unity-design team mailing list archive

[Matt Richardson <m.richardson.1990@xxxxxxxxxxxxx>]

Thanks for that insight! I thought I had read somewhere that it just 
matches the centre pixel, so this is really useful.
I agree that this is broken at the moment. As you say, it gives extra 
weighting to diagonal lines (not good news for potential patriots with a 
diagonal cross flag haha), which may not be representative of the 
background. I'd be interested to have a look to see how GIMP works when 
it indexes images down to two colours (two is the minimum it can do). 
Perhaps we could use something similar?
For now I will use the existing algorithm such that the theme tries to 
match the colour chosen by Unity

Thanks for the heads up!
Matt


On 22/05/12 19:28, Alan Bell wrote:
> I believe the current chameleonic algorithm is a bit broken. I was 
> trying to trace what it actually does, it looks at the 4 corners of 
> the screen and the center spot, then if there is much of a difference 
> it recursively does this for each quadrant in turn until the points 
> are roughly even in colour. This sounds good in theory, and works 
> reasonably well in practice, but it overemphasises a diagonal cross in 
> the center of the screen and basically diagonal lines growing from 
> each point it looks at. There is probably a much better algorithm that 
> could be dropped in, perhaps just one sampling 16 random points or 
> points on a grid would be just as good in effect and would run in a 
> shorter, constant, time. Currently if the center point and 4 corners 
> match then it stops very fast, but doesn't see any variations 
> elsewhere in the image, for other images it will recursively average 
> quite a lot of points.
>
> Alan.