4 ms·
Yeah. I wrote the long one of those. What you should use for best results is something like CIELAB lightness. Basically the steps are: 1. Take "gamma-compresse
by jacobolus 16y ago
Yeah. I wrote the long one of those. What you should use for best results is something like CIELAB lightness. Basically the steps are:
1. Take "gamma-compressed" RGB and scale to the range [0, 1] (if it's an 8-bit integer representation, divide by 255), and convert to linear RGB: undo the sRGB nonlinearity or for the lazy, take each component to the 2.2 power.
2. Take a weighted sum of R, G, B components. The weights for an sRGB/HDTV image are .2126, .7152, and .0722 for R, G, and B respectively. This gives you the luminance, Y.
3. Apply some non-linearity to the luminance to get a decent correlate of human-perceived lightness. A gamma curve w/ 2.2 gamma (that is, raise your value of Y to the (1/2.2) power) actually is an okay function to use. If you want a better one, feel free to use the CIELAB definition: L = 116 Y^(1/3) - 16, where L ranges from 0 to 100.
4. Anything with L >= 50, use black, anything with L < 50, use white.
(properly L*, but hacker news will think I'm using the asterisk to indicate emphasis if there is more than one of them)
Since this is likely only being calculated for a few colors (even up to dozens or hundreds), there's really not enough overhead in doing this math to worry about the speed.
- nitrogen 16y agoI found the HSV/HSL article immensely useful on a semi-recent project, as Google seems to have forgotten about the original references I used when I first learned about those two colorspaces (i.e. students' and professors' personal web sites, found using the Altavista search engine).
- jacobolus 16y agoActually, now that I think about it, if the only goal is to decide on black or white text, steps 3–4 aren't really necessary. Just undo the gamma compression (1), take a weighted sum or linear RGB intensities (2), and then pick a cut-off for Y of 18% or 18.4% (or whatever you prefer), about at mid gray.