RE: [Algorithms] Heightfield to NURBS conversion

[Brian M. <bma...@ra...>]

This might not help a lot but here are couple of references.

Try:
	Pages 101-106 in Advanced Animation and Rendering Techniques - by Alan Watt
/ Mark Watt.

and	Piecewise Smooth Surface Reconstruction - Hughes Hoppe, Tony DeRose, Tom
Duchamp, Mark Halstead
(A Siggraph paper - its available on Hughes Hoppe's sight at Microsoft
research.)

	The first reference goes a little into building a b-spline patch by fitting
individual curves - hard to get the error control it sounds like you need,
but relatively easy to implement.
	The paper details arbitrary surface fitting for b-splines. Thankfully a
regular height field has a nice regular structure so you don't have to worry
about the tricky continuity fitting that arises in general patch networks.
This paper includes subdivision techniques to guarantee the error of the
patch approximation.

	One other thing to think about is fitting bezier patches to small blocks of
the heightfield (something like 4x4 blocks). Then do the standard trick of
using multiple knots to represent the bezier patches inside one b-spline
patch for the terrain. Then you remove the surplus multiple knots. Messy,
but something that's easier to get going with the kind of patch toolkit
people normally have lying around.

	Unfortunately patch fitting is tricky. :-)

-Brian.