Hi folks,
one of the things that makes we always wonder is how the normalization
from "absolute" color coordinates XYZ to "relative" or "normalized"
coordinates xyz is done. One can of course simply accept the definition,
namely
x = X / (X + Y + Z) etc
but why is it done this way? First of all, if Y is the luminance (and
that's as far as I know how the Y channel is defined, approximately),
then why hasn't x been defined as "luma normalized" coordinate, namely
x = X / Y etc.
which, clearly, would set y = 1 (relative luminance = identity).
On a related thought, why does one divide by the *sum* of the three
coordinates. If I would consider XYZ as a vector in R^2, then wouldn't
it be "more natural" to divide by its length, i.e.
x = X / sqrt(X^2 + Y^2 + Z^2) etc.
I understand the way things are done, and I'm ready to accept "this is
only a definition", but is there a specific reason behind the way it was
done? I find that the two approaches above (either of them) seem to be
more natural than the one taken by the CIE, so what's the rationale here?
Thanks,
Thomas
|
