This is a quick writeup of some work in progress that might never actually progress any further… so I thought I’d put it out here just in case it’s interesting to anyone.
If you think of colour as computer people do, you probably think of it as red, green and blue values. In digital images, colours go from 0 to 255 in the red, green and blue channels, with 0,0,0 being black, and 255,255,255 being white. 255,0,0 is bright red. 255,255,0 is yellow. And so on. We can think of colour this way making up the colour cube: here’s a video I have put together which provides a visualisation of that. You start off with black top left, and end up with white bottom right, and you can think of it as being a cube where you move through the cube over time (although I’m not sure that’s a particularly clear explanation…).
That video includes every colour that it’s possible to have in an ordinary 24 bit (“true colour”) digital image.
Different types of image use different parts of the colour cube. We can build a statistical model – specifically a Gaussian Mixture Model – which learns the way in which colours are distributed in a set of images. Feeding this kind of model a set of photographs of landscapes downloaded from Flickr, we can then re-colour the colour cube with the mean colours of the elements of the mixture. This gives some kind of idea of what landscape photographs look like, statistically speaking.
In the next video clip there’s the same animation/visualisation, but this time for the paintings of Kyffin Williams:
And here’s David Hockney:
I’m not quite sure where I’m going to go with this yet, but I think the videos look kinda cool.