Geometric Modelling

Geometric Modelling
2 1 0 -1 1
-1 -1 0 1 2
Lecture in Winter Term by G. Greiner Transscript by J. Kaminski Revision by S. Bergler
ii
iii
Contents
1 General Remarks on Curves and Surfaces 2 Polynomial Curves 2.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Different Basis of . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Lagrange Polynomials . . . . . . . . . . . . . . . . . 2.2.2 Bernstein Polynomials . . . . . . . . . . . . . . . . . 2.2.3 Comparison of Lagrange and Bernstein Representation 2.2.4 Change of Basis . . . . . . . . . . . . . . . . . . . . 2.3 Polar Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Bzier Curves 3.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.1 Shape Properties for Bzier Curves . . . . . 3.2 Algorithms for Bzier Curves . . . . . . . . . . . . . 3.2.1 The de Casteljau Algorithm . . . . . . . . . 3.2.2 The de Casteljau Subdivision Algorithm . . . 3.2.3 How to draw a Bzier Curve . . . . . . . . . 3.2.4 Degree Elevation . . . . . . . . . . . . . . . 3.2.5 Degree Reduction . . . . . . . . . . . . . . . 3.3 Bzier Splines . . . . . . . . . . . . . . . . . . . . . 3.3.1 Construction of complex Curves . . . . . . 3.3.2 -Continuity of two Bzier Curves . . . . . 3.3.3 Geometrical Construction of continuous Joins 3.3.4 Interpolation with Bzier Splines . . . . . . .
1 3 3 4 4 5 7 7 8 11 11 13 15 15 17 18 19 21 23 23 24 26 28 31 31 34 35 36 37 38 41 41 45 45 47 48
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
4 B-splines 4.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 B-spline Curves . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 Shape Properties of B-spline Curves . . . . . . . . 4.2.2 Comparison of B-Spline and Bzier Curves . . . . 4.3 Control Points and Polar Forms . . . . . . . . . . . . . . . 4.4 Multiple Knots . . . . . . . . . . . . . . . . . . . . . . . 4.5 Algorithms for B-spline Curves . . . . . . . . . . . . . . . 4.5.1 The de Boor Algorithm . . . . . . . . . . . . . . . 4.5.2 Conversion of B-spline Curves into Bzier Splines 4.5.3 The Boehm Algorithm . . . . . . . . . . . . . . . 4.5.4 Knot Deletion . . . . . . . . . . . . . . . . . . . . 4.5.5 Subdivision for uniform B-spline curves . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
Contents
4.6 Interpolation and Approximation . . . . . . . . . . . . . . . 4.6.1 B-spline Interpolation . . . . . . . . . . . . . . . . 4.6.2 Special case: Cubic spline interpolation at the knots 4.6.3 B-spline approximation . . . . . . . . . . . . . . . . 4.6.4 Parametrization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
iv 51 51 52 54 55 57 57 58 59 59 61 62 64 65 67 67 68 70 72 74 77 77 78 79 79 80 82 83 83 83 84 84 84 85 87 91 91 92 95 96 96
5 Rational Curves 5.1 Rational Bzier Curves . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.1 Properties of rational Bzier curves . . . . . . . . . . . . . . . 5.1.2 Evaluation of rational Bziercurves . . . . . . . . . . . . . . . 5.1.3 Inuence and properties of the weights . . . . . . . . . . . . . 5.1.4 Quadratic rational Bzier Curves as Conic Sections . . . . . . . 5.1.5 Segments of Conic Sections as quadratic rational Bzier Curves 5.1.6 Rational Bzier-Spline Curves . . . . . . . . . . . . . . . . . . 5.2 Rational B-Spline Curves (NURBS) . . . . . . . . . . . . . . . . . . . 6 Elementary Differential Geometry for Curves 6.1 Reparametrization . . . . . . . . . . . . . . . . 6.2 Planar Curves . . . . . . . . . . . . . . . . . . 6.3 Space Curves . . . . . . . . . . . . . . . . . . 6.4 Geometric and Parametric Continuity . . . . . 6.4.1 Geometric Continuity of Bzier Curves 7 Tensor Product Surfaces 7.1 General Approach . . . . . . . . . . . . . . . . 7.2 Tensor Product Bzier Surfaces . . . . . . . . . 7.3 Algorithms for TP Bzier Surfaces . . . . . . . 7.3.1 Evaluation with the TP Approach . . . 7.3.2 The direct de Casteljau Algorithm . . 7.4 Tensor Product B-Spline Surfaces . . . . . . . 7.5 Algorithms for TP B-Spline Surfaces . . . . . . 7.5.1 Evaluation of TP B-Spline Surfaces . . 7.5.2 Knot insertion . . . . . . . . . . . . . 7.6 Interpolation and Approximation . . . . . . . . 7.6.1 General interpolation . . . . . . . . . . 7.6.2 Special case: Interpolation at the knots 7.7 Trimming of parametric Patches . . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
8 Ray Tracing of TP-Bzier Surfaces (Bzier Clipping) 9 Triangular Bzier Patches 9.1 Barycentric Coordinates in . . . . . . 9.2 Bernstein Polynomials over Triangles . . 9.3 Triangular Bzier Patches . . . . . . . . . 9.4 Algorithms for Triangular Bzier Patches 9.4.1 The de Casteljau Algorithm . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
v 9.4.2 Degree Elevation . . . . . . . . . . . . . . . . . . . . 9.4.3 Subdivision . . . . . . . . . . . . . . . . . . . . . . . -Continuity of Triangular Bzier Patches . . . . . . . . . . Interpolation with piecewise Bzier Triangles (Split Surfaces) 9.6.1 Clough-Tocher . . . . . . . . . . . . . . . . . . . . . 9.6.2 Powell-Sabin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Contents
. . . . . . . . . . . . 96 97 100 102 102 103 105 105 109 111 111 112 112 115 117 120 121 121 124
9.5 9.6
10 Triangle meshes 10.1 Basics . . . . . . . . . . . . . . . . . . . . . 10.2 Computer representation of triangle meshes . 10.3 Parametrization of triangle meshes . . . . . . 10.3.1 The spring modell . . . . . . . . . . 10.3.2 Parametrization of the border vertices 10.3.3 Parametrization of the inner vertices .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
11 Subdivision Surfaces 11.1 Approximative schemes: Loop-Subdivision . . . . . . . . . . . . 11.2 Interpolatory schemes: Modied Buttery-Subdivision . . . . . . 11.3 Limit points for approximative subdivision . . . . . . . . . . . . . 11.3.1 Limit points for the subdivision of cubic splines . . . . . . 11.3.2 Limit points for the subdivision of regular triangle meshes
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
Contents
vi
vii
List of Figures
1.1 1.2 2.1 2.2 2.3 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10 4.11 4.12 4.13 Curves in the plane and in space . . . . . . . . . . . . . . . . . . . . . . . . . A torus and a free form surface . . . . . . . . . . . . . . . . . . . . . . . . . .
Quadratic Lagrange polynomials with respect to paramter values and . . Bernstein polynomials for , and . . . . . . . . . . . . . . . Polar Form of a Curve of degree . . . . . . . . . . . . . . . . . . . . . .
1 1 5 6 10 12 13 14 15 15 16 18 19 19 19 20 22 24 25 26 27 27 29 31 32 32 33 35 35 35 36 39 42 42 43 43
A Bzier curve . . . . . . . . . . . . . . . . . . . . . Convex hull . . . . . . . . . . . . . . . . . . . . . . . Variation diminishing property . . . . . . . . . . . . . de Casteljau schemas for and . . . . . . . Graphical representation of the de Casteljau algorithm Evaluation of with de Casteljau . . . . . . . . . . Left partial curve after de Casteljau subdivision . . . . Multiple subdivision . . . . . . . . . . . . . . . . . . Recursive subdivision . . . . . . . . . . . . . . . . . . ........... Degree elevation . . . . . . . . . . . Degree elevation ........... Degree reduction Joins of Bzier curves . . . . . . . . . . . . . . . . . . Join conditions in the de Casteljau algorithm . . . . . . de Casteljau conditions for - and -continuity . . . -continuation of the standard parabola . . . . . . . . -join of several cubic Bzier segments . . . . . . . . Interpolation with Catmull-Rom Splines . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A knot vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-splines for . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-splines for The B-spline . . . . . . . . . . . . . . . . . . . . . . . . . . Local convex hull . . . . . . . . . . . . . . . . . . . . . . . . . . . control points coincide . . . . . . . . . . . . . . . . . . . . . control points on a line . . . . . . . . . . . . . . . . . . . . . control points on a line . . . . . . . . . . . . . . . . . . B-splines over a knot vector with multiplicities (emphasized: ) . Bzier representation of a B-spline curve with endpoint interpolation Evaluation schema of the de Boor algorithm for . . . . . . . Visualization of the coefcients for the de Boor algorithm ( ) . Example for the de Boor algorithm (1): coefcients . . . . . . . . .
List of Figures
4.14 4.15 4.16 4.17 4.18 4.19 4.20 4.21 4.22 5.1 5.2 5.3 5.4 5.5 5.6 6.1 6.2 6.3 6.4 6.5 6.6 7.1 7.2 7.3 7.4 7.5 7.6 7.7 7.8 7.9 7.10 8.1 8.2 8.3 8.4 8.5 9.1 9.2 9.3 9.4 9.5 9.6 Example for the de Boor algorithm (2): evaluation . . . . . . . Example for the de Boor algorithm (3): graphical construction Schema for the Boehm algorithm . . . . . . . . . . . . . . . . Example for the Boehm algorithm . . . . . . . . . . . . . . . Knot insertion at . . . . . . . . . . . . . . . . . . Geometrical construction: cutting the corners . . . . . . . . . Subdivision with open knot vector and . . . . . . . . . Illustration of parameter values . . . . . . . . . . . . . . . . . Chordal parametrisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
viii 44 44 46 47 48 49 50 51 56 57 58 59 60 61 64 68 69 71 72 73 75 77 78 80 80 81 83 83 85 85 86 87 88 89 90 90 92 93 93 95 96 97
Homogeneous coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . Interpretation of rat. curves with homogeneous coordinates (proj. ) de Casteljau scheme for rational Bzier curves . . . . . . . . . . . . . . . . Effect of the weights ( ) . . . . . . . . . . . . . . . . . . . . . . . . . Conic sections: ellipsis, parabola und hyperbola . . . . . . . . . . . . . . . Intersection of a cone with a pyramid and projection of the intersection line Arc length . . . . . . . . . . . The 2D Frenet frame . . . . . The 3D Frenet frame . . . . . Spiral . . . . . . . . . . . . . Continuity of Bzier splines . Generalized A-frame property . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A tensor product Bzier surface . . . . . . . . . . . . . . . Parameter grid, control net and resulting surface . . . . . . . de Casteljau scheme for tensor product Bzier surfaces . . . Bilinear interpolation . . . . . . . . . . . . . . . . . . . . . Direct de Casteljau scheme . . . . . . . . . . . . . . . . . . Knot insertion with the Boehm algorithm . . . . . . . . . . Building a hierarchical knot grid for B-spline patches . . . . Trimming a patch . . . . . . . . . . . . . . . . . . . . . . . Specifying inner and outer contours in the parameter domain Determining inside and outside of a trimmed patch . . . . . Bzier clipping, step 1 . . . Bzier clipping, step 2 . . . Bzier clipping, step 3 . . . Bzier clipping, steps 5 and 6 Bzier clipping, step 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Barycentric coordinates . . . . . . . . . . . . . . . . . . . . Divide ratios and barycentric coordinates . . . . . . . . . . for Bernstein polynomials Arrangement of indices Triangular control nets for and . . . . . . . . . de Casteljau scheme for triangular Bzier patches . . . . . . . . . . . . . . . . . . . . . Degree elevation
ix 9.7 9.8 9.9 9.10 9.11 9.12 9.13 10.1 10.2 10.3 10.4 10.5 10.6 10.7 11.1 11.2 11.3 11.4 11.5 11.6 11.7 11.8 11.9 Splitting triangles: 1-to-2 and 1-to-4 split . . . . A control net and the rst 1-to-4 subdivision step Farin-Prautzsch 1-to-4 subdivision scheme . . . . Attaching two Bzier triangles continuously . . . Diamond condition for . . . . . . . . . . Clough-Tocher split surface scheme . . . . . . . Powell-Sabin split surface scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
List of Figures
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 98 99 100 101 102 103 105 106 107 107 108 108 114 115 117 117 118 119 120 120 121 122 123
1-neighbourhood and 2-neighbourhood of a vertice . . . . . . . . . . . . . . not possible in manifold triangulations . . . . . . . . . . . . . . . . . . . . . . open and closed fan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Moebius Strip and Klein Bottle . . . . . . . . . . . . . . . . . . . . . . . . . . Tetrahedron . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Two-torus: surface of genus 2 . . . . . . . . . . . . . . . . . . . . . . . . . . A local part of a triangular mesh (a) is mapped into the plane by expontial mapping (b) and the local weights are determined (c). . . . . . . . . . . . . . . . .
1-to-4 split . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mask for the calculation of odd and even vertices . . . . . . . . . . . . . . . . masks for the calculation of irregular even vertices . . . . . . . . . . . . . . . Loop-Subdivision: top: coarse triangle mesh, bottom: after 1 subdivision step . Loop-Subdivision: top: after 2 steps, bottom: after 3 steps . . . . . . . . . . . Buttery: Mask for the calculation of the odd vertices . . . . . . . . . . . . . . one irregular neighbour vertice with valence 3, 4 and 5 . . . . . . . . . . . . . 4-point-scheme for boarder vertices . . . . . . . . . . . . . . . . . . . . . . . Buttery-Subdivision: top: coarse triangle mesh, bottom: after 1 subdivision step . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.10Buttery-Subdivision: top: after 2 steps, bottom: after 3 steps . . . . . . . . .
List of Figures
xi
Bibliography
[1] C. de Boor. A Practical Guide to Splines (Applied Mathematical Sciences Vol. 27). Springer, Heidelberg 1987 [2] M. do Carmo. Differential Geometry of Curves and Surfaces. Prentice Hall, Englewood Cliffs, 1976 [3] J. Hoschek, D. Lasser. Grundlagen der geometrischen Datenverarbeitung. Teubner, Stuttgart, 1992 [4] G. Farin. Curves and Surfaces for Computer-Aided Geometric Design: A Practical Guide. Academic Press, San Diego, edition, 1997 [5] L. Piegel. The NURBS Book. Springer, 1995
Bibliography
xii
1 General Remarks on Curves and Surfaces

A curve is a one-dimensional connected point set in a two-dimensional plane or in threedimensional space. It is at least piecewise smooth. A surface is a two-dimensional point set in three-dimensional space with the same properties.
Figure 1.1: Curves in the plane and in space
Distinguish between regular and free form (= non-regular) curves and surfaces. Regular curves are e. g. lines and line segments, circles and arcs; regular surfaces are e. g. triangles, rectangles, polygons in the plane and spheres, cylinders, cones and parts of them in 3D space. Non-regular curves and surfaces is everything else (e. g. fender or hood of a car).
Figure 1.2: A torus and a free form surface
There are three ways to describe curves and surfaces mathematically.
1 General Remarks on Curves and Surfaces

example circle: implicit

explicit
Graph of
parameterized
example sphere: implicit
explicit
parameterized

Graph von
Proposition: "Reasonable" curves and surfaces can be described (at least piecewise) in each of the tree ways. Note that and are not unique! Change of representation: : : Invert

and dene
Remark: reasonable means here: differentiable and
e. g.
: implicit function theorem. If the following statements are true, : solved to

continuous has continous partial derivations and and
can be
and it holds

All statements are also true for surfaces.
Remark:
regular forms (circles, straight lines, spheres, cones usw.) are mostly described explicitly free form geometries mostly parametric
2 Polynomial Curves
2.1 Basics
Polynomial of th degree:
set of all polynomials
If
, then is the degree of the polynomial
are coefcients of the polynomial

:
degree
Set of all polynomials of degree
Denition 2.1 (Polynomial Curve) A two-dimensional (three-dimensional, -dimensional) polynomial curve is a parameterized : curve
with

(2.1)
The set of all -dimensional polynomial curves of degree is denoted by . Example:
Evaluation with equation 2.1 requires multiplications and additions, thus it is of type . A smarter way to evaluate the equation is to save and reuse the intermediary results of the multiplication of the parameter . In this case we have multiplications and additions (complexity ). An even faster evaluation is possible using Horners Rule: Algorithm 2.1 (Horners Rule)
Complexity: with multiplications and additions. Pseudo code: for
end for
to do
2 Polynomial Curves
2.2
Different Basis of
Note that and are linear spaces (vector spaces).

(polynomial functions)

2D curve 3D curve D curve
Remark: The dimension of a space corresponds to the number of degrees of freedom in this space. These are the coefcients for polynomial functions.
The standard basis for is the monomial basis presented in the next sections. 2.2.1 Lagrange Polynomials
. Other possible basis for are
Denition 2.2 Given parameter values

is called the
th
. The polynomial

with
(2.2)
Lagrange1 polynomial of degree .
Theorem 2.1 The Lagrange polynomials
form a basis of :

Sketch of Proof:
(2.3)
Interpretation: The points are interpolated at parameter values . There is only one polynomial of th degree, that interpolates these points (number of conditions = number of unique solvable system of linear equations), therefore the Lagrange repdegrees of freedom resentation is unique. Example: Given
Joseph Louis de Lagrange (17361813)
2.2 Different Basis of
0.8
0.6
0.4
0.2
0.2
0.4
0.6
0.8
-0.2
Figure 2.1: Quadratic Lagrange polynomials with respect to paramter values
and
2.2.2
Bernstein Polynomials
Denition 2.3 The polynomial

(2.4)
is called the th Bernstein2 polynomial of degree . Additional agreement: , if or .

Remark: Properties of the binomial coefcients:
(1) denition: (2) recursive denition:

(3) Pascal triangle: recursion:
and symmetry:
Theorem 2.2 Properties of Bernstein polynomials: (1) (2) (3)

2
(Binomial Theorem:
(partition of 1)

(positivity)
Serge N. Bernstein, 1912
2 Polynomial Curves
(4) (5) (6)
1 0.8 0.6 0.4 0.2
6 (recursion)
(symmetry) has a maximum in at .

1 0.8 0.6 0.4 0.2
1 0.8 0.6 0.4 0.2
0.2
0.4
0.6
0.8
0.2
0.4
0.6
0.8
0.2
0.4
0.6
0.8
Figure 2.2: Bernstein polynomials for
and
Example:

Denition 2.4 (General Bernstein Polynomial) : Given the parametric interval
th
(2.5)
is called the
Bernstein polynomial for the interval
Theorem 2.2 holds also true. How we get the general Bernstein Polynomial:
Remark: Note that with is represented here in barycentric coordinates with respect to and .
Evaluating formula 2.3 at yields:
Theorem 2.3 The Bernstein polynomials
form a basis of :
(2.6)
7 2.2.3
2.2 Different Basis of

Comparison of Lagrange and Bernstein Representation
Different representations:
will be interpolated
will be interpolated
The location of or determine the shape of the curve somehow intuitively. This is not true for the monomial coefcients . 2.2.4 Change of Basis
How to switch from one representation to another? A general rule from Linear Algebra claims: Theorem 2.4 Given two linear combinations representing one vector
.
. . . . .
.
. . . . .
. . . . .
. . . . .
(2.7)
Proof:

Example:
Supplement: Afne Maps

Denition 2.5 A map
is called afne, if
(2.8)
2 Polynomial Curves
for all with . The linear combination is called afne combination in this holds true for all , then it is called convex combination. case. If additionally An afne map especially preserves afne combinations, i. e.

Denition 2.6 A map
(2.9)
is called multiafne, if for any xed arguments the map
is afne. (Diction: Example:

bi-afne,
(2.10)
tri-afne etc.)
is multiafne. is multiafne. is not multiafne.

exist such that
Theorem 2.5 (Characterization of Afne Maps) is afne a linear map and a translation vector
Remark: afne maps map lines onto lines parallelity is preserved division ratios are preserved angles are not preserved examples for afne maps: rotation, scaling, shearing
2.3
Polar Forms
with the
Theorem 2.6 (Polynomials and Polar Forms) For each polynomial of degree exists a unique map following properties: (1) is n-afne:

if
9 (2) is symmetric in each variable, i. e.

2.3 Polar Forms

(diagonal property) .
(3)
Diction: Example:
is the polar form (blossom) of

1 1

1 1
General formula for polynomials of degree in monomial representation:
:
(2.11)
Example:

Consider a degree 2 polynomial curve: ing way:
. This can be decomposed in the follow-
2 Polynomial Curves

10
Figure 2.3: Polar Form of a Curve of degree
Corollary 2.1 are coefcients with respect to the Bernstein basis. With degree we have coefcients , .
Corollary 2.2 This gives reason to a geometric construction to get the point corresponding to the curve param eter :

This geometric construction is called the de Casteljau algorithm. (see section 3.2.1).
11
3 Bzier Curves
3.1 Basics
Denition 3.1 (Bzier Curve) A polynomial curve
(3.1)
with the gereralized Bernstein polynomial is called Bzier curve1 of degree over . The points are called control points or Bzier points, the polygon formed by the control points is called control polygon.
Theorem 3.1 (Bzier Points and Polar Form) Given a polynomial of degree , the Bzier points with respect to
are (3.2)
Proof: Write as an afne combination of and :
Now we have:

decompose s
Pierre Bzier, 1966
3 Bzier Curves
Example: Given:
12

Figure 3.1: A Bzier curve
Theorem 3.2 (Derivative of a Bzier Curve) a Bzier curve over Let
. Then:

(3.3)
Proof: Consider
(i)
13
3.1 Basics
(ii)
Corollary 3.1

etc. for higher derivatives. 3.1.1
(3.4)
Shape Properties for Bzier Curves
Theorem 3.3 (Lemma of Bzier) For Bzier curve over (1) For curve point
we have:
, is contained in the closed convex hull of the control polygon. (The is a convex combination of the ).
the line .
Remark: A set of points is called convex, if for any two points
Figure 3.2: Convex hull
(2) End-point interpolation: and the last control point.
and
, i. e. the curve runs through the rst
(3) Tangency condition: In the end-points the curve is tangent to the control polygon. und
3 Bzier Curves
(4) Afne invariance: Let
14
an afne map, then
I. e. to transform the curve with the afne transformation , it is sufcient to transform the control points. (5) Variation diminishing property: A Bzier curve does not vary more than its control polygon. More precisely: For any hyperplane in (line in , plane in etc.), then

control polygon

Figure 3.3: Variation diminishing property
Proof: (1) convex hull: For
(2)
set
und
(Theorem 3.1, diagonal property of the polar form) (3)
because
sonst
The same for (4) Let
.
an afne map
(5) see [4].
15
3.2 Algorithms for Bzier Curves
3.2
3.2.1
Algorithms for Bzier Curves

The de Casteljau Algorithm
The de Casteljau algorithm is a method for evaluating Bzier curves. Given: control points . of a Bzier curve Algorithm 3.1 (de Casteljau)
Then:
is a point on the Bzier curve.

depend on parameter for
(3.5)
Remark: The points
Figure 3.4: de Casteljau schemas for
and
Sketch of Proof: The proof of theorem 3.1 on page 11 is just the application of de Casteljaus algorithm. Only the intermediate results are renamed:
Graphical representation of the de Casteljau algorithm for

and

Figure 3.5: Graphical representation of the de Casteljau algorithm
3 Bzier Curves
16
Example:
Bzier representation with respect to
Evaluation of

with de Casteljau at
Figure 3.6: Evaluation of
with de Casteljau
We can compute the polar form of any value using the de Casteljau algorithm. Note that due to the symmetry of the polar form, there are two possibilities. Find: (1)

de Casteljau: evaluate rst , then :

(2) symmetric variation: rst , then
17

Remark: The de Casteljau algorithm is stable (only evaluation of convex combinations). Complexity is (in contrast to with Horners rule). Improvement is possible by recursive subdivision.
3.2.2
The de Casteljau Subdivision Algorithm

Theorem 3.4 Let Dene:
a Bzier curve over
and
Bzier curve over Bzier curve over
where are points of the de Casteljau algorithm evaluated at . Then:
(3.6) is:
Proof: According to the proof of theorem 3.1 the left part of

3 Bzier Curves
( are Bzier points with respect to
18
). Similar for the right part.
. See gure 3.7 for evaluation of the Example: According to gure 3.5 set and de Casteljau algorithm. The control points of the Bzier curve over are .

Figure 3.7: Left partial curve after de Casteljau subdivision
By repeated subdivision (e. g. uniform subdivision at the midpoints) we get a sequence of polygons that converges fast to the Bzier curve. Lemma 3.1 Let Let
a -continuous curve (i. e. two times differentiable). a line dened by . Then:
(3.7)
(Proof follows from Taylor expansion.) Consequence: The uniform subdivision for Bzier curves converges (at least) quadratically.
Remark: The evaluation of Bzier curves at equidistant parameter values converges quadratically, too.
3.2.3
How to draw a Bzier Curve

.
Given: degree , , Find: polygon for piecewise linear approximation of
Algorithm 3.2 (1) Evaluate at parameter values with evaluation of Berstein polynomials). (2) Draw the polygon Example:
(e. g. with de Casteljau or by

Algorithm 3.3 (1) Subdivide Casteljau.
at the parametrical midpoint
in
and
using de
19
Figure 3.8: Multiple subdivision
(2) Continue recursively until a certain recursion depth . (3) Draw the control polygon of the last recursion step of (2).
Figure 3.9: Recursive subdivision
3.2.4
Degree Elevation
Theorem 3.5 Let Assume
a Bzier curve of degree over
as Bzier curve of degree over
. Then:
(3.8)
(where
Figure 3.10: Degree elevation
Proof: (i) Let
the -afne polar form of
(as polynomial of degree ).
3 Bzier Curves
Denition 3.2
20
(3.9)
(ii)
(a) afne in every (b) symmetric (c) diagonal
is the polar form of (as polynomial of degree ).

Example:
Remark: The repeated degree elevation yields a sequence of polygons then converge slowly to the Bzier curve. One can use this fact to prove the variation diminishing property of Bzier curves (see [4]).
21 3.2.5 Degree Reduction
Given: a Bzier curve
Find: a representation with reduced degree:
This is not possible in general; therefore nd the best approximation . The degree elevation of . Find control points yields , where in general such that
1st Method of Solution: Evaluation of equation 3.8 yields:
Given: control points The new control points
This leaves a linear system to solve. Example:

are obtained as follows:

To receive the minimum we have to set the gradient of be written as
to zero. The partial derivatives of
can
with Then, the gradient of
is:
grad
We receive two sets of linear equations: one for the x-coordinate and one for the y-coordinate:
3 Bzier Curves
with the solution:
22
The Bzier curve with reduced degree is depicted in gure 3.12.
Figure 3.12: Degree reduction
2nd Method of Solution: Another method is based on the fact that degree elevation is a linear porcess, i. e. the control points of the degree-elevated curve can be obtained by an appropriate matrix multiplication on the initial control points. E. g., the degree elevation with the initial points of the degree elevated curve can be described by and the control points the following matrix:
To nd the degree reduction of a curve, one has to solve the linear system . In general, this is not possible (over-determined linear system). Therefore nd the best possible solution satisfying the condition .
Theorem 3.6 (Gauss Normal Equation) The best approximation for the unsolvable linear system

can be obtained by (3.10)
Proof:
min

min
23 To receive the minimum set the derivation to zero:
3.3 Bzier Splines
Remark: Due to the structure of can be solved very efciently. , the square matrix is tri-diagonal. Therefore, the linear system
Example:
(see gure 3.12)
Given: control points The following matrix
describes the degree elevation
To calculate the new control points use Gauss Normal Equation

with the result
3.3
3.3.1
Bzier Splines
Construction of complex Curves
To model more complex shapes with Bzier curves, there are two possibilities: Method 1: use a Bzier curve of high degree. This method may cause numerical problems. Method 2: split curve into several simple parts and represent every segment as a low degree Bzier curve. Denition 3.3 A Bzier spline is a piecewise polynomial curve. The segments are represented by Bzier curves.
3 Bzier Curves
24
Figure 3.13: Joins of Bzier curves
Problem: continuity of the joins? Discontinuities or non-differentiable joins cause nonsmoothness of the curve. Denition 3.4 Given two polynomial curves continuous at if Interpretation: 3.3.2

and
. Then the curves are

-continuous = continuous, no jumps -continuous = -continuous =

-continuous + same tangent vector -continuous + same osculating circle.
-Continuity of two Bzier Curves
Theorem 3.7 Let ,
. Then:
are
and
Bzier curves over

and
-continuous at
(3.11)
Proof: Any Bzier curve has a unique representation over every interval. (In this case we only .) Therefore consider over the interval , but in fact the curve has the domain we can dene curve over the interval without changing it:

. . .

25
3.3 Bzier Splines
. . .

are
-continuous at
for all
Remark:
Figure 3.14: Join conditions in the de Casteljau algorithm
3 Bzier Curves
3.3.3 Let (1) (2) Geometrical Construction of continuous Joins and like above and are are

26
-continuous
-continuous -continuous
. Then:
.
(3)
are
A-frame property:

Example: Given: standard parabola Find: continuation

? over
over
-continuity: -continuity:
Result: see gure 3.16

Figure 3.15: de Casteljau conditions for
and
-continuity
Remark: The gure 3.17 is fully described by the location of are B-Spline control points (see chapter 4). Note: The points
the endpoints and the intervals .
27
3.3 Bzier Splines
Figure 3.16:
-continuation
of the standard parabola

Figure 3.17:

-join
of several cubic Bzier segments
3 Bzier Curves
3.3.4 Interpolation with Bzier Splines
28
Given: Points
and parameter values
Find: A curve that interpolates the points at the corresponding parameter values, i. e. all must be on this curve. A simple idea are Catmull-Rom Splines 2 : nd a (e. g. cubic) Bzier curve between two neighbour points , specify the tangents (= derivatives) at the end points as follows:
(3.12)
(3.13) Theorem 3.8 If
, you get the control points of an interpolating Bzier curve as follows:

(3.14)
(3.15) Sketch of Proof:
Remark: Problem: What happens at the end points? 1st case: closed curve (i. e. 2
nd
).
case: open curve ( hold true.
). In this case the natural end conditions
and
29

3.3 Bzier Splines

Figure 3.18: Interpolation with Catmull-Rom Splines

When interpolating points, you have to consider the order (or topology = neighbourhood relation) of the points. Only if the points are uniformly distributed, you can assume that all parameter intervals have the ). In general, this is not true. length 1 (e. g.
E. Catmull, R. Rom. A Class of interpolating Splines, 1974
3 Bzier Curves
30
31
4 B-splines
4.1 Basics
B-spline curves are piecewise polynomial curves with built-in -continuity at the joins. Mainly we deal with curves of degree that are -continuous, e. g. a B-spline curve of degree 1 is piecewise linear and -continuous. Given: degree and a sequence of knots of control points of the curve. Thus, the knot vector .
where is the number denes parameter intervals

Figure 4.1: A knot vector
The B-spline basis functions are dened recursively by degree :

General recursion formula:
otherwise
Denition 4.1 (Normalized B-spline Basis Functions)

. . .

The are computed according to the follwing scheme:

. . .

otherwise
(4.1)
(4.2)

. . .

. . .

. . .

4 B-splines
Example: We choose the knot vector

32 ), i. e.
Figure 4.2: B-splines for
Figure 4.3: B-splines for
Remark: The B-spline is tangent-continuous at the joins.
33
4.1 Basics
Figure 4.4: The B-spline
Properties of B-splines (1)
is piecewise polynomial of degree .
(2) The support of : of intervals per section. (3) Positivity: (4) Partition of 1:
, i. e. the support consists
for
. for .
(5) The derivative of a basis function is given by

(4.3)
Proof by induction on k: initial condition: and are either or (see g. 4.2), and thus is either
or
(see g. 4.3). induction step: Inductional assumption: Equation 4.3 is true for Use the product rule to differentiate the basis function:

Substitute equation 4.3:

4 B-splines
Because of

34

, we obtain:
Example: Support for cubic B-splines:

. . .
. . .

. . .
4.2
B-spline Curves
Denition 4.2 Given: degree knot vector control points
or .
(4.4)
is called B-spline curve. The points control polygon.

1
are called control points or de Boor points 1. They form a
Carl de Boor, 1972
35 4.2.1 Shape Properties of B-spline Curves
4.2 B-spline Curves
Theorem 4.1 (Shape properties of B-spline curves) a B-spline curve of degree Let following properties hold true: (1) In general there is no endpoint interpolation.
over the knot vector .
Then the
(2) For , lies in the convex hull of the control points (local convex hull).
Figure 4.5: Local convex hull
(3) Local control: For
(4) If control points coincide, then the curve goes through this point and is tangent to the control polygon (Fig. 4.6).
the curve is independent of
for
and
Figure 4.6:
control points coincide
(5) If control points are on a line, then the curve touches this line (Fig. 4.7).
Figure 4.7:
control points on a line
4 B-splines
(6) If control points are on a line , then for segment of the curve coincides with (Fig. 4.8).
36
, i. e. a whole
Figure 4.8:
control points on a line
(7) Derivative:
mit
(4.5)
(8) If knots coincide, then a control point and is tangent to the control polygon. (9) If is an afne map, then:
, i. e. the curve goes through
(afne invariance).
(10) Variation diminishing property: If is a hyper plane in , then
control polygon
4.2.2
Comparison of B-Spline and Bzier Curves
Bzier Curves + control polygon gives an impression of the possible run of the curve + the run of the Bzier curve can be changed by the translation of control points but: control points have global inuence on the run of the Bzier curve more control number of control points depends on the degree of the Bzier curve points, i. e. more possibilities to inuence the run of the curve, only possible by degree elevation
37 B-Spline Curves
4.3 Control Points and Polar Forms
B-Spline curves do not have the disadvantages of Bzier curves. Geometric properties of Bzier curves like "convex hull" or "variation diminishing" are also valid for B-Spline curves. Additional advantages are: + B-Spline functions are dened locally + Translation of a control point has only local inuence on the run of the curve + Insertion of additional control points without degree elevation + Smoothness of the junction of adjoining segments can be inuenced "easily"
Basis Functions polynomial piecewise polynomial Bernstein polynomials
Curves Bzier curves B-spline curves
(normalized) B-splines over knots
Table 4.1: Review: curve schemes mentioned so far
4.3
Control Points and Polar Forms
Theorem 4.2 (Control points and forms) polar of degree over a knot vector Given: a B-spline curve . restricted to the interval is a polynomial curve and has a unique polar form . Then:
for
(4.6)
e. g. for
this means:
Special case: The Bzier representation of the curve segment over the interval can be obtained as usual by inserting the interval borders und into the polar form:

An application of this property is the conversion of a B-spline curve into a Bzier spline curve.
4 B-splines
Example: . In this segment,
38
. Consider the curve segment where is the polar form of the B-spline curve . Control points:

Bzier points of
Given: Find:
The Bzier points are obtained by construction of suitable afne combinations, e. g.

Similar computation for the other control points.
4.4
Multiple Knots
Multiple Knots are allowed, if the multiplicity of the knots is less or equal to the degree. Generalized knot vector:
(4.7)
Exception: The rst and last knot may have the multiplicity . We have to extend the denition of the B-spline basis functions to the case of multiplicities:

otherwise
(4.8)
The recursion formula reduces B-splines of degree
to B-splines of degree
39 (1) case
4.4 Multiple Knots
, i. e. the multiplicity of ist greater or equal to

(2) case
(4.9)
.
, i. e. the multiplicity of is greater or equal to

(3) case
(4.10)
, i. e. the multiplicity of is greater or equal to

Example:
(4.11)
Figure 4.9: B-splines over a knot vector with multiplicities (emphasized:
Using a knot vector with multiple knots, some B-spline basis functions can vanish: Example: Quadratic B-spline basis functions over does not exist!
Remark: The higher the multiplicity of the knots the smaller is the continuity (see Theorem 4.3).
Theorem 4.3 (Curry-Schoenberg) . These knots have multiplicities Given: a sequence of knots and the border knots , where the inner knots have the multiplicity
4 B-splines
have the multiplicity knots
40
The normalized B-splines of degree with respect to this knot vector make a basis of the follow -continuous ing space: is a polynomial of degree and is at the value , .
. Now consider this knot vector as a sequence of single
with the corresponding multiplicities
Remark: The most important special case are the proper splines, where
If we insert a new knot , the dimension of the space increases by 1, i. e. the space of all piecewise polynomial functions over the old knot vector is a linear subspace of that over the new knot vector.
Multiple Knots and Endpoint Interpolation The curves are to be at least -continious. So from Theorem 4.3 follows, that the maximal multiplicity of the knots may be . At the end points we can equate , because the curve does not depend on (see scheme on page 34: For , does not contribute to the computation of , but does). Therefore the maximum multiplicity allowed at the endpoints is . Theorem 4.4 (Endpoint interpolation) If an end knot has multiplicity , then the B-spline curve interpolates the corresponding control point and is tangent to the control polygon. Sketch of Proof: Consider a B-spline curve of th degree over the knot vector
Similar, it can be shown that Because of and Tangential property:
For the basis function for we obtain: This is the rst Bernstein polynomial of th degree with respect to the interval with .
with . the end points are interpolated.
since does not exist and . All the other basis functions must be zero at the parameter value (afne combination). So the line is the tangent to the Bzier curve at control point . Similar for .
41 Example:
4.5 Algorithms for B-spline Curves
. over this knot vector that lives on three not
We construct a piecewise polynomial curve vanishing parameter intervals:
control points: control points: control points:
Consider the interval
. B-spline control points that affect
are:
The Bzier representation of
is obtained as follows:

4.5
Algorithms for B-spline Curves
In this section,
is a B-spline curve of degree , where multiple knots are allowed.
over a knot vector
4.5.1
The de Boor Algorithm
The de Boor algorithm is a method for evaluating B-spline curves. Given: control points of a B-spline curve and knot vector and a parameter intervall . Find: value of at
4 B-splines

42
Figure 4.10: Bzier representation of a B-spline curve with endpoint interpolation
Algorithm 4.1 (de Boor)

with Then:
and
(4.12)
The evaluation scheme for the de Boor algorithm is depicted in Figure 4.12.The coefcients can be visualized as divide ratios of line segments (see Figure 4.11).

Figure 4.11: Evaluation schema of the de Boor algorithm for
Remark: The de Boor algorithm is the analogue to the de Casteljau algorithm for Bzier curves.
43

Figure 4.12: Visualization of the coefcients for the de Boor algorithm (
Example: (see g. 4.13, 4.14 and 4.15).

Figure 4.13: Example for the de Boor algorithm (1): coefcients
Now consider the following knot vector:
Here:
, and have all the multiplicity . Then:
This is the de Casteljau algorithm! Corollary 4.1 As we only consider afne transformations, assumption of the interval loss of generality.
is not a
4 B-splines
44
Figure 4.14: Example for the de Boor algorithm (2): evaluation
Figure 4.15: Example for the de Boor algorithm (3): graphical construction
45

If two neighbouring knots have the multiplicity , then the corresponding B-spline curve is a Bzier curve between these knots. By insertion of all knots until multiplicity we can construct the Bzier polygon.
4.5.2
Conversion of B-spline Curves into Bzier Splines
Given: B-spline curve
over a knot vector

Find: Bzier representation for each polynomial segment There are two possibilities: (1) Insert knots and (2) Use the polar form. Example: Knot vector: The control points
,
until both have the multiplicity .
. Consider interval .
are given.
To nd the Bzier points combinations, e. g.
and , construct suitable afne
4.5.3
The Boehm Algorithm
How can we insert a new knot into a knot vector without changing the curve? Insertion of knots is useful for: adding local detail as basis of other algorithms (e. g. de Boors algorithm for curve evaluation) to convert B-splines into Bzier splines. Insertion of an additional knot , lets say at .
, gives a new knot vector
With the theorem of Curry-Schoenberg (theorem 4.3) we get: every B-spline curve knot vector has also a unique representation over the knot vector .
over the
4 B-splines
How to compute the new control points algorithm. With theorem 4.2, we get from the given
46 ? An answer is given by the Boehm2
and also

. Then:
By construction of suitable afne combinations we get e. g.
The other control points are obtained similar. More general: Algorithm 4.2 (Boehm)
(4.13)
The coefcients can be visualized as divide ratios of line segments. In the following schema, there is , and .
Figure 4.16: Schema for the Boehm algorithm
Wolfgang Boehm, 1980
47 Example:
Figure 4.17: Example for the Boehm algorithm
Remark: According to Theorem 4.3 we can insert inner knots up to multiplicity to mulitplicity .
and at the end points up
Insertion of an existing knot (i. e. increasing the multiplicity of the knot to ) gives only new control points.
Insertion of knots until multiplicity makes the B-spline curve interpolate control points. This gives the de Boor algorithm to evaluate B-spline curves.
4.5.4
Knot Deletion
Knot deletion is the inverse process to knot insertion. In general, it is not possible to remove a knot without changing the curve. Therefore, try to remove a knot and nd a curve that is as close as possible to the original curve (see degree elevation degree reduction for Bzier curves). Knot insertion is a linear process:

such that
(4.14)
. Find control points Short: have to solve the linear system
is minimal. With theorem 3.6 we (4.15)
4 B-splines
and get
48
4.5.5
Subdivision for uniform B-spline curves
Intention: Create a series of polygons that converge fast to the curve. The subdivison of Bspline curves is achieved by insertion of many additional knots or by doubling the number of knots. Let
is called uniform knot vector. Insert new knots at
with
a B-spline curve of degree over the knot vector for every or simply . A knot vector with this property
Then:
Repeating this process gives a sequence of polygons that coverges fast to (compared with subdivision for Bzier curves). are obtained again with suitable afne combinations. For this gives Chaikins 3 The algorithm. Example 1: and knot vector with multiplicity at the endpoints Knots are inserted where the curve has full support, in this example at
over the new knot vector .
Figure 4.18: Knot insertion at

3
G. Chaikin, 1995
49
According to Figure 4.18 we obtain the following new control points:

Therefore we can calculate the new control points from old control points as follows:

for for
The graphical evaluation is also called "Corner Cutting". You can see the result after two steps of iteration in Figure 4.19.

Figure 4.19: Geometrical construction: cutting the corners
Example 2: and open knot vector e.g. Evaluation according to the following scheme (see Figure 4.20):

for for
4 B-splines

50
Figure 4.20: Subdivision with open knot vector and
For the algorithm is more complicated, as new control points can depend on up to three old control points: Example 3: and knot vector with multiplicity at the endpoints, control points Evaluation according to the following scheme:

for for
Example 4: and open knot vector, control points Evaluation according to the following scheme:

for
for
51
4.6 Interpolation and Approximation
4.6
Given:
Interpolation and Approximation
Remark: The given points are to be interpolated / approximated at the parameter values . Imagine the knot vector is the time axis and the parameter the time: The curve we want to nd, should run through the point at time (see Figure 4.21).

data points with corresponding parameter values an -dimensional space with basis functions over a knot vector Find: Curve that interpolates / approximates the data points
Figure 4.21: Illustration of parameter values
There are three cases to distinguish: (1) (2) (3)
tion/approximation 4.6.1 Given:
: number of conditions = number of degrees of freedom interpolation : overdetermined system of equations approximation : underdetermined system of equations "constrained" interpola-
B-spline Interpolation
In matrix notation:
points with parameter values an -dimensional space: with knot vector Find: control points such that the curve , i. e. the given points at the parameter values
interpolates .

. . .
. . .
. . .
. . .
(4.16)
4 B-splines
Therefore, the interpolation problem is always solvable in case the matrix ible.
Remark: Matrix has at most non-zero entries per row (support of :
52 is invert-
).
Matrix is a band diagonal matrix. There are efcent numerical methods to solve a linear system consisting of a band diagonal matrix and a vector.
Theorem 4.5 (Schoenberg-Whitney)
Matrix A is invertible
Proof: See [1]. A choice of the parameter values Theorem 4.5 are the Greville abscissas. Denition 4.3 The parameter values
Remark: For the Greville abscissas holds:

that always fullls the requirements of
are called Greville abscissas.

have multiplicity
fullls , if all inner knots in .
. Reason for this:
Applications typically only give the , but no parameterization. In this case, choose a parameterization that is suitable for the application.
4.6.2 Given:
Special case: Cubic spline interpolation at the knots
degree
a -dimensional space
data points knot vector
Find: cubic B-spline curve That means: we choose those knots as parameter values that lie in the area where the curve has full support. Then the conditions of Theorem 4.5 are fullled. But the number of conditions is smaller than the number of degrees of freedom. Hence for the uniqueness of the curve we need additional conditions.
over with
53 Possible additional conditions: (1) Specication of tangents at the end points: (2) Natural end condition:

):
(3) In the case of closed curves (i. e. The degrees of freedom are used to create
-continuity at the point
Example 1: Interpolation at the knots of an arbitrary knot vector by a cubic spline with additional condition 2
Linear system to solve:

.. .

.. .
..
. . .
. . .
Example 2: Interpolation at the knots of an end point interpolating knot vector
by a cubic spline with additional condition 2

4 B-splines
54
analogue:
The linear system that we have to solve is derived by modication of the linear system of Example 1:

.. .
..
. . .
. . .
Example 3: Interpolation at the knots of an end point interpolating and uniform knot vector by a cubic spline with additional condition 2 The linear system is:
with
..
..
. . . . . . . . .
. . . . . . . . .
Remark: In the case of end point interpolating splines the matrix
is tridiagonal (see Example 2 and 3).
4.6.3
B-spline approximation
If there are given more data points than control points available in the -dimensional space, is overdethen interpolation of the data points is usually impossible. The linear system termined and therefore not solvable.
55
We can use the Gauss Normal Equation (Theorem 3.6) to minimize the error it, to approximate the data points at the best:
and with
derivation
is minimal
Thus, the approximation has exactly one solution if and only if the matrix
exists
is invertible:
Theorem of Schoenberg-Whitney is fullled columns of are linear independent
4.6.4
Parametrization
An important item of interpolation and approximation of data points is the choice of the parameter values (parametrization). By choosing a proper parametrization great deections and loops of the curve can be avoided in some degree. The easiest way to determine the is to set . This is called the uniform or equidistant parametrization. The disadvantage is the disregard of the position of the data points. For this reason the interpolating curve will have unwanted loops and deections in most cases. Interpretation: Imagine the parameter value is the time. Because of the -continuity the speed and the acceleration is continuous. On account of the uniform parametrization large distances between data points must be managed in the same time as distances between data points that are close to each other. Therefore the speed is higher between points with large distance, but speed and acceleration are continuous and we receive loops and deections of the curve between data points with short distance. One way to take the positions of the data points into account is to choose the distance of the parameters proportional to the distance of the data points:
with
uniform for This parametrization is called
centripetal for chordal for
The chordal parametrization is depicted in Figure 4.22: The distances between the parameter values and are equal to the distances between the data points und . Afne transformation does not change the parametrization. Hence, the parameter values can determined as follows for a given interval :
for
4 B-splines

56
Figure 4.22: Chordal parametrisation Remark: It is also possible to choose the knot vector case:
depending on the parameter values . Cubic
57
5 Rational Curves
A more general approach are rational curves. Using the polynomial curves we have examined so far it is not possible to represent conic sections (e. g. circle arcs). This becomes possible if we extend the schemes to rational curves. Of special interest in this chapter are rational Bzier curves and rational B-spline curves (NURBS). A rational curve can be described in two ways: as a quotient of polynomials as central projection of a -dimensional curve onto a -dimensional plane. Example: A two-dimensional point is extended by another dimension to and can be represented by a line through the origin:
Figure 5.1: Homogeneous coordinates
By projection onto the plane
(division by ) we get back the 2D point.
Remark: This representation is called homogeneous coordinates. The aim is to represent -dimensional afne maps as linear homogeneous transformations (i. e. matrices) in the -dimensional space.
5.1
Rational Bzier Curves
Denition 5.1 (Rational Bzier curve) is the projection of a polynomial Bzier curve in A rational Bzier curve in Denition 5.2 (Rational Bzier curve, analytic denition) Given: control points
(e. g. )
(projections of the -dimensional control points
5 Rational Curves
58
Figure 5.2: Interpretation of rat. curves with homogeneous coordinates (proj.
weights
(homogeneous component of the control points) We can assume without loss of generality: Then:
(5.1)
is called rational Bzier curve over the interval . 5.1.1 Let Properties of rational Bzier curves
a rational Bzier curve and
(1) If all (2) (3)
, then we get the well known polynomial Bzier curves.
(convex hull property)
(endpoint interpolation)
(4) The curve is tangent to the control polygon in the endpoints. (5) Afne (even projective) invariance afne yields:

I. e. (6)
shows the variation diminishing property with respect to
59 5.1.2 Evaluation of rational Bziercurves
5.1 Rational Bzier Curves
The de Casteljau algorithm in homogeneous coordinates can be used to evaluate and subdivide the curve. 1st possibility: Evaluate de Casteljau in the next higher dimension and project the result onto the -plane. The computation is done according to the scheme in Figure 5.3.
(5.2)
Figure 5.3: de Casteljau scheme for rational Bzier curves
Problem: The Algorithm is numerically unstable if the weight differ very much. 2nd possibility: Rational de Casteljau algorithm

with 5.1.3
for

(5.3)
Inuence and properties of the weights
Theorem 5.1 (Effect and properties of the weights) the rational Bzier curve is equal to a polynomial Bzier (1) For curve. (2) Multiplication of all weights with a common factor does not change the curve. (3) More general: If for all ratios
const.
then the curve has the same shape but another parametrization. (4) Increasing the weight pulls the curve to the control point . Decreasing the weight pushes the curve away from the control point (see Figure 5.4).
5 Rational Curves
(5) In the limit case:
60
(5.4)
Remark: From the properties (2) and (3) follows that we can assume without loss of generality that (proof given in the exercices).
Figure 5.4: Effect of the weights (
Supplement: Conic Sections Denition 5.3 (Conic section) analytic: given a quadratic polynomial

quadratisch linear
konstant
then a conic section is dened as the point set

i. e. a conic section is the geometric: section of a plane with a double cone. The coefcients determine the type of the conic section.
The type of the conic section can be seen directly from the determinant of this matrix. Graph ellipsis parabola hyperbola Angle between cone axis and plane greater than cone opening equals cone opening less that cone opening determinant of

greater than 0 equals 0 less than 0
61
Figure 5.5: Conic sections: ellipsis, parabola und hyperbola
Example:

unit circle standard parabola hyperbola
Conic sections can not be represented as polynomials but as rational functions. Theorem 5.2 Each rational quadratic Bzier curve is part of a conic section. For we have: ellipsis parabola hyperbola
and
Each part of a conic section can be represented as a rational quadratic Bzier curve. 5.1.4 Quadratic rational Bzier Curves as Conic Sections
over with

Theorem 5.3 Given: a rational quadratic Bzier curve

(5.5)
Choose coordinate system
such that
5 Rational Curves
Then:
is part of a conic section with the dening equation
62
Proof: Let
(5.6)
Thus, in the coordinate system
we have
and additionally:

(2) From equation 5.6 follows

and therefore
ellipsis parabola hyperbola
5.1.5 Given:
Segments of Conic Sections as quadratic rational Bzier Curves
conic section
63

two points on the same component of branch of a hyperbola) with intersecting tangents. (e. g. on the same
Find: Representation of the corresponding segment of a conic section as a rational quadratic Bzier curve . Procedure: (1) Choose (2) Choose

as given.
as intersection point of the tangents and :
Intersection point: solve
(3) Express the given polynomial
in the coordinate system
with
Inserting this into gives a new polynomial (also of degree 2) with . describes the conic section in the coordinate system . (4) Determine the weight using equation 5.6
solutions
These are two equations for two variables comparison of the coefcients gives the

Example: Unit circle:
(1)
5 Rational Curves
64

(3)
(2)
(4) Comparison of coefcients to determine and :
Remark: For a circle arc of an arbitrary opening angle
we get
5.1.6
Rational Bzier-Spline Curves
A rational Bzier-spline curve is a curve that is composed of single rational Bzier curves. Example: Representation of the circle with four segments. Control points:

The weights are and (see example in the previous section). The corresponding 3D curve is the intersection curve of a cone (opened at top) with a pyramid (opened at bottom).
Figure 5.6: Intersection of a cone with a pyramid and projection of the intersection line Remark: The higher dimensional curve is not smooth while the projected curve is smooth. This is quite typical for rational spline curves. By projection one can gain differentiability!
65
5.2 Rational B-Spline Curves (NURBS)
Theorem 5.4 and with control points Given: Rational Bzier curves , and weights , . Additionally, set . The curve composed from and is continuously differentiable
(5.7)
Geometric interpretation: . Proof: We have can summarize:
divides the line segment from to
by the ratio
and
Because
, we
5.2 Rational B-Spline Curves (NURBS)
no local support rational polynomial rational Bzier curves Bzier curves
local support NURBS B-splines
Table 5.1: Summary: polynomial and rational curve schemes
NURBS = non-uniform rational B-spline. Denition 5.4 (Rational B-spline curve) A rational B-spline curve in is the projection of a B-spline curve in Denition 5.5 (Rational B-spline curve, analytic denition) Given: degree , knot vector , control points weights

with
5 Rational Curves
Then:
66
(5.8)
is called rational B-spline curve. Advantages: compared with rational Bzier-spline curves: of th degree
-continuity of rational B-spline curves
compared with B-spline curves: more degrees of freedom Disadvantages: is linearly dependent of the , but not of the . For tasks of approximation and interpolation NURBS are improper, because you cant build a linear system of equations.
67
6 Elementary Differential Geometry for Curves

We are not (really) interested in the parametric representation of curves but in the shape of a curve. What characterizes shape?
6.1
Reparametrization
Denition 6.1 (Regular curve) a differentiable curve . Then: is called regular Let is the (unique) tangent direction at the point . all .
for
Now we consider a reparametrization or parameter transformation of a regular curve:
Denition 6.2 (Equivalence of curves) Let and regular curves. Then: ) there exists a differentiable mapping (notation: all such that .
and
are called equivalent with for
Example:

Example: Arc length
Let
with , then:

a regular curve. If we denote the length of the curve segment
(6.1)
This leads to a nice parametrization (arc length parametrization). In this special case we have for all and therefore

for all
68
Figure 6.1: Arc length
Example: Circle:
Parameter transformation:
yields
Theorem 6.1 Every regular curve can be reparametrized by arc length parametrization. This can be achieved by the following parameter transformation: Let
then
(6.2)
is a reparametrization according to the arc length. Example: In the previous example we had
, thus
Remark: Bzier curves are rarely arc length parametrized.
6.2
Planar Curves
Denition 6.3 (2D Frenet frame) a regular parametric curve. Given: Let The local coordinate system
with
(6.3)
rotated by
(6.4)
is called 2D Frenet frame1 of

1
Jean Frdric Frenet (1816-1900)
69

6.2 Planar Curves
Figure 6.2: The 2D Frenet frame
Theorem 6.2 (Frenet) For the derivatives of and
the following holds:

The quantity
is called curvature of the curve.
For arc length parametrized curves:

curve bends to the left curve bends to the right
Sketch of Proof: The theorem states that and (respectively und ) are perpendicular; const., because from this is the case if and only if follows and const. This is true by the normalization of .
Remark:

is the radius of the osculating circle with the midpoint
Theorem 6.3 For a curve point
it holds true that
(6.5)
Example: Ellipsis:
70
Theorem 6.4 The curvature characterizes the curve uniquely short of rotation and translation. Additionally:
is a line segment. const.
is a circle arc (with radius
).
Remark: Distinguish between geometric and parametric properties of a curve. Geometric means independent of the parametrization. Example: geometric unit tangent vector Frenet frame curvature
curvature vector
tangent vector all derivatives
parametric
6.3
Space Curves
Denition 6.4 (3D Frenet frame) Given: curve . The local coordinate system

with
unit tangent vector main normal vector binormal vector (6.6)
shows the following properties:
is orthonormal

with the same orientation
is positively oriented (right handed system)
Theorem 6.5 (Frenet-Serret) For the derivatives the following holds:
(6.7)
71
6.3 Space Curves
Figure 6.3: The 3D Frenet frame
In matrix notation:
Denition 6.5 (Curvature and torsion)
curvature: torsion: Theorem 6.6 Formulas to compute
(6.8) (6.9)
and :
quantity

arc length parameter
arbitrary parameter
Example: Spiral: Derivatives:
(6.10)
72
Figure 6.4: Spiral
Remark: The osculating plane is curves). The normal plane is

(see osculating circle for planar
We have assumed throughout that )
are linearly independent! (i. e.
Theorem 6.7 (Base theorem for space curves) Curvature and torsion characterize a space curve uniquely short of rotation und translation. Additionally:
is a planar curve.
6.4
Geometric and Parametric Continuity

and are called (parametric)
.
Remember: continuity of composed Bzier splines Two polynomial curves at , if
-continuous
73
6.4 Geometric and Parametric Continuity
Figure 6.5: Continuity of Bzier splines
Denition 6.6 (Geometric continuity) Two polynomial curves and at , if the arc length parametrizations and
Remark:
are called (geometric)

are -continuous.
-continuity does not!
-continuous
-continuity depends on parametrisierung ab,
Theorem 6.8 (Charakterization of geometric continuity) The following statements are equivalent: (1) and are or -continuous can be reparametrized such that and (or and ) are -continuous.
(2) either
(3) There exist constants

(Beta-constraints) such that
(6.11)
Proof: Let and parameter transformations such that

: continuous.
and : Assume:

are and
-continuous
F and

und
. are

. . .

are
-continuous at .
Note that
In matrix notation:

. . .

. . .

..

(3)
74
(2): Dene locally by

Remark: Normally, of interest are only:

fr
6.4.1
Geometric Continuity of Bzier Curves and the interval lengths
Given: Bzier curves
and
with
and
and
Theorem 6.9 (Characterization of Equivalent are: (1) (2) and and are
-continuity (tangent continuity))
-continuous
have the same tangent direction (i. e. the same tangent unit vector) at
(3) divide ratio:
Sketch of Proof: (1) (1)
(3): use equation for -constraints and
(2): follows from
Theorem 6.10 (Characterization of Equivalent are: (1) (2)
-continuity (curvature continuity))
and and
are
-continuous (
have the same unit tangent vector and the same curvature vector at )
(3) There exist constants holds.
with
such that the generalized A-frame property
75
6.4 Geometric and Parametric Continuity
Figure 6.6: Generalized A-frame property
Sketch of Proof: (1)
and and for the control points

(3): Use -constraints:
generalized A-frame condition:

respectively
76
77
7 Tensor Product Surfaces

7.1 General Approach
Given: two curve schemes in 3D space:
and
with scalar basis functions and . The tensor product of these schemes is:

2 1 0 -1 1
(7.1)
-1 -1 0 1 2
Figure 7.1: A tensor product Bzier surface
Of special interest are: Denition 7.1 (Tensor Product Bzier Surface)
(7.2)
Denition 7.2 (Tensor Product B-Spline Surface)

(7.3)
with knot vectors
and

Remark: We can factorize:
78
Each xed gives a control point of the curve
The tensor product surface is a curve of curves!
With the same approach we get rational tensor product Bzier and B-spline surfaces: consider a TP Bzier or a TP B-spline surface in 4D space (homogeneous coordinates) and divide by the component. This gives:
Denition 7.3 of a tensor product surface is called the control net of the surface The array of control points (analogon of control polygon for curves).
Figure 7.2: Parameter grid, control net and resulting surface
7.2
Tensor Product Bzier Surfaces
Given: a tensor product (TP) Bzier surface
Theorem 7.1 (Shape properties of TP Bzier surfaces) (a) Convex hull property: convex hull for
(b) Interpolation of the four corners of the control net:
79
7.3 Algorithms for TP Bzier Surfaces
(c) Tangency in the corner points, e. g. the tangent plane at and .
is spanned by
(d) The boundary curves are Bzier curves, their control points are the boundary points of the is Bzier curve). control net (e. g. (e) Afnce invariance (f) Variation diminishing property does not hold! Sketch of Proof: afne invariance: combination:
must be a barycentric (afne)
convex hull:
and
is a convex combination.
7.3
7.3.1 Given:
Algorithms for TP Bzier Surfaces

Evaluation with the TP Approach
TP Bzier surface parameter values , Proceeding:
(1) Apply the univariate de Casteljau algorithm in -direction times:

(2) Apply the univariate de Casteljau algorithm in -direction once:
or rst in -direction, afterwards in -direction. The evaluation with the TP approach rst in -direction, then in -direction is depicted in Figure 7.3. Choice of the best order of evaluation (smallest number of needed arithmetic operations): If
: rst in -, afterwards in -Richtung If : rst in -, afterwards in -Richtung
80
Figure 7.3: de Casteljau scheme for tensor product Bzier surfaces
7.3.2
The direct de Casteljau Algorithm
Idea: use bilinear interpolation instead of linear interpolation. Theorem 7.2 (Bilinear interpolation formula)
(7.4)
Figure 7.4: Bilinear interpolation
The bilinear interpolation formula (see Figure 7.4) is applied to the whole control net in the direct de Casteljau Algorithm (see Figure7.5). Therefore the evaluation scheme looks like a three-dimensional pyramid.
81 Algorithm 7.1 (Direct de Casteljau)
7.3 Algorithms for TP Bzier Surfaces
(7.5)
Then:
Figure 7.5: Direct de Casteljau scheme
Problem: If you have to use the univariate de Casteljau algorithm on from a certain step. This makes the implementation considerably more difcult. Comparison of direct de Casteljau and TP approach according to the number of convex combinations (CC): TP approach CC CC
direct de Castejau CC CC
In practice the TP approach is used, because it is more efcient and you dont have to do any distinction of cases.
Remark: Other algorithms like degree elevation or subdivision can be derived from the corresponding algorithms for curves.
82
7.4
Given:
Tensor Product B-Spline Surfaces
knot vector knot vector
array of control points
surface

Theorem 7.3 (Shape properties of TP B-spline surfaces) (a) Local convex hull property:

convex hull
(b) Afne invariance (c) Endpoint interpolation: if
then
and
In this case: tangency in the corner points, and the boundary curves are B-spline curves with the boundary points of the control net as curve control points. The tangential plane at the corners is spanned by the control points, e. g. .
(d) Local control: a control point has inuence only on a part of the surface. Changing will change
only for
(e) Variation diminishing property does not hold!

Remark: Remember from curves: a Bzier curve is a special case of a B-spline curve. A Bzier curve over the interval with the control points is a B-spline curve over the knot vector with the same control points. The same property holds true for tensor product surfaces.

83
7.5 Algorithms for TP B-Spline Surfaces
7.5
7.5.1
Algorithms for TP B-Spline Surfaces

Evaluation of TP B-Spline Surfaces
with knot vectors and . Given: The proceeding is similar to the TP approach of TP Bziersurfaces:
(1) Evaluate B-Spline curves of degree with de Boor in -direction. (2) Evaluate one B-Spline curve of degree with de Boor in -direction. 7.5.2 Knot insertion
The Boehm knot insertion algorithm for increasing modelling parameters (i. e. control points) in a local area will produce overhead, because the renement of the control net can only be done by adding new rows and columns.
Figure 7.6: Knot insertion with the Boehm algorithm
One alternative are hierarchical B-splines (by Forsey and Bartels, Siggraph88). In this solution, a B-spline surface is composed of a patch over a coarse parameter grid and, local detail dened over a locally rened parameter grid. You can see in Figure 7.7 that the grid is rened only where needed.
Figure 7.7: Building a hierarchical knot grid for B-spline patches
84
7.6
7.6.1 Be
Interpolation and Approximation

General interpolation

. Consider

as basis functions
Task: Interpolation/approximation of data points
with parameter values
Interpolation and approximation are exactly the same as for Bzier and B-Spline curves. Problem: For curves we have the Theorem of Schoenberg-Whitney, that says if the interpolation/approximation matrix is invertable. But there is no comparable theorem for surfaces.
7.6.2
Special case: Interpolation at the knots
Interpolation over grids: Given: data points with parameter values Find: surface
for with

. . .
. . .
. . .
. . .
. . .
. . .
. . .
. . .
The resulting matrix system is very large. For for example you have to do operations. Therefore the following proceeding is better:
(1) Solve (2) Solve
Tasks (1) and (2) are interpolation problems of curves, for whom the corresponding algorithms can be applied. Interpolation at the knots: If , then tasks (1) and (2) are tridiagonal matrices and therefore solvable in a fast way.
85
7.7 Trimming of parametric Patches
7.7
Trimming of parametric Patches
One disadvantage of tensor product Bzier and B-spline surfaces is that all patches are squarelike; it is not possible to model surfaces that does not have a well-dened tangent plane (i. e. if ).
Figure 7.8: Trimming a patch
A solution are trimmed patches. Specify a set of closed contours in the parameter are inner contours (see Figure 7.9). By domain, where is the outer contour and denition: shall belong to the trimmed surface if is inside with respect to and is outside with respect to
.
Figure 7.9: Specifying inner and outer contours in the parameter domain
How to determine if a point is inside or outside the trimmed surface (see Figure 7.10)? Theorem 7.4 Given: a closed contour is inside even.
a ray emanating from intersects the contour times with being odd. is outside a ray emanating from intersects the contour times with being
with
86
Figure 7.10: Determining inside and outside of a trimmed patch
87
8 Ray Tracing of TP-Bzier Surfaces (Bzier Clipping)

In this chapter, we want to discuss an algorithm to compute intersections between rational tensor product Bzier surfaces and rays. In particular, this is important for rendering Bzier patches using the ray tracing approach. For more information see Nishita et al., Siggraph90. Given: rational TP Bzier patch

(8.1)
Find: intersection, i. e. nd and such that
ray
Remark: This problem is analytically solvable only for low polynomial degrees. Therefore an iterative approximation algorithm as presented here is normally a better approach.
1st step: Represent the ray as an intersection of two planes:
Figure 8.1: Bzier clipping, step 1
Example: Select plane normal vectors that are perpendicular to and . Then determine :
, e. g. and by applying

2nd step: Find intersection of
88
We get
with the two planes:

with
and

Remark: This is a projection of
and
along onto a plane that is perpendicular to .
3rd step:
89 Choose two directions in space:
direction direction
perpendicular to direction, perpendicular to direction,

const. const.
In case of a biquadratic patch:
Alternate the directions and and iterate on the following steps: 4th step: Determine the (signed) distances of the projected control points to . This gives a Bzier representation of the distance-to- -function that can be used to determine the distance of an arbitrary point to :

5th step: Plot the values 6th step: Determine the convex hull of the points in this diagram (by connecting maxima and minima of the columns) and intersect it with the -axis. This gives an interval .
in a - -diagram (see gure 8.4)
90
Figure 8.4: Bzier clipping, steps 5 and 6
7th step: Clip away the parametric domain and Casteljau algorithm for Bzier patches).
by two subdivision steps (using the de
Switch direction and repeat steps 4 to 7. The parameter values of the . intersection point must be within the box
Stop when the size of the box is smaller than a threshold value or the convex hull in step 6 does not intersect with the axis; this indicates that there is no intersection of and .
91
9 Triangular Bzier Patches

The tensor product surfaces discussed in the previous chapter have been dened over rectangular parameter domains. Now we will consider Bzier patches dened over triangular parameter domains. There are two major advantages: triangles give more topological exibility tensor product surfaces have high polynomial degree, e. g. a biquadratic TP patch has actually degree 4 (see table below for all monomials occurring in a biquadratic TP patch).

9.1 Barycentric Coordinates in
Normally, a point is given by an origin and two directions and such that . We call and the standard coordinates of with respect to the coordinate system . Now we introduce another type of 2D coordinates based on three points: Denition 9.1 (Barycentric coordinates) Given: Three points , and that are not on a line. An arbitrary point can be represented as an afne combination
The coefcients .
(9.1)
are called barycentric coordinates of with respect to the triangle
Theorem 9.1 Given: Triangle and point . The barycentric coordinates to are unique and can be determined by

of with respect
(9.2)
where


Sketch of Proof: Equation 9.1 can be rewritten as a linear system with tree variables:
92
So we have

and are linear dependent.

The formulas for and follow from Cramers rule.
Figure 9.1: Barycentric coordinates Remark: Because of coordinates also as ratios of areas, i. e.

we can represent the barycentric

is inside
and
If
then is the barycenter of
9.2
Bernstein Polynomials over Triangles
A 1-dimensional analogon to barycentric coordinates: if divides the interval such that then and . th The Bernstein polynomial of degree over the interval is dened as follows:
93
9.2 Bernstein Polynomials over Triangles
Figure 9.2: Divide ratios and barycentric coordinates
Denition 9.2 (Bernstein polynomial over triangles) Given: Point having barycentric coordinates with respect to a triangle , a triple of indices with degree and . Then:
(9.3)
is called a Bernstein polynomial over the triangle . Example:

Figure 9.3: Arrangement of indices
for
Bernstein polynomials
Theorem 9.2 The Bernstein polynomials of degree ).
form a basis of (= bivariate polynomials
Sketch of Proof: The number of Bernstein polynomials is , which is the dimension of bivariate polynomials of degree . Example: for we have 10 linear independent polynomials: . Theorem 9.3 (Properties of triangular Bernstein Polynomials) . (1) If is inside , then

(2) Partition of 1:
94
(9.4)
(3) Recursion:

(9.5)
Sketch of Proof: (1) follows from denition; (2)
if is inside
(3)
Similar for
and
Theorem 9.4 (Polar Forms in 2D) of degree there exists a unique polar form For each polynomial with is symmetric is multiafne, i. e.

is diagonal, i. e.

Example:
95
9.3 Triangular Bzier Patches
9.3
Triangular Bzier Patches

with
Denition 9.3 (Triangular Bzier Patch) Given: degree , parameter domain , points
. Then:
(9.6)
is called triangular Bzier patch over the triangle . The points are called control points or Bzier points. They form the control net.

Figure 9.4: Triangular control nets for
and
Theorem 9.5 (Control Points and Polar Form) Given: triangular Bzier patch
. Then: (9.7)
Sketch of Proof: Let Assume that , then
the barycentric coordinates of such that

Theorem 9.6 (Shape Properties of Triangular Bzier Patches) for (a) convex hull property: convex hull of (b) end point interpolation:

,
.
.
(c) end point tangency: e. g. the tangent plane at is
(d) The boundary curves are Bzier curves formed by the boundary control points. (e) afne invariance
96
9.4
9.4.1
Algorithms for Triangular Bzier Patches

The de Casteljau Algorithm
, we need the barycentric coordinates
To evaluate
of .
Algorithm 9.1 (de Casteljau for Bzier triangles)
(9.8)
Then:
with
Sketch of Proof: According to theorem 9.5, we have
This is a solution for the recursion formula, it leads to
Figure 9.5: de Casteljau scheme for triangular Bzier patches
9.4.2
Degree Elevation
As an example, we rst consider here the degree elevation from quadratic to cubic triangular Bzier patches:
To nd the
, we need a mapping for the polar forms

:
(9.9)
97
9.4 Algorithms for Triangular Bzier Patches
It is easy to see that this is symmetric, multiafne and diagonal, so Example:

is a polar form.
Algorithm 9.2 (Degree elevation) Given: Bzier triangle

Find: control points
such that
(9.10)
9.4.3
Subdivision
into two triangles (1-to-2 split).
There are two possibilities to perform a subdivision for triangular Bzier patches: 1st possibility: divide E. g. dividing the line segment with gives

2nd possibility: divide
into four triangles (1-to-4 split, see gure 9.7).
98
Figure 9.7: Splitting triangles: 1-to-2 and 1-to-4 split
Figure 9.8: A control net and the rst 1-to-4 subdivision step
Algorithm 9.3 (1-to-2 split) Perform de Casteljau algorithm:
for . Then: the outer faces of the resulting tetrahedron are control points of the two subdivided triangles (see gure 9.5). Algorithm 9.4 (1-to-4 split, Farin-Prautzsch method) Perform four 1-to-2 splits with de Casteljau (see gure 9.9). Instead of subsequently using the de Casteljau algorithm for subdivision, it is also possible to pre-calculate the split points using polar forms:
99
9.4 Algorithms for Triangular Bzier Patches

Figure 9.9: Farin-Prautzsch 1-to-4 subdivision scheme
Example: If we want to split at
and
we get:
Supplement: Derivatives of Multivariate Functions

Considering multivariate (here: bivariate) functions, there are three types of derivatives: partial derivatives are and directional derivative: the derivative of is
at a certain point in direction
total derivative: the Jacobian matrix
For bivariate functions, the Jacobian matrix is a matrix with the dimension of the . space
100
9.5
-Continuity of Triangular Bzier Patches

and which have the parameter domains
and
Given: two Bzier triangles .
Figure 9.10: Attaching two Bzier triangles continuously
How to determine that and are -continuous along the edge ? Assume a point (and assume that and are also given in on the edge in cartesian coordinates cartesian representation).
: : :
and and
and
for all
and .
In general:
-continuity can be characterized using polar forms. Assume .

: :
and are polar forms of
and
. . .
. . .
101
9.5
-Continuity of Triangular Bzier Patches
for all . An equivalent condition for tions hold for .

:
-continuity:
holds and the above equa-
and .
for all In general:
. Equivalent:
holds and
Theorem 9.7 and with the parameter domains The Bzier triangles are -continuous if
and
(9.11)
for all
and
.
):
Geometric interpretation (for
is a representation of in barycentric coordinates with respect to . Then:

Figure 9.11: Diamond condition for
These quadrilaterals (highlighted in gure 9.11) are afne images of the parameter quadrilateral , therefore they have to be planar (diamond condition).
102
9.6
Interpolation with piecewise Bzier Triangles (Split Surfaces)
One application of triangular Bzier patches is the interpolation of scattered data points using smooth triangular surfaces. and corresponding data values . Given: data points Find: a smooth (i. e. . General procedure: 1st step: triangulate the set of data points, e. g. using Delaunay triangulation (see lecture on Graphical Algorithms for details). 2nd step: split each triangle. We want to discuss here two methods of splitting, the Clough-Tocher method that gives three cubic Bzier patches and the Powell-Sabin method that gives six quadratic Bzier patches. 3rd step: estimate the normal vectors (or partial derivatives) at each data point. These are used to determine the control points. 9.6.1 Clough-Tocher
) scalar function
that interpolates the given data, i. e.
Note that there exist many approaches to solve this problem.
Figure 9.12: Clough-Tocher split surface scheme
This is how the control points in gure 9.12 are obtained:
: Assume that the orthogonal derivative along an edge is linear and linearly interpolate
: specied by the directional derivatives obtained in 3 rd step. the orthogonal derivative at the end points.
: given data points .
103
9.6 Interpolation with piecewise Bzier Triangles (Split Surfaces)
: diamond condition at the interior edges for
: choose the barycenter of the triangle. 9.6.2 Powell-Sabin
-continuity
Figure 9.13: Powell-Sabin split surface scheme
Here, the control points according to gure 9.13 are obtained as follows:
: diamond condition at the outer edges for
: given data points .
: specied by the directional derivatives obtained in 3 rd step.
-continuity
: diamond condition at the interior edges for
: barycenters of the given triangles (incenters can also be used; the only condition is that has always the same barycentric coordinates with respect to its corresponding triangle).
-continuity
104
105
10 Triangle meshes
Triangle meshes are used for modelling geometric objects. The representation of triangles is supported by the graphics hardware and there are efcient algorithms. Triangle meshes consist of: a set of vertices a set of edges a set of triangles
with
10.1
Basics
Notations: For vertices the edges and triangles with For edges all triangles with
resp.
are called adjacent.
are called adjacent.
(1-)neighbourhood of a vertice: all adjacent triangles and the edges and vertices they consist of (see Figure 10.1) 2-neighbourhood of a vertice: neighbourhood of the neighbourhood (see Figure 10.1) n-neighbourhood: neighbourhood of the (
)-neighbourhood
valence of a vertice : number of adjacent edges resp. number of neighbour vertices
Figure 10.1: 1-neighbourhood and 2-neighbourhood of a vertice
We only consider manifold triangulation, i. e. triangle meshes with the following properties:
10 Triangle meshes
(1) The intersection of two different triangles is a common edge, a common vertice or empty. (see Figure 10.2)
106
Figure 10.2: not possible in manifold triangulations
107 (2) Edges have
10.1 Basics
one adjacent triangle
border edge oder two adjacent triangles inner edge.
(3) If the adjacent triangles of a vertice build a open fan (see Figure 10.3), the vertice is a vertice at the border: adjacent triangles closed fan (see Figure 10.3), the vertice is a inner vertice: adjacent triangles
Figure 10.3: open and closed fan
(4) orientable surfaces, i.e. the normals can be oriented constantly no Moebius Strips and Klein Bottles (see Figure 10.4)
Figure 10.4: Moebius Strip and Klein Bottle
(5) The triangle is connected
the border edges build a close polygon.

Example: Tetrahedron
10 Triangle meshes
108
Figure 10.5: Tetrahedron
Theorem 10.1 (EULER formula for general polyhedrons) If is the number of vertices, the number of edges, the number of faces, the genus of the surface (number of holes (see Figure 10.6)) and the number of border polygons die Anzahl der Randpolygone, we have:
Figure 10.6: Two-torus: surface of genus 2
Example: sphere: tetrahedron: torus:

109
10.2 Computer representation of triangle meshes
For the estimation of the required memory of a triangle mesh the following theorem is important. Theorem 10.2 For large triangle meshes holds: Proof: i) Each triangle has 3 edges, each inner edge belongs to 2 triangles. ii)

(because the genus and the number of border polygons is vanishing in large triangle meshes)
10.2
Computer representation of triangle meshes
The information of triangle meshes consists of geometry: where are the vertices, edges and triangles? topology: how are the vertices connected to edges and triangles? This information should be suitably stored for the application and as efcient as possible.
Remark: In the following holds:
Below there are presented three possible representations of triangle meshes in the computer: (1) explicit representation For all triangles the vertices oats
are stored.
Remark: Maximal redundant, because the geometry is multiply stored.
(2) "shared vertex" or "indexed face set" list of coordinates of all vertices
list of all triangles by referencing the vertex list, i. e. for integers

Remark:
oats
we store
Efcient because geometry and topology is separated. Therefore it is suitable for archiving the data. Edges are only stored implicitly. Explicit access is difcult.
10 Triangle meshes
Frequent operations are expensive (e. g. access to the neighbour of a vertex). Vertices of the triangles are normally stored counter clockwise.
110
(3) extended shared vertex list of vertices with reference to one adjacent triangle. (convention for border vertices: right border triangle) oats integers list of triangles with reference to adjacent triangles (at the border: integers
Remark: The extended shared vertex structure is suitable especially for the processing triangle meshes. E. g. neighbour vertices can be found easily because of the stored lists.
Example: tetrahedron (1) explicit data structure
0 1 2 3 (2) shared vertex structure
1.0, 1.0, 1.0 1.0, 1.0, 1.0 1.0, 1.0, 1.0 1.0, -1.0, -1.0
-1.0, 1.0, -1.0 -1.0, -1.0, 1.0 1.0, -1.0, -1.0 -1.0, -1.0, 1.0
-1.0, -1.0, 1.0 1.0, -1.0, -1.0 -1.0, 1.0, -1.0 -1.0, 1.0, -1.0
0 1 2 3
1.0 -1.0 -1.0 1.0
1.0 1.0 -1.0 -1.0
1.0 -1.0 1.0 -1.0
0 1 2 3 0 0 0 3 1 2 3 2 2 3 1 1
111 (3) extended shared vertex structure
10.3 Parametrization of triangle meshes
0 1 2 3
1.0 -1.0 -1.0 1.0
1.0 1.0 -1.0 -1.0
1.0 -1.0 1.0 -1.0
0 0 0 1
0 1 2 3 0 0 0 3 1 2 3 2 2 3 1 1

3 3 3 0 1 2 0 2 2 0 1 1
10.3
Parametrization of triangle meshes
Given: data points Find: parameter values for which the angles and splitting ratios of the triangle mesh are preserved. Notations:
border points:
(indices of) neighbour of : 10.3.1 The spring modell
In order to parameterize the inner points we use the spring modell from physics. Interpretation: Springs are attached to the inner points and xed along the edges at the neighbour points. The springs are idealised and have the minimum size 0. The mesh is getting at and the springs are as minimal tensed as possible. Parameterization of data points by spring modells requires the minimization of a quadratic functional. That corresponds to solving a linear system of equations. The spring equation is:

Minimization:

sonst
10 Triangle meshes
112

with
Remark:

sonst
. . .
. .
is invertible because it is diagonal dominant.
Proceeding of parameterization: (1) parameterization of the border vertices (2) parameterization of the inner vertices 10.3.2 Parametrization of the border vertices
First the border vertices of the triangle mesh have to be parameterized by one of the following possibilities. i) method for curves over a circle ii) specify corners and parameterize over a polygon with corners iii) projection of the border vertices onto the least squares plane
Remark: It is possible that method iii) leads to convolutions of the border polygon. The projected border polygon is not convex necessariliy.
10.3.3
Parametrization of the inner vertices
Given: vertices Find: parameter values
, which fulll the following:
(1) identity property: parameterization is identical with the triangle mesh, that has been pressed into the plane. (2) regularity resp. no convolutions
113 Choice of the a) :
10.3 Parametrization of triangle meshes
are called
uniform parametrization zentripetal parametrization chordal parametrization
Disadvantage: If the given triangle mesh lies in a plane, the mesh is getting deformed. The parameterization does not fulll (1), but (2). b) discret harmonic parameterization

Disadvantage: parameterized triangle mesh might be convoluted. The parameterization fullls (1), but not (2). c) shape preserving parameterization: fullls (1) and (2) i) Flattening of the 1-neighbourhood of a inner point v by exponential map (see Figure 10.3.3 (a) and (b)):

and
and
with
ii) determine by averaging the barycentric coordinates (see Section 9.1) For each we nd one , so that , with i. e. all other (see Figure 10.3.3 (c)) Finally: Dieses ist das zwischen den Punkten und . Es ist klar, dass i)
denn f"ur alle gilt
10 Triangle meshes
ii)

114
ist damit Konvexkombination der

Remark: Diese sind i. d. R. nicht symmetrisch, d. h.
(a)
(b)
(c)
Figure 10.7: A local part of a triangular mesh (a) is mapped into the plane by expontial mapping (b) and the local weights are determined (c).
115
11 Subdivision Surfaces
Motivation: Comparison TP-Surfaces - Triangle Meshes TP-Surfaces memory requirements smooth surface display on screen free forms complex forms + + Triangle Meshes + + +
Subdivision surfaces are a compromise between TP-surfaces and triangle meshes. In the following, we learn that Subdivision Surfaces offer a good possibility for surface modelling (e. g. in movies). The modelling process consists of two steps: First build a coarse triangle mesh of the object, then apply several subdivision steps. We receive a smooth surface, which is - or -continuous in the limit and interpolates or approximates the original mesh. Basic idea: 1-to-4 split (see Figure 11.1)
Figure 11.1: 1-to-4 split
Analogously: subdivision of curves useful terms: "old" points: even vertices "new" points: odd vertices There are two kinds of subdivision schemes: (1) interpolative subdivision: Even vertices are kept, odd vertices are added. (2) approximative subdivision: Also the position of the even vertices changes.
Remark: The more subdivision steps, the smaller the triangles (and the smoother the surface). In the limit (after subdivision steps): smooth surface
116
Advantages and disadvantages of subdivision surfaces: + arbitrary topology + scalability + compromise between triangle meshes and TP-surfaces + numerical stability + code simplicity + adaptivity no simple method with global
-continuity in the limit
artifacts at vertices with high valence Objectives for the development of subdivision schemes: efcient calculation of even and odd vertices simplicity and locality of this rules afne invariance
convex combinations
continuity of the limit surface
117
11.1 Approximative schemes: Loop-Subdivision
11.1
Approximative schemes: Loop-Subdivision
The Loop-Subdivision scheme is approximative. It is the most simple subdivision scheme. Even and odd vertices in regular triangle meshes are calculated with the following masks: odd vertices:
even vertices:

odd
even

at the boarder:

at the boarder:

odd
even
Figure 11.2: mask for the calculation of odd and even vertices
For irregular even vertices the Loop-Subdivision uses the scheme in Figure 11.3.
with and
valence of
Figure 11.3: masks for the calculation of irregular even vertices
At regular vertices the limit surface of the Loop-Subdivision scheme is -continuous. At irregular vertices it is -continuous. You can see a triangle mesh and three Loop-Subdivision steps in the Figures 11.4 und 11.5.
118
Figure 11.4: Loop-Subdivision: top: coarse triangle mesh, bottom: after 1 subdivision step
119
11.1 Approximative schemes: Loop-Subdivision
Figure 11.5: Loop-Subdivision: top: after 2 steps, bottom: after 3 steps
120
11.2
Interpolatory schemes: Modied Buttery-Subdivision
Buttery-Subdivsion is a interpolatory scheme. It reaches -continuity of the limit surface. It only needs one rule for the calculation of the odd vertices, because the even vertices are interpolated. Inner points:
(1) regular neighbour vertices
Figure 11.6: Buttery: Mask for the calculation of the odd vertices
(2) one irregular neighbour vertice
k = 3:

k 5:
k = 4:
Figure 11.7: one irregular neighbour vertice with valence 3, 4 and 5
(3) two irregular neighbour vertices Use mask from (2) for both neighbours and average the results
Remark: Case 3 is only possible in the rst subdivision step.
121
11.3 Limit points for approximative subdivision
Figure 11.8: 4-point-scheme for boarder vertices
Boarder vertices: At the border the 4-point-scheme for curves is applied (Figure 11.8). You can see a triangle mesh and three Buttery-Subdivision steps in the Figures 11.9 und 11.10.
11.3
Limit points for approximative subdivision
In this section subdivision is written as a linear process. By analysing the so-called subdivision matrix the position of the vertices on the limit surface and propositions about the continuity of the limit surface can be deduced. 11.3.1 Limit points for the subdivision of cubic splines
The formulas for subdividing cubic splines from section 4.5.5 example 4 for open knot vectors are applied. These formulas for the calculation of the shifted vertices can be displayed in matrix form as follows:
By repeated insertion we get:
We have to know the matrix for the calculation of the limit points. Therefore we have to diagonalize the subdivision matrix S:
122
Figure 11.9: Buttery-Subdivision: top: coarse triangle mesh, bottom: after 1 subdivision step
123
Figure 11.10: Buttery-Subdivision: top: after 2 steps, bottom: after 3 steps
R: matrix consisting of eigenvectors D: eigenvalues of the diagonal matrix
124
Remark: The subdivision matrix of good subdivision schemes for curves has the eigenvalues
11.3.2
Limit points for the subdivision of regular triangle meshes
Analogously to curves:
Subdivision matrix:
. . . . . .
. . . . . .
. . .

..
..
..
125
In the same way as for curves we diagonalize the subdivision matrix. The eigenvalues of the matrix are . In the limit we receive the following subdivision matrix and the limit point of :
. . . . . .
. . .
. . .
. . .
. . .
. . .
Remark: Eigenvalue criterias for the determination of the continuity of the limit surface for Eigenvalues of the subdivision matrix

-continuous

-continuous -continuous
The analysis of the Loop scheme for irregular vertices and the Buttery scheme for all vertices only yields -continuity.

Geometric Modelling

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Geometric Modelling

Uploaded by

Copyright:

Available Formats

Geometric Modelling

Lecture in Winter Term by G. Greiner Transscript by J. Kaminski Revision by S. Bergler

1 General Remarks on Curves and Surfaces

Figure 1.1: Curves in the plane and in space

Figure 1.2: A torus and a free form surface

There are three ways to describe curves and surfaces mathematically.

1 General Remarks on Curves and Surfaces

example sphere: implicit

Remark: reasonable means here: differentiable and

: implicit function theorem. If the following statements are true, : solved to

set of all polynomials

, then is the degree of the polynomial

are coefcients of the polynomial

Set of all polynomials of degree

The set of all -dimensional polynomial curves of degree is denoted by . Example:

Complexity: with multiplications and additions. Pseudo code: for

Note that and are linear spaces (vector spaces).

2D curve 3D curve D curve

. Other possible basis for are

Denition 2.2 Given parameter values

Lagrange1 polynomial of degree .

Theorem 2.1 The Lagrange polynomials

Joseph Louis de Lagrange (17361813)

2.2 Different Basis of

Figure 2.1: Quadratic Lagrange polynomials with respect to paramter values

Denition 2.3 The polynomial

is called the th Bernstein2 polynomial of degree . Additional agreement: , if or .

(3) Pascal triangle: recursion:

Theorem 2.2 Properties of Bernstein polynomials: (1) (2) (3)

Serge N. Bernstein, 1912

(symmetry) has a maximum in at .

1 0.8 0.6 0.4 0.2

Figure 2.2: Bernstein polynomials for

Denition 2.4 (General Bernstein Polynomial) : Given the parametric interval

Bernstein polynomial for the interval

Evaluating formula 2.3 at yields:

Theorem 2.3 The Bernstein polynomials

2.2 Different Basis of

Supplement: Afne Maps

is called multiafne, if for any xed arguments the map

is afne. (Diction: Example:

is multiafne. is multiafne. is not multiafne.

9 (2) is symmetric in each variable, i. e.

2.3 Polar Forms

is the polar form (blossom) of

General formula for polynomials of degree in monomial representation:

Consider a degree 2 polynomial curve: ing way:

. This can be decomposed in the follow-

Figure 2.3: Polar Form of a Curve of degree

Denition 3.1 (Bzier Curve) A polynomial curve

Proof: Write as an afne combination of and :

Pierre Bzier, 1966

Figure 3.1: A Bzier curve

Theorem 3.2 (Derivative of a Bzier Curve) a Bzier curve over Let

Shape Properties for Bzier Curves

Remark: A set of points is called convex, if for any two points

Figure 3.2: Convex hull

(2) End-point interpolation: and the last control point.

, i. e. the curve runs through the rst

an afne map, then

Figure 3.3: Variation diminishing property

Proof: (1) convex hull: For

(Theorem 3.1, diagonal property of the polar form) (3)

The same for (4) Let