159.

235 Graphics 1
159.235 Graphics & Graphical
Programming
Lecture 28 - Curves & Surfaces II

159.235 Graphics 2
Uniform Non-rational B-Splines
For each i > 4 , there is a knot between Q
i-1
and Q
i
at t = t
i
.
Initial points at t
3
and t
m+1
are also knots. The following
illustrates an example with control points set P
0
P
9
:
Knot.
Control point.
m=9 (10 control points)
m-1 knots
m-2 knot intervals.
159.235 Graphics 3
Uniform Non-rational B-Splines
First segment Q
3
is defined by point P
0
through P
3
over the
range t
3
= 0 to t
4
= 1. So m at least 3 for cubic spline.
Knot.
Control point.
P1
P2
P3
P0
Q
3
m=9 (10 control points)
m-1 knots
m-2 knot intervals.
159.235 Graphics 4
Uniform Non-rational B-Splines
Second segment Q
4
is defined by point P
1
through P
4
over
the range t
4
= 1 to t
5
= 2.


Knot.
Control point.
Q
4
P1
P3
P4
P2
m=9 (10 control points)
m-1 knots
m-2 knot intervals.
159.235 Graphics 5
Uniform Non-Rational B-Splines
The unweighted spline set will look as follows, 10 control
points, 10 splines, but only 8 knots and 7 knot intervals.
You can see why t
3
to t
4
is the first interval with a curve
since it is the first with all four B-Spline functions.
t
9
to t
10
is the last interval, t
3
to t
m+1
are the knots.
4 3
t
8 6 m
m+1
0
159.235 Graphics 6
Cubic B-Splines
We have talked about a bi-infinite set of splines.
What does this look like?



The unweighted B-Splines are shown for clarity.
How does a weighted set affect the shape of the
curve?
t 4 8 12
159.235 Graphics 7
Generating a Curve
X(t)
t
t
Opposite we see an
example of a shape to be
generated.


Here we see the curve
again with the weighted
B-Splines which
generated the required
shape.
159.235 Graphics 8
B-Spline Example
A closed curve is rather like a bi-infinite set
a periodic set of N knots in which:
for all i, t
i+N
= t
i
and t
i
= i for i = 0..N-1
Consider a closed curve with 5 control
points with 5 knots. N=5, ie. x
1
x
4,
etc.

Control points are: x
0
= (2,0); x
1
= (1,1);
x
2
= (-1,1); x
3
= (-1,-1) and x
4
= (1,-1).
159.235 Graphics 9
B-Spline Example
Draw the interval: t = [4..5] , the others are handled the same way.
| |
(
(
(
(

(
(
(
(

=
1 1
1 1
1 1
1 1
0 1 4 1
0 3 0 3
0 3 6 3
1 3 3 1
6
1
1 ) 4 ( ) 4 ( ) 4 ( ) (
2 3
t t t t X
) 0 , 12 11 (
) 3 2 , 3 2 ( ) 5 ( ) 3 2 , 3 2 ( ) 4 (
) 4 6 6 4 , 6 6 4 (
6
1
) ) 4 ( 4 ) 4 ( 6 ) 4 ( 6 4 , ) 4 ( 6 ) 4 ( 6 4 (
6
1
3 2 2
3 2 2
=
= =
+ + =
+ + =
midway and
X and X Hence
t t t t t
or
t t t t t
x
0

x
1

x
2

x
3
x
4

x
4

x
1

159.235 Graphics 10
How Smooth is a B-Spine?
Smoothness increases with order k in B
i,k
Quadratic, k = 3, gives up to C
1
order continuity.
Cubic, k = 4 gives up to C
2
order continuity.
However, we can lower continuity order too with Multiple Knots, ie.
t
i
= t
i+1
= t
i+2
= Knots are coincident and so now we have non-
uniform knot intervals.
A knot with multiplicity m is continuous to the (k-1-m)th
derivative.
A knot with multiplicity k has no continuity at all, ie. the curve is
broken at that knot.


159.235 Graphics 11
Non-Uniform Non-Rational B-Splines
Note: as m increases by 1, an extra B-Spline
function is attached to that knot: q
i
B
i,k
B-Splines have special forms at multiple knots.
Obtained from a recursive formula defining how
B-Splines are built, by setting m successive knots
to be equal: t
k+1
= t
k+2
= = t
k+m
No point in having m>k
Thus for endpoints to be on curve, they must have
a multiplicity of k, multiplicity of 4 in cubic case.
159.235 Graphics 12
B-Splines at Multiple Knots
The multiple-knot B-Splines are as follows:
t
B
3,4
(t)
B
0,4
(t)
B
1,4
(t)
B
2,4
(t)
i=0,1,2,3 4 5 6 7
B
0,4
is totally discontinuous at t = t
0
, B
1,4
is position continuous,
B
2,4
is gradient continuous, B
3,4
is curvature continuous.

159.235 Graphics 13
B-Spline Continuity Example
First knot shown with 4
control points, and their
convex hull.
P1
P2
P0
P3
159.235 Graphics 14
B-Spline Continuity Example
First two curve segments
shown with their respective
convex hulls.

Centre Knot must lie in the
intersection of the 2 convex
hulls.
P1
P2
P0
P4
P3
159.235 Graphics 15
Repeated Control Point
First two curve segments
shown with their respective
convex hulls.

Knot is forced to lie on the
line that joins the 2 convex
hulls.

Curve is only C
1
continuous
P1=P2
P0
P3
P4
159.235 Graphics 16
Triple Control Point
First two curve segments
shown with their respective
convex hulls.

Both convex hulls collapse
to straight lines all the
curve must lie on these lines.

Curve is only C
0
continuous

(Curiously it is actually C
2

continuous because tangent
vector magnitude falls to
zero at join. )
P1=P2=P3
P0
P4
159.235 Graphics 17
Non-Uniform Non-Rational B-Splines
Parametric interval between knots does not have to be
equal.
Blending functions no longer the same for each interval.
Multiple knots may have to be spaced apart.
Advantages
Continuity at selected control points can be reduced to
C
1
or lower allows us to interpolate a control point
without side-effects.
Can interpolate start and end points.
Easy to add extra knots and control points.
159.235 Graphics 18
Summary of B-Splines
Functions that can interpolate a series of control points with C
2

continuity and local control.
Dont pass through their control points, although can be forced.
note that if an order k B-Spline has k points locally colinear then a
straight-line section will result midway in the set, and will touch
(tangentially) the convex hull for (k 1) colinear control points.
Uniform
Knots are equally spaced in t.
Non-Uniform
Knots are unequally spaced
Allows addition of extra control points anywhere in the set.
159.235 Graphics 19
B-Spline Summary Continued
For interactive curve modelling
B-Splines are very good.
To interpolate a number of positions
Catmull-Rom splines are best.
To interpolate with tangent control
Hemite or Bzier forms are useful and most often used.
To draw a spline.
Brute force method : Evaluate matrix to get parametric
expressions for coordinates. Then for small increments of
t, join with line segments.
Better : use method of forward differences to express
spline in At form.
159.235 Graphics 20
Surfaces a Simple Extension
Easy to generalise from cubic curves to bicubic
surfaces.
Surfaces defined by parametric equations of two
variables, s and t.
ie. a surface is approximated by a series of
crossing parametric cubic curves
Result is a polygon mesh and decreasing step size
in s and t will give a mesh of small near-planar
quadrilateral patches and more accuracy.
1 0 1 0 s s s s t and s
159.235 Graphics 21
Example Bzier surface
159.235 Graphics 22
Control of Surface Shape
Control is now a 2D array of control points.
The two parameter surface function, forming the
tensor product with the blending functions is:


Use appropriate blending functions for Bzier and
B-Spline surface functions.
Convex Hull property is preserved since bicubic is
still a weighted sum (1).
) , ( ) , (
) ( ) ( ) , (
t s Z and t s Y f or similarly
q t f s f t s X
ij j
ij
i
=
159.235 Graphics 23
Bzier Example
Matrix formulation is as follows:
coords z of array is q
t M q M s t s z
coords y of array is q
t M q M s t s y
coords x of array is q
t M q M s t s x
z
T
B z B
T
y
T
B y B
T
x
T
B x B
T
4 4
. . . . ) , (
4 4
. . . . ) , (
4 4
. . . . ) , (

=

=

=
Substitute suitable values for
s and t (20 in the above ex.)
159.235 Graphics 24
B-Spline Surfaces
Break surface into 4-sided patches choosing suitable
values for s and t.
Points on any external edges must be multiple knots of
multiplicity k.
Lot more work than Bzier.
There are other types of spline systems and NURBS
modelling packages are available to make the work much
easier.
Use polygon packages for display, hidden-surface removal
and rendering. (Bzier too)
159.235 Graphics 25
Continuity of Bicubic Patches
Hermite and Bzier patches
C
0
continuity by sharing 4 control points
between patches.
C
1
continuity when both sets of control points
either side of the edge are collinear with the
edge.
B-Spline patch.
C
2
continuity between patches.

159.235 Graphics 26
Displaying Bicubic Patches
Can calculate surface normals to bicubic surfaces
by vector cross product of the 2 tangent vectors.
Normal is expensive to compute
Formulation of normal is a biquintic (two-variable,fifth-
degree) polynomial.
Display.
Can use brute-force method very expensive !
Forward differencing method very attractive.
159.235 Graphics 27
Curves & Surfaces II - Summary
Bezier Curves
Bicubic patches

Acknowledgments - thanks to Eric McKenzie, Edinburgh, from whose
Graphics Course some of these slides were adapted.