[Radiance-general] Re: Color appearance and color temperature
Greg Ward
[email protected]
Thu, 7 Feb 2002 10:35:30 -0800
> Does anybody know tools that might help in tuning chromaticity (color
> temperature) adaptation for Radiance renderings? For example, if you
> define
> a tungsten source (2856 K), the result is far too reddish to be
> plausible.
> This question arises when mixing sources of different temperatures,
> such as
> daylight and incandescent light. (When this is not the case, I suppose
> it
> is reasonable to use neutral source color, which in the Radiance RGB
> system
> is near 5500 K).
>
> In pcond you have an opportunity to define, besides the RGB primaries,
> also
> the white point, with the -d option. Does it include any "color
> appearance
> model" or something of a kind which could be useful? Any other ideas? In
> photography there are filters defined in "mired" units, which mean
> reciprocals of megakelvins. They shift the overall color temperature
> range.
> So, at least in the spectral domain a feasible transformation might be
> quite straightforward.
>
> Regards,
> Markku N.
>
One of the new features in Radiance 3.4 is a von Kries white-point
transform built into the -p option of pcond (and other programs). This
transform does a reasonable job of simulating the apparent colors under
a different illuminant, and is an improvement over the straight color
transform that was used previously. However, when you have a mixture of
illuminants in your rendered scene, there is no accepted adaptation
transform for such a condition. A number of researchers have worked on
this problem, and no one has proposed a workable solution to my
knowledge. The best you can do is pick an average of the illuminants
that seems right for the part of the scene you are viewing.
The issue of white balance is separate from the issue of spectral
rendering, but as long as we're on the topic, one of the things people
have requested over the years is a means for full spectral rendering in
Radiance. As it turns out, this is difficult for the very silly reason
that all the materials and patterns have arguments whose count would
change with the number of spectral samples. Choosing anything other
than three samples will make Radiance input files incompatible with the
change. In hindsight, it would have been much better to do things the
way I did them in the Material and Geometry Format (MGF), which
separates color into its own entity, with a number of different ways to
specify it.
Be that as it may, it is possible to obtain accurate spectral renderings
of a scene with a single illuminant, provided you precompute the
adjusted RGB color of the surfaces using a technique I described in my
most recent paper at the Color Imaging Conference. You can get a PDF
copy of the paper through the link below if you want to read about the
details.
-Greg
http://viz.cs.berkeley.edu/gwlarson/papers.html