clipka <ano### [at] anonymousorg> wrote:
> Hmmm... why do you use an uneven gray curve in the first place? (Unless
> that's your official whitepoint, in which case officially there's no
> shift in hue.)

Because I'm aiming for approximations of real life curves, and most of the white
samples I've seen have a sharp drop-off in the extreme violet.  But as I noted,
the visual impact is minimal; under default Lightsys IV color system settings,
the white curve yields rgb <0.9985, 1.0017, 0.9848>, or srgb <255, 255, 253> /
255 in 8 bits.

As it turns out, there are a couple of parameters that can be used to even out
the white curve, although I had introduced them for a different reason.

I should point out that, at least for now, I'm only doing reflective curves.
The "whitepoint" is whatever is reflected by an object that reflects all
wavelengths (mostly) equally.

>  From what I understand now, you have precomputed spectral curves for
> "pure" colours, and you simply compute a weighted average of one of
> those precomputed colours, the whitepoint and black; so according to my
> knowledge "you /can/ be sure that the actual percieved hue is stable [in
> relation to the whitepoint] WRT the other parameters of the model"
> (emphasis added).

Exactly, except that the spectral curves themselves are not precomputed.  The
parameters that match them to sRGB hues are precomputed.

> OTOH, you're obviously still imposing some artificial limit on the
> saturation of colours there, as even reflective colours can - at least
> in theory - be arbitrarily narrow-band (albeit at reduced brightness).

I've decided to use the term "brightness" instead of "saturation."  There will
be some minor rewriting of SDL and major rewriting of documentation.

Post a reply to this message