Comment by caseyross | Hacker News Reader

Comment by caseyross | Hacker News Reader

I think for people like us engineers who don't spend a lot of time developing our "visual sense", it can be hard to understand what the big deal is about painting and other visual arts.

For me the breakthrough was when I realized that the colors I think I see in the world are all lies. Try this: look at something near you that's a uniform color. Looks like one color, right? Wrong! Light hits the object, and bounces off in a million different ways depending on physics. Your brain interprets these light rays as a single composite color for each object, but it's not the physical truth - it's just a simplification to deal with overwhelming detail. If you want proof of this, just look at what a digital camera sees. Pixels that the brain would consider the same color... aren't.

A big part of the artist's journey, perhaps the main part, is in developing the ability to break through the brain's natural conceptual abstraction layer and see the world much closer to the sensory raw data that normally goes unnoticed. Ironically, the better an artist manages to reproduce these low-level details, the more their work will seem obvious and like something anyone could do!

Oh the colors are even more lies than you are letting on here.

Every object reflects different frequencies of light to different degrees, as I imagine you know. If you think of the range of frequencies in the visible light spectrum, you can construct for any object the “spectral power distribution” (SPD) that says how much light is reflected for a given frequency. So far so good.

But we don’t see “frequencies” of light. An object that reflects only yellow and an object that reflects red and green but no yellow will look the same to us, despite the fact that they are reflecting completely different light, thanks to the way humans perceive light. In fact, there is an infinite variety of different SPDs that to humans look like exactly the same color — called metamers — for any given color we perceive.

A color is an equivalence class, not a point on a spectrum.

Adding to this (I'm sure you are aware), that it is a bit more complex in reality. The SPD of a non emitting medium depends on the reflectivity (as function of wavelength) of the medium and the SPD of the illuminating light source. So two materials with different reflectivities would be metamers (appear the same color) with one light source but not for a second, while if they had the same reflectivity they would appear the same color for every light source.

Absolutely! Thanks for pointing this out — it’s in fact one issue with light sources that are not full-spectrum, like fluorescents and (even “white”) LEDs. They can make objects appear to be different colors!

Ooooh DAMN. Interesting idea. That part that pisses me off is that I was very familiar with all the concepts that you used, but I doubt I ever came close to the conclusion.

Ok, I concur, I was probably not aware that color mixing is a human physiology thing and not a physics thing.

So to clarify, when we mix red and green playdough as kids, what happens is all the little particulates/molecules interweave together, but each is reflecting a different signal. And when that variation/entropy back-and-forth between red and green is too high, the brain just melds them to yellow?

> And when that variation/entropy back-and-forth between red and green is too high, the brain just melds them to yellow?

Not “melds them”: it is yellow. A specific yellow, to a human, is any of the SPDs that triggers the same amount of “red” and “green” cone activity. It’s not some kind of brain interpretation trick, it’s literally the same signals reported by your eyes.

If my interpretation is correct:

human_eyes.input(emit_light(580))

-> ((0.5, 0.5, 0), "Pure Yellow")

human_eyes.input(emit_light([[650, 0.3456], [550, 0.4567], [450, 0.0]]))

-> ((0.5, 0.5, 0), "Pure Yellow")

human_eyes.input(emit_light([[670, 0.00625], [640, 0.03125], ... [450, 0.0]]))

-> ((0.5, 0.5, 0), "Pure Yellow")

... because input is a list but output is a weighted average in a single 3D vector.

Experienced a feeling of awe, thinking through the ramifications of this. Thanks for sharing zamfi!

And in addition, the classes are of different sizes. There is a such thing as yellow light, and red and green looks like it. There is no such thing as magenta light -- light is a linear spectrum, not a circle -- and yet red and purple together look like this nonexistent color. We see a continuity where there's a hard break.

Be very careful. Digital cameras have a lot of noise that has nothing to do with the actual reality. Using that as your barometer is likely to mislead unless you’ve got a very good camera (or just use a microscope).

And they are not good at certain kinds of colour subtlety -- even the Foveon sensors struggle with the colours of dim, diffuse light (sunsets etc.) and deep muted colours (the grey-green of ivy leaves for example).

Yep. I was an artist before I became a programmer, and one thing we were taught is that we have to learn to draw/paint from direct observation rather than from photographs. Cameras reduce so much from the real world.

one of my favorite art exercises in school when I was first learning to paint with physical paints, we took a canvas, covered it in roughly 1 inch grid lines and then had to left to right top to bottom paint a scene by first mixing the correct color and simply painting that 1 inch square, it REALLY teaches you how light bounces around a scene, reflects through objects and so on... the other thing that really got me into learning how light works funny enough was moving on to 3d art! being able to isolate a single sphere in a room, with accurate light simulation and light let me really hone in that sense. (and yeah colors are VERY psychological as well, you percieve them very differently based on what they are near for exmple)

What you describe is about technique, and can already be done by a computer.

Reply on news.ycombinator.com