somewhere to talk about random ideas and projects like everyone else

stuff by antimatter15

Hi. I’m Kevin Kwok, and I’m currently a junior at MIT, and this is my obligatory online presence or whatever.

This is my little corner of the internet, where I talk about random projects and ideas and stuff.

If you want to contact me or read obnoxiously nostalgic autobiographical trivia, check out the about section.


Archive

Tags

3D actionscript ajax ajax animator api apple arduino art awesome blog browser bugfix canvas change chrome code color colors css design doge elitist evil experiment export extension ExtJS fast features firefox flash format function future GAE github google google code google wave graphics hacking hardware hosting html5 idea image internet exploder ipad javascript json library light lines linux login magical meta microsoft microwave mirror mit multitouch new nostalgia ocr offline old onlypaths pi progress prototype python raphael release rewrite school security shiny ShinyTouch simple site small speed SVG touch update upload user interface vector vectoreditor version video vision vp8 vX JS wave webcam wikify wikipedia wordpress

Perceptual color distance and RGB and L*a*b 31 May 2014

CIEsmic-Inflation

I was doing some experiments with clustering colors, and it turns out that RGB euclidean distance is woefully inadequate, so here’s a neat little visualization of the colors in an certain image (an obligatory cat) in Lab space (specifically the axes represent L and a).

So it turns out that comparing the difference between RGB colors is actually a really shitty way to determine color distance. There are colors which have similar RGB representations while being perceptually distant, and perceptually similar colors with vastly different representations.

So I decided to run a series of large scale double-blind (heh) experiments on human perception, analyzing the output with a dizzying array of statistical techniques and— oh wait, no I didn’t, because scientists have gotten that all figured out since the late ‘70s.

I was doing some experiments with clustering colors, and it turns out that RGB euclidean distance is woefully inadequate, so here

That group of color scientists, the International Commission on Illumination (CIE, acronyms are a bit weird when they’re international, and by international, I mean French), have been working tirelessly since the early 20th century to create increasingly nuanced mathematical models of color perception.

Rather than give up and use tables (of a slightly different sort), they’ve elected to add increasingly nuanced corrections to the originally elegant euclidean distance metric. Somewhere on that trek, they’ve actually abandoned the notion of actually being a metric- the DeltaE function, which defines how to calculate the perceptual distance between any two colors, is actually only a quasimetric— that is, it isn’t guaranteed that the distance from a point A to a point B is the same as the distance from point B to point A.

I originally wrote this blog post to be parody of all the buzz over the discover of evidence for Alan Guth’s cosmic inflationary theory (hence the diagram), but I kind of waited a while for the content to ferment (albeit not the decades requisite to be worthy of a Nobel).

So here’s a little library for converting between L*a*b* and sRGB, as well as computing DeltaE. There’s probably stuff like configuring the white point that I haven’t adequately considered, but this was good enough for my purposes of filtering colors for Project Naptha.


« Archives