Wednesday, November 18, 2009

A color never comes alone

This is a motto I was using 25 years ago. At that time I was working on VLSI design automation tools at Xerox PARC, more specifically on design rule checkers for full custom CMOS. The designers were doing so many layout errors that I could not understand how a top notch designer could do them. I had the suspicion that some of the designers could have a color vision deficiency — and 20 years later I discovered one of them is a dichromat — but that was not explaining the the type and volume of errors. I decided to investigate.

I visited several designers and asked them to explain me their layouts and how they design them. To show me all the details, they had to move around pieces of the cells, because they had a hard time figuring out the topology. In my naivité of a Young Turk, I decided there must be some order in the set of colors and one should be able to sort them so that the layout becomes very readable.

I was naive, because the color scientists simply handed me a copy of Wyszecki and Stiles telling me "just read it and you will know everything you have to know." Fortunately I had studied physics and was actually able to read and understand a big chunk of the book, enough to solve the VLSI layout problem — the design error rate dropped to almost zero — and to switch my research area to color reproduction.

This experience taught me to never look at individual colors and to always consider color palettes. For example, I did a plug-in for an illustrator called Gargoyle that would allow you to preview the output on a variety of printers (at that time we were using dry and liquid xerography as well as thermal transfer and ink jet). In a small panel I was displaying a metric chromaticity diagram with a point for each color in the illustration. You could now drag each point and see the simulated output in real time. Although each printer technology had a very different gamut, you could always make your illustration look good because you could edit the entire palette in one swoop: a color never comes alone.

When printers became a mass product, the first color experts like Brigitte Ruetz at Canon and Jay Gondek at HP took the holistic approach of optimizing the color rendering table so that all or most images would print well for all colors simultaneously. It was only later, when the graphic arts adopted digital color management and required a large standardized set of tables that ICC profiles were invented and adopted. In the ICC approach, each color is optimized individually to reduce the ∆E error, there is no holistic approach of mapping the color palette in an image.

Why is this history relevant today?

I was just looking up the now forgotten literature of those early days and came across this interesting paper that is very relevant to what we are playing around with today: James Shilts Boster, Can individuals recapitulate the evolutionary development of color lexicons?, Ethnology 25 (1986), no. 1, 61–74.

There, at the end of the paper Boster refers to a 1984 paper by Kay and Kempton with a similar result to his namely that when informants are shown diads of color swatches, their response patterns suggest that they respond according to the perceptual distances between the chips. However, when they are shown a triad of chips, they overwhelmingly overrode perceptual affinities (as measured by minimal perceptible differences) in favor of categorical ones.

The chips in a diad can be the original color and the reproduced color. When mapping color the ICC way, the correct paradigm is colorimetry and JNDs. However, in holistic color reproduction, a global metric like the lexical metrics induced by categorical perception are the better approach. They lead then to rules like "in gamut mapping a color is not allowed to cross a name boundary."