Title: Bézier Curves and Surfaces

Name: Michael S. Floater1

Affil./Addr.: Department of Mathematics

University of Oslo

Moltke Moes vei 35, 0851 Oslo, Norway

Phone: (+47) 22854036

Fax: (+47) 22854349

E-mail: {michaelf}@math.uio.no

Bézier Curves and Surfaces

Definition

Computer-aided geometric design (CAGD) is the design of geometrical shapes using

computer technology, and is used extensively in many applications, such as the au-

tomotive, shipbuilding, and aerospace industries, architectural design, and computer

animation. A popular way of modelling geometry in CAGD is to represent the outer

surface, or curve, of the object as a patchwork of parametric polynomial pieces. Bézier

curves and surfaces are a representation of such polynomial pieces that makes their in-

teractive design easier and more intuitive than with other representations. They were

developed in the 1960’s and 1970’s by Paul de Casteljau and Pierre Bézier, for use in

the automotive industry.

Curves

A Bézier curve, of degree n, on some interval [a, b], is a parametric polynomial p :

[a, b] → Rd given by the formula

 2
n
X
p(t) = ci Bin (u), t ∈ [a, b],
i=0

where: u is the local variable, u = (t − a)/(b − a); the points ci ∈ Rd are the control

points of p; and Bin is the Bernstein (basis) polynomial

 
n i
Bin (u) = u (1 − u)n−i , u ∈ [0, 1].
i

The Euclidean space will often be R2 or R3 . The polygon formed by connecting the

sequence of control points c0 , c1 , . . . , cn is known as the control polygon of p. The shape

of p tends to mimic the shape of the polygon, making it a popular choice for designing

geometry in an interactive graphical environment. Figure 1 shows a cubic Bézier curve

Fig. 1. A cubic Bézier curve

and its control polygon. The cubic Bernstein polynomials are

B03 (u) = (1 − u)3 , B13 (u) = 3u(1 − u)2 , B23 (u) = 3u2 (1 − u), B33 (u) = u3 ,

shown in Figure 2.

Various properties of Bézier curves follow from properties of the Bernstein poly-

nomials. One is the endpoint property: p(a) = c0 and p(b) = cn . Another is that since

the Bin are non-negative and sum to one, every point p(t) is a convex combination of

the control points and p lies in the convex hull of the control points. Similarly, p lies

in the bounding box

[x1 , y1 ] × [x2 , y2 ] × · · · × [xd , yd ],

where, if the point ci has coordinates c1i , . . . , cdi ,

 3

0.8

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
Fig. 2. The cubic Bernstein polynomials

xk = min cki and yk = max cki , k = 1, . . . , d.

0≤i≤n 0≤i≤n

Bounding boxes are useful for visualization, and for detecting intersections between

pairs of objects and self-intersections.

Due to the recursion formula,

n−1
Bin (u) = uBi−1 (u) + (1 − u)Bin−1 (u),

one can evaluate (compute) p(t) for some t ∈ [a, b] using de Casteljau’s algorithm.

After the initialization c0i = ci , i = 0, 1, . . . , n, we compute

cri = (1 − u)cir−1 + uci+1

r−1
,

for r = 1, . . . , n, and i = 0, 1, . . . , n − r, the last point being the point on the curve:

p(t) = cn0 . This can be viewed as the following triangular scheme, here arranged row-

wise, with each row being computed from the row above:

c00 c01 c02 ··· c0n

c10 c11 ··· c1n−1

... ...

c0n−1 c1n−1

cn0
 4

Derivatives of p can be computed by expressing them as Bézier curves of lower degree:

n−1
dp n X
p′ (t) = = ∆ci Bin−1 (u),
dt (b − a) i=0

where ∆ is the forward difference, ∆ci = ci+1 − ci , and more generally,

n−r
(r) dr p n(n − 1) · · · (n − r + 1) X r
p (t) = r = r
∆ ci Bin−r (u).
dt (b − a) i=0

Complex curves are often modelled by joining several Bézier curves together. If q :

[b, c] → Rd is another Bézier curve,

n
X t−b
q(t) = di Bin (v), t ∈ [b, c], v= ,
i=0
c−b

then p and q join with C k continuity at t = b, i.e., q(r) (b) = p(r) (b) for all r = 0, 1, . . . , k,

if and only if
∆r d0 ∆r cn−r
= , r = 0, 1, . . . , k.
(c − b)r (b − a)r

Tensor-product surfaces

A tensor-product Bézier surface in Rd is a parametric polynomial p : D → Rd of degree

m × n, given by the formula

m X
X n
p(s, t) = ci,j Bim (u)Bjn (v), (s, t) ∈ D,
i=0 j=0

where D is a rectangle, D = [a1 , b1 ] × [a2 , b2 ], and

s − a1 t − a2
u= , v= .
b 1 − a1 b 2 − a2

The Euclidean space is usually R3 . The control net of p is the network of control points

ci,j ∈ Rd and all line segments of the form [ci,j , ci+1,j ] and [ci,j , ci,j+1 ].

On each of the four sides of D, the surface p is a Bézier curve whose control

polygon is one of the four boundaries of the control net of p. At the four corners of

D, the surface p equals one of the corners of the control net. Like Bézier curves, these
 5

surfaces have the convex hull and bounding box properties. The point p(s, t) can be

evaluated by applying de Casteljau’s algorithm to the rows of points in the control net,

in each of the two directions, using m steps with respect to u and n steps with respect

to v. These m + n steps can be applied in any order.

Triangular surfaces

A triangular Bézier surface, of degree n, is a polynomial p : T → Rd , in the form

X
p(t) = ci Bin (u), t ∈ T,
|i|=n

where: T ⊂ R2 is a triangle, with vertices a1 , a2 , a3 ∈ R2 ; i = (i, j, k); |i| = i + j + k;

u = (u, v, w), and the values u, v, w ≥ 0 are the barycentric coordinates of the point t

with respect to T , i.e., the three values such that

u + v + w = 1,

ua1 + va2 + wa3 = t,

and Bin is the Bernstein polynomial

n! i j k
Bin (u) = uv w .
i!j!k!

For example, with n = 3 there are 10 such polynomials,

3
B003
3 3
B102 B012
3 3 3
B201 B111 B021
3 3 3 3
B300 B210 B120 B030

given by the formulas

 6

w3

3uw2 3vw2

3u2 w 6uvw 3v 2 w

u3 3u2 v 3uv 2 v3
The points ci ∈ Rd are the control points of p, which, together with all line segments

that connect neighbouring points, form the control net of p. Two control points are

neighbours if they have one index in common and the other two indices differ by

one. A point that is not on the boundary has six neighbours. On each of the three

sides of T , the surface p is a Bézier curve whose control polygon is the corresponding

boundary polygon of the control net of p. At the corners of T , the surface equals

one of the corner control points. Triangular Bézier surfaces have the convex hull and

bounding box properties. There is a de Casteljau algorithm for evaluating p and there

are formulas for derivatives and conditions for joining pairs of such triangular surfaces

together with a certain order of continuity.

Figure 3 shows a biquadratic surface, where m = n = 2, and a quadratic surface,

where n = 2, with their control nets.

Fig. 3. Biquadratic and quadratic surfaces

 7

References

1. Farin G (1990), Curves and Surfaces for Computer Aided Geometric Design (2nd edition), Aca-

demic Press, San Diego.

2. Prautzsch H, Boehm W, and Paluszny M (2002), Bézier and B-spline techniques, Springer, Berlin.