Am 22.12.2011 21:31, schrieb waggy:
> clipka:  Actually, a true curvature pattern is what I really want (despite
> boring CSG seams) and you're correct that it would be better to calculate it
> precisely for each of the two-dozen or so primitives, basically following the
> same idea as how the surface normal is calculated at a ray-object intersection
> point.  As with the slope pattern, I was thinking in terms of a directional
> derivative, so the slopes at two nearby points on a surface are sufficient to
> calculate the curvature in that direction.  (That's pretty much the definition
> of curvature in the limit as the distance between the points approaches zero.)

Be aware that curvature in 3D space can be (and in most cases is) 
anisotropic, so you'll need to measure the surface normal of at least 
three points. (Note how this is similar to approximating the surface 
normal by measuring the position of three nearby points; two points 
won't do there either.)

Post a reply to this message