Color is a sensation produced by the human eye and nervous system. It is related to light, but an understanding of the properties of light is not sufficient to understand color, and is especially not sufficient to understand the art of color reproduction. Overwhelming experimental evidence tells us that the perception of a color is related to the strength of three signals which are transmitted along the optic nerve to the brain. The importance of this is that:

It is useful to represent a color by a set of exactly three numbers.

In practice, the set of three numbers must be related to some actual color reproduction process. The numbers commonly specify portions of some set of primary colors such as:

As everone learned in grade school, mixing varying portions of primary colors enables one to perform color reproduction. There are two big problems with using numbers that correspond to these primary colors to communicate about color.

My magenta may be a little different from your magenta.

The specification 20% Cyan, 60% Magenta, 47% Yellow fully specifies a color only if everyone uses inks that are of the same color. Offset printers have acheived this (to some degree) for many years by insisting that everyone use the same "process inks". Too bad the makers of digital printers couldn't do the same, but hey - process inks are pretty restrictive anyway.

A little shift in yellow makes a bigger difference in color than a little shift in magenta.

In the above mentioned color; 20% Cyan, 60% Magenta, 47% Yellow, a change of a few percent in Yellow will have a much bigger affect on the perceived color than a similar change in Cyan. This is true in general, and makes it very tough for a computer program to figure out what is and isn't important about a color specification. Color scientists say that the color model (space) is not "perceptually uniform".

It would be very useful, especially when computers need to determine the differences between colors, if we could measure colors in such a way that the "distance" between two sets of numbers corresponded to the perceived difference between the colors.

A computer could then make better decisions about the "goodness" of color reproduction.

Fortunately, such a "scientific" system of color measurment was devised way back in the 1930's, the CIE color system. These numbers can be measured by instruments called "colorimeters" (as opposed to "densitometers"). The famous CIE "chromaticity diagram" illustrates some relationships between colors in this "color space" and is illustrated below.

The real magic of CIE numbers goes beyond the "perceptual uniformity" issue. An even bigger advantage is that this was defined decades ago and nobody owns it so everybody can agree on it. Makers of scanners and makers of printers, even the most entrenched competitors, can all form some agreement on what is meant by a particular set of CIE color numbers.

Considering the situation of the last ten years, that is real magic!

A "color model" is any method of associating names or numbers with colors. Examples include CMYK dot densities as reproduced by an offset press, RGB voltages sent to the electron guns of a cathode ray tube display, CIE LAB numbers as measured by a "colorimeter", or Pantone color numbers as seen on the "swatch books" published by that company.

The importance of a color model is that it allows communication about what color should be produced. When a graphic designer calls a printer on the telephone to specify a color by referring to a swatch book, he is doing this. When two computer programs communicate through a digital CIE LAB specification, they're doing much the same thing.

Some color models allow us to determine very effectively whether two colors are "similar" just by examining these numbers. This is very useful when we want a computer to be able to decide if two colors are "similar". In the case of the Pantone numbers, where numbers do nothing other than name the colors, this is scarcely possible at all. "Device" color models such as RGB and CMYK, which use numbers that are very directly related to the physical color reproduction process are better. "Scientifically designed" color models such as CIE LAB are better still, making them preferred for computer calculation of color reproduction problems.

When it is possible to define a "distance" between two sets of numbers that bears some relation to the difference between two colors, the color model may be referred to as a color space.