-hump of the red response curve? Do you believe Rob that it has nothing to do with the perception of purple, or any perceived closeness between purple and violet?

> The color wheel is not naturally a circle, purple’s a synthetic addition, used for the purpose of making the colors of the rainbow into a circle by tying the ends, red and blue, together.

It's a natural consequence of seeing what happens when you mix different colors. Magenta-purple is the only synthetic color that exists, and it's adjacent to both blue and red. And then you can ask people how far apart different hues are, and you get results consistent with a circle with a certain spacing.

> In any case, to stand by your claim that either red is close to purple or that all these models are wrong, you must first define what red being close to purple means. Rob couldn’t do it, can you?

All these entirely different ways of laying out the colors in the article seem to put purple and red a lot closer than green and red.

And you have the observational charts like https://www.handprint.com/HP/WCL/IMG/shepard.gif

> This doesn’t seem to be particularly helpful here. Do you have an opinion on the left-hump of the red response curve? Do you believe Rob that it has nothing to do with the perception of purple, or any perceived closeness between purple and violet?

He didn't say it has nothing to do with it, he said it's "insignificant" for a lot of the closeness to red.

For my point of view, if the opponent process is relatively accurate for how vision works, then the signal from the blue cones is likely factored in to the "red/green" signal and tilts it red at very short wavelengths independent from any bump coming from the red cone itself.

> Magenta-purple is the only synthetic color that exists, and it’s adjacent to both blue and red.

Exactly right. It has to be adjacent to both blue and red, because it’s the middle of the gradient mixture of blue and red. We don’t have any other mixture gradients on the color wheel. Do you see the problem yet? What if we did have other mixture gradients?

> He didn’t say it has nothing to do with it

Hmm, that’s not my read, it was stated multiple times and quite clearly and flatly that the left hump is “not” the reason, and that nothing changes if it didn’t exist. The top of the thread declared unambiguous disagreement.

> All these entirely different ways of laying out the colors in the article seem to put purple and red a lot closer than green and red.

These are abstract organizational tools you’re referring to, they do not prove anything about the proximity of purple and red. On the color wheel, red and blue are placed just as far apart as red and green (even though we know they are different perceptual distances). Since purple is a red-blue mix, and therefore in between the two at the midpoint, then it must be closer to red than green is to red, by color wheel definition. So your point about green is therefore tautological and uninteresting, if we’re talking about color wheels.

BTW, there’s a pretty important subtlety you seem to have missed in my argument. I am not claiming that purple and red aren’t somehow close. The actual claim of this thread being debated is that left hump is not involved (splitting hairs between zero and insignificant is irrelevant in my view). I’m asking for a definition of closeness, a metric that proves it, because I believe that whether the left-hump matters may hinge on such a definition. Rob’s argument depends on how you define closeness. He is right within his definition of closeness based on the color wheel, red is close to purple, the problem is the circular logic of this definition, it means the question of whether red is close to purple is a basically meaningless question, another uninteresting tautology that dodges explaining why red and purple might seem close (or more importantly why red and violet might seem close.)

> if the opponent process is relatively accurate for how vision works, then the signal from the blue cones is likely factored in to the “red/green” signal and tilt is it red at very short wavelengths independent from any bump coming from the red cone itself.

Interesting thought! I’ll have to think about it. What if it’s not independent, but two sides of the same effect? Maybe you’re right, and the left hump is caused by the mechanisms people call opponent process. This still begs the question even more acutely of how you would define color proximity. Opponent process color theory points out that we can see completely non-physical colors, colors that don’t exist in nature and cannot be displayed, like red-blue. How far apart is opponent red-blue from the purple that is red+blue mix?

My reading is that Rob’s argument tosses your opponent process theory out as well, right? He already claimed multiple times that non-zero perception of red in blue colors is not the reason that purple is close to red. That’s the problem I have with Rob’s argument, that he ruled out perception of red from blue colors, and just cited the color wheel, without even stopping to ponder the question of whether the color wheel is the way it is because of the left hump.