Postcard from the Gamut Edge
by Michael H. Brill
Occasionally, I have written summaries of the SID/IS&T Color Imaging Conference (CIC) for this magazine. This year, I will try to use this briefer and less-formal format to convey a sense of CIC-12, held November 9–12, 2004, in Scottsdale, Arizona. I will focus on one aspect that may interest members of the display community: nuances of color rendering on multiprimary displays that were designed for increased color gamut. Here, then, is a postcard from the "gamut edge."
In June 2004, we saw a product announcement for the Genoa five-primary display. Unsurprisingly, additive-color displays with more than three primaries (multiprimary displays) then appeared in almost every part of CIC-12. All but one of these presentations came from a group at the Akasaka Natural Vision Research Center in Japan.
Nagaaki Ohyama's keynote speech, "Concept and Technology of Natural Vision System," described the integration of the group's six-primary cameras with their six-primary displays. H. Kanazawa's oral presentation, "Color Conversion for Multiprimary Displays Using a Spherical Average Method," discussed a way of commanding the six display primaries so only one six-vector of activations – also called a primary setting – would yield any given color. The spotlight poster by J. Takiue and his co-authors, "Evaluation of Smoothness in Color Gradation on Multiprimary Displays," emphasized the important point that perceived smoothness of a spatial gradient of color, independent of quirks of the observer's color-matching functions, depends on smoothness of commanding the more-than-three primaries. Finally, my paper (with J. Larimer of NASA/Ames Research Center), "Avoiding On-Screen Metamerism with N-Primary Displays," offered a different theoretical perspective from that of the Akasaka group.
I cannot resist explaining that perspective here. Although the gamut polygon of a display with more than three primaries can extend beyond any RGB triangle, such a display has an embarrassment of riches. More than three primaries makes the representation of most colors non-unique, i.e., more than one primary setting can give the same screen color – a phenomenon I call on-screen meta-merism. Externally applied constraints such as those reported by Kanazawa ensure that only one primary setting can make any screen color.
However, this uniqueness applies to the CIE Standard Observer, an "average observer" whose color-viewing characteristics are the result of experiments in which observers were asked to match monochromatic lights with mixtures of three primaries. For actual observers whose characteristics are quite far from those of a CIE Standard Observer – older people with eye lenses that have yellowed with age, for example, or people wearing tinted lenses – more than one set of primaries may again yield the same screen color.
It is important that no two observers – even non-standard ones – disagree on whether two colors match on the display screen. We can assure this only if, starting with N primaries, we use only three at a time for any color. This condition is not as limiting as it sounds because these primaries may be composites – fixed linear combinations of the original N primaries.
With that background, I am ready to explain that our algorithm divides the display color gamut into triangles whose vertices are (possibly composite) primaries, and then commands a test color using only the three primaries of the triangle in which the test color lies. Any color within the triangle can be rendered by a combination of these primaries. In contrast, the Akasaka group makes any color in the entire gamut by using all the primaries at once, with a smooth transition between the settings at adjacent colors.
The Akasaka group demonstrates its color-coding algorithm on its display using a test pattern whose colors comprise a smooth gradient. Commanding the N-primary display one triangle at a time, as we recommend, does not guarantee the kind of smoothness one sees in the Akasaka group's mapping from N display activations to three tristimulus values. However, the triangle-by-triangle mapping is continuous even if its derivative is not; and the behavior is not disturbing. On the other hand, Akasaka mappings can lead – for some observers – to a color being visited multiple times (although smoothly) as one traverses the gradient.
Of course, practical realities may invade the dream world of multiprimary theory. Display colors may not be additive enough to use our algorithm, i.e., the actual display white may not be exactly the same color one would get from using all the primaries that are supposed to make the white. More generally, displays with more than three primaries will face significant challenges reaching a price point that will make them competitive with RGB displays, even for high-end users. However, at this moment I am optimistic about the multiprimary future.
To keep up with that future, please attend CIC-13 to be held November 8–11, 2005. As General Chair, I promise to let someone else summarize it. •