It started with a idea of writing a posterizer – or in this case a ‘Obamizer’:
Main principle is to compare each color’s distance in rgb-space to some predefined colors – it’s like doing a voronoi diagram, but in color space, and things remain easy while dealing with only a few colors.
xy-voronoi :
I had already earlier though of what if I extend the voronoi algorithm so it uses the x- and y-coordinates, but considers r-,g- and b-values as well, like a 5-dimensional distance Δx²+Δy²+Δr²+Δg²+Δb² ? Anyway,here’s some – in my opinion – quite interesting results I came up with, xyrgb-voronois with random points and a regular grid:
Different weights for colors: d²=x²+y²+f(r²+g²+b²) with f= 4.0 and f=-0.25, the last one seems to seach nearby area for a very different color on each pixel, creating like some partial negative:
‘Obamizer’ was written with Pixel Bender, other images with actionscript – I used a very brute-force code. Not like for instance Frank’s voronoi processor, mine is so far more like 1/15 fps !
… an article of colour metrics with a more scientific approach.
that last .25 thing stands out for some reason. maybe because trees aren’t black.
By: makc3d on February 27, 2009
at 6:59 pm
The thing about the last one is that it’s a very strange metric (almost Minkowski). Two points far away can still be at distance 0. In fact, an infinite number of points can be at distance 0 from each other. I’m not sure a pixel-based algorithm will give you the true Voronoi diagram, but it’s sure is interesting
By: Frederik on March 6, 2009
at 1:21 pm
really interesting, i think you might want to couple it with some AA of some sorts. I think it could be exploited a bit further.
By: peutichat on March 19, 2010
at 2:23 pm