Ask HN: Measuring the Aesthetic Quality of a Set of Colors? | Hacker News Reader

Ask HN: Measuring the Aesthetic Quality of a Set of Colors?

18 pointsfp17 comments

For a side project I would like to automatically determine the aesthetic quality of a set of colors. I was wondering if there are metrics by which I could measure it.

For instance if I had these two 'color schemes':

  http://beta.dailycolorscheme.com/archive/2006/09/20
  http://beta.dailycolorscheme.com/archive/2006/09/19

could I objectively (for some value of objective) tell which one is 'better'? The only metrics I found on the net so far seem to be rather ad-hoc measurements of how well the set of colors is 'analogic' or 'complementary'. Since I want to write a program to make the measurements, a more structured apporach would help me.

Do you have any pointers or ideas for this problem, HN? Thank you!

View on HN

This is going to be a little more technical than I'd like, but it should be concrete enough to implement on a computer.

So, start with the color sphere: a unit sphere, and each point in the volume represents a color. Around the equater is arranged the hues, roygbiv. Moving north or south changes value, by adding white or black, respectively. Moving inwards toward the north-south axis reduces saturation; the exact center of the sphere is a perfectly neutral 50% gray.

Traditionally a set of colors is supposed to be 'harmonious' together if, when one constructs a polyhedron inside the color sphere, with each vertex being one of the chosen colors, the polytope formed has it's center of balance at the center of the sphere.

That is a 'designer' set of colors --- emotionally flat but visually pleasing. Emotional content is added by moving the centroid off center, charging the whole color set with a color 'cast'.

This definition, of course, fails when monochromatic schemes are chosen, but monochromatic schemes are almost always highly emotional unless a suitable accent color is chosen (usually a complementary, so as to move the COB to the center of the sphere).

For more information, read Itten, Keely, Albers, &c.

How can we know if this method of choosing colours results in colours 'aesthetically better' than, say, generating rgb values by flipping a coin?

Re: gray. Yes. I didn't say it was a medium gray, I said it was a neutral gray (neither warm nor cool). But you're absolutely right, this model is merely an approximation of perceptual color. You could instead build a complicated manifold representing perceptual color space, and define a criterion for a cloud of points in that manifold that resulted in pleasing colors, but it would be a lot more work than using the sphere, and you'd not gain that much, I don't think.

Re: how do we know? Well, mostly this system is based on the work of a number of thinkers, starting with Goethe and running through Munsell, Itten, Kandinsky, and a bunch of others, who spent a great deal of time putting colored squares next to each other. They discovered physiological effects of color, like simultaneous contrast. They attempted to learn about psychological effects of color. They admitted that, while some color combinations are attractive only to individuals, there were other sets that seem to have universal appeal. One of the things that those universal combinations had in common was that they centered themselves on the color wheel.

We know, more or less, by science. It is a largely empirical question (put some colors together, see if people like it). You can repeat the experiments yourself, if you like --- you asked 'how can we know'? Either take their word, or try it yourself.

And finally, in reply to the my use of the word 'emotional' --- if I were to say that colors whose center of balance is far from the center axis of the sphere are more surprising, and less comfortable, is that acceptable? I don't mean to say that red and green don't evoke different emotions than yellow and purple --- but they are both comfortable, obvious, familiar combinations; while, say, magenta and brown is less comfortable and takes more work to accept.

Less by science, then.

It seems like a particular combination of colours, say a particular red and green, may be a comfortable, obvious, familiar combination right now, in a specific culture - but 100 years ago may have been shocking and outrageous. And, who knows, in 100 hundred years may again be 'less comfortable'.

Goethe, Munsell, Itten, Albers, Kandinsky may have spent a great deal of time putting coloured squares next to each other, but none of them ever conducted anything like an objective scientific study on the 'aesthetic quality' of a palette. Goethe's theory was based heavily on the work of Aristotle. Munsell's views on aesthetics seemed laughably conservative at the time, let alone now. Neither Albers or Kandinsky, ever put foward a 'theory' for evaluating harmonious colours. Albers said that each colour 'goes' equally well with any other colour. Itten & Kandinsky were obsessed by the occult 'meanings' of colours. Certainly they were genius colourists.. but scientists?

Absolutely true. They were all mystics and goofballs. But I'm not justifying the mysticism. I never mentioned 'blond types' and 'black types', I didn't touch on the 'passion' of red, or the 'purity' of white.

I'm talking about the bits that seem to have held up after 100 years. The bits these men found from observation, rather than mysticism. This is, more or less, the method taught to design and art students today. You can see it in paintings from around the world, and going back hundreds of years. Simultaneous contrast exists and is physiological, not cultural. The vibration of complements is build into our visual system, too.

Saying that these people had some bad ideas doesn't make all their ideas wrong. We all have head about Newton's alchemy --- but we praise him for calculus and optics nevertheless. Kepler had his Harmonices Mundi, but we still can call him a scientist; because sometimes he was.

And if someone comes along and makes color theory even more scientific, that would be great! I would be overjoyed, because so much of what's been written about it is absolute bunk. However, at the moment, this is the best answer to the question I can give; and I think it works pretty well when compared to randomness.

50% gray isn't a medium gray, it is actually a fairly light gray. On an average PC monitor it is rgb(186,186,186) (assuming sRGB). Photographers often use 18% or 12.5% gray (0.18 or 0.125 albedo) as a medium gray depending on the tonal range they're working with -- these values are the geometric average of white and black in photographs with five and six stops of range, respectively.

Then there's the issue of print versus the web. In print, you can call a color neutral because it reflects light which is the same color as the illuminant. I'm skeptical about "neutral gray" on a website, viewed in a variety of different environments with a variety of different backlights. Fill your screen with pure white, look at it on a sunny day and then again late at night. You'll find it's shifted from a yellowish white to a bluish white.

I'm also skeptical about a mathematical formula for "emotionally flat colors". It seems I could choose a bright red, yellow, and blue and get them to center on neutral gray, and use that for my american hot dog stand. Then I could take a few shades of black and white, centered on neutral gray again, and use them to paint the corporate office. Neither seems emotionally flat to me.

First you need to understand the domain, then you can automate it:

http://en.wikipedia.org/wiki/Color_theory

Another challenge is what is it you want to achieve? Do you want people to say "ah, nice", or do you want them to "react" (click something for instance). Those two may be mutually exclusive in some instances, as generations of neon signage, "sales" banners and similar are living proof of.

http://colorschemedesigner.com lets you choose a main color and then generates a color scheme based on semi-objective criteria.

Hi there.

It's been quite a few years for me since I took color theory in college for my art degree. I think that you can automate color compliments with 2, 3, 4, or x colors by using "vectors" to compute the opposite hues... cutting the HSV model in halves for two colors, thirds for three, etc.

To get to the aesthetics, you'll have to use voting or surveys, and the reason is somewhat straightforward. Here are a few resons (not all):

The appeal of colors change over time with fashion and trends. In the US, mint green and canary yellow in the 1930s, white and cherry red in the 1940s, the list goes on for consumer goods and their colors. Think of vintage goods and their colors. Pantone produces color forecasts and tweaks their colors over time to reflect tastes.

Cultural differences. Color symbolism plays a large role - after all part of a designer's skill is selecting the right color for the specific context of an assignment. White for mourning in Asia != white for a wedding. One thought is that the region or location may be significant in the aesthetic judgment. If I remember correctly, the ArthurAnderson logo was orange because it was a color that could be used globally without negative meaning.

One last note, which comes from my training in a modernist driven school, is the form and the function of a design in black and while should be rock solid and work before adding color and the complexity... I think this was taught so that the novice students we were would learn the basics of typography, form, and balance before getting into colors.

As to the aesthetics: You can use the voting model like colorlovers (I hope that older votes are when tracking current color trends), use A/B testing if you are a web designer, or fall back to a company like Pantone to help you find current trends. Someone like google might be able to tie click paths to color schemes ... in other words data mine to establish what is a trend.

Unfortuantely aethetics are subjective - the best you can do is some machine-learning to recreate a "subjective" mind. As you've discovered, some combinations can be ruled out immediately on objective grounds, but deciding why one set of colours is "better" than another when they have similar properties is something that is not objectively understood. Even more difficult, the context in which you use the colours affects their "aesthetic quality".

This is very true - despite phrases like 'colour theory' aesthetic quality is not a matter of fact. Colour choices are not only subjective they are also highly context-dependent. You can completely transform the appearance of a set of colours by adding an accent in another and colour choices that are great for one purpose may be hopeless for another.

There is an existence proof "for some value of objective": Your and my own subjective tastes are also an objective determinant. With enough data from people, one might find some function over the domain of colour that yields pleasing combinations.

I'd suggest you put your candidates up on http://www.colourlovers.com/ and have them vote.

Aesthetics is a silly metric. We evolved color combination reactions to find food, like fruits in a forest, or to find healthy mates.

Every situation is different. A color palette like on kuler or colorlovers doesn't represent the real environment those colors will be found in. It doesn't include amounts, gradations and lighting, all things our brains evolved to process. For example we need to compensate for changes in lighting on a forest with fruits, because we'll be there at different times of day.

You could develop an objective aesthetic rating for a domain like picking fruits, based on what time of day should you pick a fruit to get the best taste, nutritional content.

Giving arbitrary ratings to color palettes has already been done on kuler and by color theorists, but when the reasons behind the ratings are not understood, you might as well use a neural net to replicate those ratings, using the palettes and ratings as training data. Black box either way.

I learned a lot with "Interaction of Color" http://www.amazon.com/Interaction-Color-Expanded-Josef-Alber...

Color Theory for Designers: http://www.smashingmagazine.com/2010/01/28/color-theory-for-...

Make a site that "learns" by asking designers.

Or just make a site that claims to, and then you have somewhere you can post your color schemes every time you need to test one.

You might be able to scrape color + ratings data from sites that offer downloadable themes / editor color schemes / etc. Or why not, a dataset of fine art paintings.

Reply on news.ycombinator.com