Realistically, transmitting material is usually still absorbing across the
entire thickness of the layer. This means that when looking at an angle, the
path of the light is longer and thus the transmitted intensity is less. Even if
the layer is really thin (so that an infinitely thin surface is a good
approximation), this results in a transmittance
real_transmittance=perpendicular_transmittance^(1/|normal*ray_direction|)

Ideally, this should be done per-color, which can result in interesting color
variations with angle - although realistic in principle, in RGB model this won't
work so well (but in spectral model it should work). Just doing this just for
filter & transmit seperately should still be a big improvement and almost
physically correct.

This might not look necessary but in fact makes a huge difference: if you
specify a filtering sphere, the specular highlight is the only thing that makes
it look round, but the part in shadow looks exactly like a disc-shaped filter
glass: completely flat. In real life, the outline transmits less and the shape
looks rounded. At extreme sharp angles, no material should transmit much. Just
imagine a shadow cast by a hollow plastic sphere. Including this effect will
make transparent objects look much more realistic.

This could actually be simulated as a thin hollow shell filled with absorbing
media, but that's overkill for this purpose. This effect is seperate from the
energy conservation and should happen even if there's no reflection. A parameter
to turn this effect on or off would be nice. Probably as a tuning parameter
beteween 0 and 1 to allow greater flexibility.

Post a reply to this message