Compare CIEDE2000 and YIQ-based color metric

[arikon]

The code is here mapbox/pixelmatch#7

[arikon]

/cc @SevInf @j0tunn

[arikon]

/cc @sipayRT @SwinX

[SevInf]

Do we have any data we can use to compare the metrics quality?

[mourner]

Also check out this alternative http://www.compuphase.com/cmetric.htm. I'm also eager to do a comparison, but perceptual differences are a very individual thing so hard to measure. Performance difference is astounding though (CIEDE2000 is really really slow to calculate).

[SevInf]

@mourner @arikon maybe we could together come up with a dataset we can use for automatic evaluation of the metrics? What we need is a fair number of similar images which we consider perceptually the same and perceptually different.
@arikon @j0tunn @SwinX will it possible to get this images from Yandex repositories somehow? You have the longest history of doing visual regression testing among us all.

[mourner]

The tricky thing is that perceptual color differences can't be evaluated automatically, — they can only be judged by a human. :) Today I put up this fun viz to play with 4 different metrics: http://bl.ocks.org/mourner/8a3968f74da743a754af

I'm also thinking about making a little quiz app that picks a random color and shows the closest one to it according to each metric, and then asks you to pick the best match (not telling which is which). And saves stats of how often you pick each metric in localStorage. :)