If you've ever did some research on colour spaces, you've certainly come 
across the ubiquitous CIE "horseshoe" gamut diagrams.

It only tells a quarter of the story though, and if you ever tried to 
apply it to /pigment/ colours it gets you nowhere: The "horseshoe" shape 
presumes that you can simply crank up the intensity in any spectral band 
/ad infinitum/ to get the desired hue. But when we're talking about 
reflections, 100% is the maximum you can get for any given wavelength.

The attached AVI shows where the spectral colours /really/ lie in the 
CIE XYZ colour space, and we're in for a few surprises there:

- All axes are normalized to equal energy; thus the strong blue peak 
indicates that the human eye is actually /very/ sensitive to light in 
the blue spectrum; so how come blue pigments are perceived as the 
/darkest/? The reason is simple: The sensors for blue only respond to a 
very narrow band of wavelengths (note the "beads" placed at intervals of 
5 nm), so there's only so much "blue energy" to stimulate the eye. Note 
how this differs from the perception of single-wavelength light sources 
like LEDs: Blue LEDs /do/ appear very bright, because they concentrate 
all their energy in a single wavelength in the blue region of the spectrum.

- The shape resembles anything but a horseshoe; in fact it is all 
closed, as the extreme red and blue wavelengths contribute virtually 
nothing to the overall brightness, ultimatively converging to black.


The diagram also includes the loci of a set of primaries (and their 
combinations) labeled "sRGB' Red", "sRGB' Green", "sRGB' Yellow" and so 
on. These are /not/ the sRGB primaries; rather, they are primaries that, 
when used as /pigments/ and /illuminated/ with the sRGB reference white 
(D65), would /look/ like the sRGB primaries. (This is a property that 
will be needed when attempting to use a less naive and more realistic 
colour comutations approach while still allowing sRGB colour input.)

Another thing to note about these primaries is that they are /not/ to 
scale with respect to the spectral loci.

Post a reply to this message

Attachments:
Download 'spectrum.avi.dat' (3276 KB)