A new technical paper has just been uploaded which shows
that an initial projection of the gradient is needed to do proper normal
perturbation of height functions.

http://jbit.net/~sparky/sfgrad_bump/

Most of the paper is aimed at rendering on GPUs but an equation (eq. 4) is
presented which produces Blinn's perturbed normal but in a unified (2D/3D bump
function) formulation.

Section 5 gives more details regading differences though the analysis is given
in section 3.

The point is that simply adding or subtracting the 3D gradient of the function
gives visual errors and is not equivalent to the normal of the displaced
function nor a good approximation. Since it is an easy fix I would recommend
that it is worked into povray.


Cheers!

Post a reply to this message