Those are good questions. In lieu of actual machinery that would return real reflectances, I'd had to come up with a scheme that would get get me close. The basis of the scheme is that RGB screen values aren't reflectances, and they shouldn't be treated as such.
I'll normalize to 0.8 or 0.9 because there is no real "pure white" in real life. Even white paint is rarely over 0.8. And with models with complicated geometry, even 0.8 requires a hefty number of ambient bounces to balance the energy---light really can penetrate into some deep spaces with 0.8. Consider that with a reflectivity of 0.9, only 10% of the energy is killed after every bounce, and you'll see that -ab will need to be prohibitively high to get an accurate rendering. And I choose Radiance for its accuracy, but I haven't got all month for a result.
The squaring of the result is to push the values down closer to zero. Most of the interesting colors are in the 0.2 range, and that's just a hack to get closer. I am sure there's a better way, but that's easy to remember and quick to compute.
On Sat, 9 Apr 2011, Terrance Mc Minn wrote:
Many wonderful points of inspirations - thanks for sharing.
Question - page 11 RGB Estimate:
1) Why are you normalising to 0.9 rather than full scale of 1?
2) Why are you squaring the result?
*Terrance Mc Minn