Computer Vision and Image Understanding 81, 1–25 (2001)

doi:10.1006/cviu.2000.0881, available online at http://www.idealibrary.com on

Conversions between Parametric and Implicit

Forms Using Polar/Spherical
Coordinate Representations
Cem Ünsalan
Department of Electrical and Electronics Engineering, Boğaziçi University, Bebek, Istanbul 80815, Turkey
E-mail: unsalan@boun.edu.tr

and

Aytül Erçi̇l
Department of Industrial Engineering, Boğaziçi University, Bebek, Istanbul 80815, Turkey
E-mail: ercil@boun.edu.tr

Received August 14, 1998; accepted September 12, 2000

Since parametric and implicit forms have complementary advantages with respect
to certain geometric operations, it can be useful to convert from one form to the other.
In this paper, a new method for converting between parametric and implicit forms
based on polar/spherical coordinate representations is introduced. °c 2001 Academic Press
Key Words: implicitization; parameterization; polar coordinates; spherical co-
ordinates.

1. INTRODUCTION

The development of computer aided graphics design has seen two distinct approaches to
representing surfaces in 3D space:
1. Parametric methods with a representation of the form (x(u, v), y(u, v), z(u, v)), which
maps a 2D domain containing (u, v) to 3D space.
2. Implicit methods that define a surface as a set of points {(x, y, z) such that
F(x, y, z) = 0}.
The use of parametric surfaces for the general representation and design of free-form
surfaces has been quite successful and remains dominant in computer graphics and geomet-
ric modeling. The implicit approach is philosophically more closely related to the concepts
of constructive solid geometry and solid modeling and is receiving increased attention.
1
1077-3142/01 $35.00
Copyright ° c 2001 by Academic Press
All rights of reproduction in any form reserved.
2 ÜNSALAN AND ERÇİL

Implicit surface functions naturally describe the interior of an object, whereas a parametric
description of an object usually consists of piecewise surface patches. Both approaches have
long lists of pros and cons [2]. Although the parametric formulation is useful for tracing,
rendering, and surface fitting, many operations such as surface intersection require one of
the surfaces to be represented implicitly. Moreover, the implicit representation can be used
to test whether a point lies on the boundary and to represent an object as a semialgebraic set,
and implicit forms are finding wider applications in computer vision, mainly in the area of
object recognition and automated tolerance inspection [3, 4, 13–16]. Algebraic invariants of
implicit polynomials have been shown to be very effective in Euclidean and affine invariant
object recognition [16].
Since parametric and implicit forms have complementary advantages with respect to cer-
tain geometric operations, it can be useful to convert from one form to the other. Conversion
between implicit and parametric forms opens new possibilities of combining the existing
vast databases of CAD models using parametric representations with the advantages of
implicit polynomials for invariant object recognition.
Conversion from parametric to implicit form is known as implicitization and every rational
surface and curve can be represented implicitly as the zero set of an irreducible1 homo-
geneous polynomial f (x, y, z, w) = 0 for surfaces, and f (x, y) = 0 for 2D curves [12].
Sederberg et al. [12] apply resultants to eliminate parameters from polynomials; Hoffman
[5] details the use of the Gröbner bases for the same purpose; and Hoffman [6] describes
the Wu–Ritt method. The conversion from implicit to parametric form is known as parame-
terization. Parameterization is not always possible, however; for example, implicit surfaces
that are defined by certain polynomials of fourth and higher degree cannot be parameterized
by rational functions [9]. Conversion is always possible for nondegenerate quadrics and for
cubics that have a singular point.
In this paper, a new approach to conversion between parametric and implicit forms
based on polar coordinate representation of the coordinate system will be outlined. The
main contribution is the implicit/parametric conversion since no techniques are available
in the literature for this type of conversion. Section 2 outlines the technique for converting
from parametric form to implicit form and Section 3 outlines the technique for the reverse
conversion. In Section 4, the conversion techniques developed are applied to automated
tolerance inspection of a 3D object.

2. CONVERSION FROM PARAMETRIC FORM TO IMPLICIT FORM

There are three known techniques for implicitization of parametric forms. All of these
techniques reduce the problem of implicitizing rational surfaces to eliminating two vari-
ables from three parametric equations. In general, it is believed that techniques based on
elimination theory can result in extraneous factors along with the implicit representation,
and their separation can be a difficult task. The technique developed in this section is valid
only for star-shaped objects.

THEOREM 1 (Conversion in 2D). Let α be a parameterized curve in <2 s.t. α : (a, b) →

< with α(t) = (h(t), g(t)), where (a, b) is an open interval in <. The implicit representation
2

1
If a change of variables cannot reduce the degree of the polynomial expression, then it is assumed to be
irreducible.
 PARAMETRIC AND IMPLICIT FORM CONVERSION 3

of this curve is given as

µ µ ¶¶ µ µ ¶¶
−1 y −1 y
h 2
f +g f
2
= x 2 + y2, (1)
x x

where f (t) is an invertible function of the form f (t) = g(t)

h(t)
.

Proof. Let (r, θ) be a polar representation of a point p = (x, y) ∈ <2 on the curve α for
some t. The value, xy , at any point (x, y) on the curve and parametric representation of the
g(t)
value, h(t) , at this point is the same. Equating these two values and defining a new function
of the form gives f (t) ¡ = h(t)
¢
g(t)
= xy . As the curve is star-shaped, f (t) is an invertible function.
We obtain t = f −1 x . Using the standard change of variables formulas from rectangular
y

to polar coordinates,

x 2 + y2 = r 2 (2)
x = r cos θ (3)
y = r sin θ, (4)

we can rewrite Eq. (2) as

µ µ ¶¶ µ µ ¶¶
−1 y −1 y
h 2
f +g f
2
= r 2 = x + y2.
x x

EXAMPLE 1. Let h(t) = sin t and g(t) = sin t cos t; then f (t) = cos t and t = cos−1 ( xy ) .
The corresponding implicit form is x 2 + y 2 = sin2 (cos−1 xy ) + sin2 (cos−1 xy ) cos2 (cos−1 xy ).
Simplifying this equation will yield x 4 − x 2 + y 2 = 0. Both parametric and implicit repre-
sentations are plotted in Fig. 1.
The advantage of this technique over existing techniques in the literature is that it is not
iterative and hence is in general faster, although no systematic time comparisons have been
carried out. Another advantage is that there are no spurious parts in the implicit function
obtained.
Theorem 1 gives the conversion of parametric functions to corresponding implicit poly-
nomial form when f (t) = h(t)g(t)
is invertible. If the parametric representation of the object
boundary is carried out using Fourier expansion fitting, which is detailed below, the implicit

FIG. 1. Plot of parametric and implicit representations.

4 ÜNSALAN AND ERÇİL

function/polynomial representation of the object can be obtained more easily. This case is
given in Theorem 2.
DEFINITION 1. Let Cn be an (n + 1)-dimensional vector such that

Cn (2i + 1) = (−1)i an , Cn (2i + 2) = (−1)i bn i = 0, 1, 2, 3 . . . ,

where
µ ¶ µ ¶
1 X 1 X 1 X
Ns Ns Ns
2knπ 2knπ
a0 = r (k) an = r (k) cos bn = r (k) sin .
2Ns k=1 Ns k=1 Ns Ns k=1 Ns

an , bn correspond to Fourier series coefficients of the parametric representation of a 2D

shape F(r (t), t) = 0 t ∈ [0, Ns ]. Ns is the total number of samples of the function r (t) of
the contour of the 2D shape.
Define ⊗ to be an operator such that
à n+1 !
X Xn+1
ai ⊗ Sn = ai Sn (i).
i=1 i=1

Both the implicit polynomial representation and the parametric representation of the object
are based on representing the boundary of the object in polar coordinates and expanding
the radius function r (t) in terms of the angle θ by Fourier series. Ns is the total number of
samples of the radius function r (t), from 0 to 2π radians. Note here that radius function is
equally sampled, so f (t) = t and t = θ .
LEMMA 1. N th degree Fourier series approximations for x(θ ) and y(θ) can be repre-
sented as

x(θ) = r cos(θ ), y(θ) = r sin(θ ),

where
∞
X
r = a0 + an cos nθ + bn sin nθ.
n=1

Proof. Since the radius function r = r (θ) for a star-shaped object boundary is periodic
and it satisfies the requirements for being a function, r can be expanded into a Fourier series
with respect to index θ as
∞
X
r = a0 + an cos nθ + bn sin nθ.
n=1

Using the polar/Cartesian conversion formulas, the representations for x(t) and y(t) are
obtained.
THEOREM 2 (Conversion from Parametric Representations Using Fourier Series Fitting).
An implicit function representation of any star-shaped object represented by N th degree
Fourier series can be obtained as

X
N
Kn
(x 2 + y 2 )1/2 = , (5)
n=0
(x 2 + y 2 )n/2
 PARAMETRIC AND IMPLICIT FORM CONVERSION 5

where

K n = (x + y)n ⊗ Cn and K 0 = a0 . (6)

The corresponding implicit polynomial representation is given as follows:

For N even:
à !2 à !2
X
N X
N
2 N /2 2 (N −m−1)/2 2 (N −s)/2
(x + y ) (x + y )
2 2 2
− K m (x + y )
2
= K s (x + y )
2
.
mod d seven

(7)

For N odd:
à !2 à !2
X
N X
N
2 (N +1)/2 2 (N − m)/2 2 (N −s−1)/2
(x + y )
2
− K m (x + y )
2
= (x + y )
2 2
K s (x + y )
2
.
mod d seven

(8)

Proof. Since radius function r = r (θ ) for a star-shaped object boundary is periodic and
it satisfies the requirements for being a function, r can be expanded into Fourier series with
respect to index θ as

∞
X
r = a0 + an cos nθ + bn sin nθ. (9)
n=1

Using the expansion formulas for cos nθ and sin nθ given in Appendix A, and substi-
tuting Eqs. (2)–(4) into the expanded formula, we obtain the above implicit polynomial
form.
EXAMPLE 2. Two star-shaped objects represented by a parametric representation using
Fourier series expansion are converted to implicit polynomial form using Theorem 2. The
80th degree implicit polynomial representations for Mig and Butterfly shapes obtained from
this conversion are given in Figs. 2 and 3.

THEOREM 3 (Conversion for Space Curves). Let α be a parameterized space curve

in <3 s.t. α : (a, b) → <3 with α(t) = (h(t), g(t), k(t)), where (a, b) is an open interval
in <.

FIG. 2. Implicit polynomial plot of Mig.

6 ÜNSALAN AND ERÇİL

FIG. 3. Implicit polynomial plot of Butterfly.

The two implicit surfaces, whose intersection represents this space curve, are given as
µ µ ¶¶ µ µ ¶¶ µ µ ¶¶
y y z
h2 f 1−1 + g 2 f 1−1 + k 2 f 2−1 p = x 2 + y 2 + z 2 (10)
x x x 2 + y2
µ µ ¶¶ µ µ ¶¶ µ µ ¶¶
−1 y −1 y −1 z
2
h f1 + g f1
2
+ k f3
2
= x 2 + y 2 + z 2 , (11)
x x y

where

f 1 (t) is an invertible function of the form f 1 (t) = g(t)/ h(t)

p
f 2 (t) is an invertible function of the form f 2 (t) = k(t)/ g 2 (t) + h 2 (t)
f 3 (t) is an invertible function of the form f 3 (t) = k(t)/g(t).

Proof. Let (r, θ, ϕ) be a polar representation of a point p = (x, y, z) ∈ <3 on the curve α.
x = h(t), y = g(t), z = k(t) for some t. f 1 (t), f 2 (t), and f 3 (t) are obtained using equal-
ities similar to those given in Theorem 1. These three functions are invertible since the
space curvepis star-shaped. We can find three different versions of t as t = f 1−1 (y/x),
t = f 2−1 (z/ x 2 + y 2 ), and t = f 3−1 (z/y). Using the standard formulas for change of vari-
ables from rectangular to spherical coordinates,

x 2 + y2 + z2 = R2 (12)
x = R cos θ sin ϕ (13)
y = R sin θ sin ϕ (14)
z = R cos ϕ, (15)

we can rewrite Eq. (10) as

µ µ ¶¶ µ µ ¶¶ µ µ ¶¶
y −1 y z
h 2
f 1−1 + g f1
2
+ k f2
2 −1
p = R2 = x 2 + y2 + z2.
x x x 2 + y2

We can rewrite Eq. (11) as

µ µ ¶¶ µ µ ¶¶ µ µ ¶¶
y −1 y −1 z
h 2
f 1−1 + g f1
2
+ k f3
2
= R2 = x 2 + y2 + z2.
x x y
 PARAMETRIC AND IMPLICIT FORM CONVERSION 7

EXAMPLE 3. α(t) = (h(t), g(t), k(t)) is a parametric space curve. h(t) = sin t, g(t) =
cos t, k(t) = cos2 t. Then

cos t y
f 1 (t) = t = arccot
sin t x
z
f 2 (t) = cos2 t t = arccos p
x 2 + y2
z
f 3 (t) = cos t t = arccos .
y

Implicit surface 1 is obtained as x 2 + y 2 + z 2 = 1 + z 2 /(x 2 + y 2 ). In simplified form

x + y 2 = 1.
2

Implicit surface 2 is obtained as x 2 + y 2 + z 2 = 1 + z 4 /y 4 . In simplified form z = ±y 2

or z = ±(1 − x 2 ).
Two implicit surfaces, which represent this space curve, are obtained as

x 2 + y2 = 1
z = 1 − x 2.

A parametric representation of the space curve is plotted in Fig. 4. Two intersecting

implicit surfaces, where their intersection represents the space curve, are plotted in Fig. 5.
It can be seen that the two representations correspond to the same space curve.

THEOREM 4 (Patch Conversion for 3D). Let α be a parameterized patch in <3 s.t.
α : [(0, 2π ), (0, 2π)] → <3 , with α(θ, ϕ) = (x(θ, ϕ), y(θ, ϕ), z(θ, ϕ)). An implicit repre-
sentation of this patch in terms of x, y, z can be found, if θ and ϕ can be obtained explicitly,
from
p p
y(θ, ϕ) y x(θ, ϕ)2 + y(θ, ϕ)2 x 2 + y2 y(θ, ϕ) y
= , = , = .
x(θ, ϕ) x z(θ, ϕ) z z(θ, ϕ) z

An implicit representation for the corresponding patch is given as

x(θ, ϕ)2 + y(θ, ϕ)2 + z(θ, ϕ)2 = x 2 + y 2 + z 2 . (15)

Proof. The proof is the same as that given for previous two theorems. The only difference
in this theorem is that an explicit slope function cannot be defined here.
EXAMPLE 4. Let α(θ, ϕ) = (x(θ, ϕ), y(θ, ϕ), z(θ, ϕ)) be a patch representation of the
shape given in Fig. 6. x(θ, ϕ) = cos θ sin ϕ, y(θ, ϕ) = cos θ, z(θ, ϕ) = cos ϕ. ϕ is obtained
from the equality xy = sin1 ϕ as ϕ = arcsin xy . θ is obtained from the equality
p p
cos2 θ (1 + sin2 ϕ) x 2 + y2
=
cos ϕ z
as
p
y2 − x 2
θ = arccos .
z
8 ÜNSALAN AND ERÇİL

FIG. 4. Parametric representation.

FIG. 5. Implicit representation (intersection of two implicit surfaces).

FIG. 6. Plot of 3D patch given in Example 4.

FIG. 13. Implicit surfaces and parametric form of space curve in Example 11.
 PARAMETRIC AND IMPLICIT FORM CONVERSION 9

FIG. 14. Water pipe and point to be inspected.

Replacing these values in the equality gives

cos2 θ(1 + sin2 ϕ) + cos2 ϕ = x 2 + y 2 + z 2 .

The corresponding implicit form is obtained as

y 2 − x 2 − y 2 z 2 = 0.

3. CONVERSION FROM IMPLICIT FORM TO PARAMETRIC FORM

Conversion of an implicit polynomial to a corresponding parametric form is not always

possible. In this section, we will give some cases for which the implicit-to-parametric
conversion can be carried out. The conversion can be carried out exactly for the cases given
in Lemmas 2–5 and approximately for more general cases.

3.1. Conversion in 2D
LEMMA 2. A polynomial of the form

X
n X
k
m (n−m)
am x y + bm x m y (k−m) = 0 (16)
m=0 m=0

(sum of two homogeneous polynomials) can be converted to parametric form α(θ ) =

(h(θ), g(θ)) as
s P
k
bm cosm θ sin(k−m) θ
h(θ) = n−k
− Pnm=0 cos θ (17)
m=0 am cos θ sin θ
m (n−m)

s P
k
bm cosm θ sin(k−m) θ
g(θ) = n−k
− Pnm=0 sin θ. (18)
m=0 am cos θ sin θ
m (n−m)

Parameter values to obtain the real-valued parametric form specify the parameter range.
Proof. Equation (16) can be represented in polar coordinates as

X
n X
k
rn am cosm θ sin(n−m) θ + r k bm cosm θ sin(k−m) θ = 0.
m=0 m=0
10 ÜNSALAN AND ERÇİL

If we solve the above equation for r , we obtain

Pk
bm cosm θ sin(k−m) θ
r n−k = − Pnm=0 . (19)
m=0 bm cos θ sin θ
m (n−m)

By using the polar-to-rectangular coordinate change-of-variable formulas α(θ ) = (r cos θ ,

r sin θ) is obtained.
EXAMPLE 5. Consider the implicit curve x 4 − x 2 + y 2 = 0:

(r cos θ )4 − (r cos θ )2 + (r sin θ)2 = 0

cos2 θ − sin2 θ
r2 =
cos4 θ
p
cos2 θ − sin2 θ p sin θ
h(θ) = g(θ ) = cos2 θ − sin2 θ θ ∈ [0, 2π].
cos θ cos2 θ
EXAMPLE 6. Superquadratics in 2D can be converted to parametric form by Lemma 2.
A general form of a superquadratic can be given as
"µ ¶ µ ¶2/ε2 #ε2 /ε1
x 2/ε2 y
+ −a = 1, (20)
a1 a2

obtaining r (θ) as
Á·µ ¶2/ε2 µ ¶2/ε2 ¸ε2 /2
ε2 /2 cos θ sin θ
r (θ) = a + .
a1 a2
The parametric form for the general superquadratic is then given as

h(θ) = r (θ ) cos θ, g(θ ) = r (θ ) sin θ.

LEMMA 3. A polynomial of the form

X
2 X
n−l
aml x m y (n−l−m) = 0 (21)
l=0 m=0

(three homogeneous polynomials of consecutive degrees) can be converted to two parametric

forms as

α1 (t) = (h 1 (t), g1 (t)) (22)

α2 (t) = (h 2 (t), g2 (t)) (23)
p
− p − p 2 − 4sq
h 1 (θ) = cos θ (24)
2s
p
− p − p 2 − 4sq
g1 (θ) = sin θ (25)
2s
p
− p + p 2 − 4sq
h 2 (θ) = cos θ (26)
2s
p
− p + p 2 − 4sq
g2 (θ) = sin θ, (27)
2s
 PARAMETRIC AND IMPLICIT FORM CONVERSION 11

where

X
n
s= am0 cosm θ sin(n−m) θ (28)
m=0

X
n−1
p= am1 cosm θ sin(n−1−m) θ (29)
m=0

X
n−2
q= am2 cosm θ sin(n−2−m) θ. (30)
m=0

Parameter values to obtain a real-valued parametric form specify the parameter range.

Proof. Equation (21) can be represented in polar coordinates as

X
2 X
n−l
r n−l aml cosm θ sin(n−l−m) θ = 0
l=0 m=0

X
2 X
n−l
r n−2 r 2−l aml cosm θ sin(n−l−m) θ = 0 (31)
l=0 m=0

sr 2 + pr + q = 0. (32)

Solving this quadratic equation in terms of r and using polar-to-rectangular coordi-

nate conversion formulas will give parametric representations of the polynomial given in
Eqs. (22) and (23). A closed-form solution of this quadratic equation can be found in [1].
EXAMPLE 7. Let x 2 + y 2 − 2x − 2y + 1 = 0, r 2 − 2r (cos θ + sin θ ) + 1 = 0. To find
real-valued roots of this equation, the following constraint has to be satisfied: 4(cos θ +
sin θ)2 − 4 ≥ 0. It is satisfied for θ ∈ [0, π2 ].
The solutions of the above equation are
√ √
r1 = cos θ + sin θ − sin 2θ r2 = cos θ + sin θ + sin 2θ

and the corresponding two parametric curves are obtained as

h 1 (θ) = r1 cos θ g1 (θ ) = r1 sin θ

· ¸
π
h 2 (θ) = r2 cos θ g2 (θ ) = r2 sin θ θ ∈ 0, .
2

The implicit polynomial form of the curve given in this example is plotted in Fig. 7. The
corresponding two parametric forms obtained from Lemma 3 are plotted in Figs. 8 and 9.

LEMMA 4. A polynomial of the form

X
3 X
n−l
aml x m y (n−l−m) = 0 (33)
l=0 m=0
12 ÜNSALAN AND ERÇİL

FIG. 7. Plot of the implicit form of curve given in Example 6.

( four homogeneous polynomials of consecutive degrees) can be converted to three para-

metric forms as

α1 (t) = (h 1 (t), g1 (t)) (34)

α2 (t) = (h 2 (t), g2 (t)) (35)
α3 (t) = (h 3 (t), g3 (t)) (36)
h 1 (θ) = (A + B) cos θ (37)
g1 (θ) = (A + B) sin θ (38)
µ ¶
A+B A − B√
h 2 (θ) = − + −3 cos θ (39)
2 2
µ ¶
A+B A − B√
g2 (θ) = − + −3 sin θ (40)
2 2
µ ¶
A+B A − B√
h 3 (θ) = − − −3 cos θ (41)
2 2
µ ¶
A+B A − B√
g3 (θ) = − − −3 sin θ, (42)
2 2

where
s r
3 b b2 a3
A= − + + (43)
2 4 27

FIG. 8. Plot of the first part of parametric curve in Example 6.

 PARAMETRIC AND IMPLICIT FORM CONVERSION 13

FIG. 9. Plot of the second part of parametric curve in Example 6.

s r
b3 b2 a3
B= − − + (44)
2 4 27
µ ¶
1 q p2
a= 3 − 2 (45)
3 s s
µ 3 ¶
1 p pq u
b= 2 3 − 9 2 + 27 , (46)
27 s s s

s, p, q are defined in Eqs. (28)–(30), and

X
n−3
u= am3 cosm θ sin(n−3−m) θ. (47)
m=0

Parameter values to obtain a real-valued parametric form specify the parameter range.

Proof. Equation (33) can be represented in polar coordinates as

X
3 X
n−l
r n−l aml cosm θ sin(n−l−m) θ = 0
l=0 m=0

X
3 X
n−l
r n−3 r 3−l aml cosm θ sin(n−l−m) θ = 0 (48)
l=0 m=0

sr 3 + pr 2 + qr + u = 0. (49)

Solving this cubic equation in terms of r and using polar-to-rectangular coordinate conver-
sion formulas will give parametric representations of the polynomial given in Eqs. (34)–(36).
A closed-form solution of this cubic equation can be found in [1].
LEMMA 5. A polynomial of the form

X
4 X
n−l
aml x m y (n−l−m) = 0 (50)
l=0 m=0
14 ÜNSALAN AND ERÇİL

( five homogeneous polynomials of consecutive degrees) can be converted to four parametric

form as

α1 (t) = (h 1 (t), g1 (t)) (51)

α2 (t) = (h 2 (t), g2 (t)) (52)
α3 (t) = (h 3 (t), g3 (t)) (53)
α4 (t) = (h 4 (t), g4 (t)) (54)
µ ¶
p R D
h 1 (θ) = − + + cos θ (55)
4s 2 2
µ ¶
p R D
g1 (θ) = − + + sin θ (56)
4s 2 2
µ ¶
p R D
h 2 (θ) = − + − cos θ (57)
4s 2 2
µ ¶
p R D
g2 (θ) = − + − sin θ (58)
4s 2 2
µ ¶
p R E
h 3 (θ) = − + + cos θ (59)
4s 2 2
µ ¶
p R E
g3 (θ) = − + + sin θ (60)
4s 2 2
µ ¶
p R E
h 4 (θ) = − + − cos θ (61)
4s 2 2
µ ¶
p R E
g4 (θ) = − + − sin θ. (62)
4s 2 2

Let l be any root of the equation

µ ¶
q pu p2 v qv u2
l − l2 +
3
− 4v l − + 4 − = 0. (63)
s s2 s3 s2 s2

Then
r
p q
R= − +l (64)
4s s
r
3 p2 q 4 pqs − 8us 2 − p 3
D= 2
− R2 − 2 + (65)
4s s 4s 3 R
r
3 p2 q 4 pqs − 8us 2 − p 3
E= − R 2−2 − . (66)
4s 2 s 4s 3 R

s, p, q are defined in Eqs. (28)–(30); u is defined in Eq. (47);

X
n−4
v= am4 cosm θ sin(n−4−m) θ. (67)
m=0

Parameter values to obtain a real-valued parametric form specify the parameter range.
 PARAMETRIC AND IMPLICIT FORM CONVERSION 15

Proof. Equation (50) can be represented in polar coordinates as

X
4 X
n−l
r n−l aml cosm θ sin(n−l−m) θ = 0
l=0 m=0

X
4 X
n−l
r n−4 r 4−l aml cosm θ sin(n−l−m) θ = 0 (68)
l=0 m=0

sr 4 + pr 3 + qr 2 + ur + v = 0. (69)

Solving this quartic equation in terms of r and using polar to rectangular coordinate con-
version formulas will give parametric representations of the polynomial given in Eqs. (51)–
(54). A solution of this quartic equation can be found in [1].
THEOREM 5 (Conversion in 2D). Exact parametric representation for implicit polyno-
mials up to degree 4 or polynomials that are sums of up to five homogeneous polynomials
can be obtained by representing them in polar coordinates and solving the corresponding
equation to obtain r (θ).
Proof. The proof follows from Lemmas 2–5.
Note that the implicit polynomials can be converted to corresponding parametric forms
if the obtained equation of r (θ) has real roots. In case the equation does not have any real
roots, the implicit polynomial does not have a corresponding parametric form.
As a special case, higher order polynomials that can be factored into terms of at most
fourth-order polynomials can also be converted to parametric form term by term. An example
is given for such a situation.
EXAMPLE 8. Consider the implicit polynomial of the form

x 22 y 10 + x 14 y 18 − 10−5 x 14 y 10 − 1.9x 20 y 12 − 1.9x 12 y 20 + 19 × 10−6 x 12 y 12 + x 18 y 14

+ x 10 y 22 − 10−5 x 10 y 14 − 0.5x 18 y 10 − 0.5x 10 y 18 + 5 × 10−6 x 10 y 10 = 0.

The plot of this implicit polynomial is given in Fig. 10.

This implicit polynomial can be factored as

(x 4 − 1.9x 2 y 2 + y 4 − 0.5)(x 8 + y 8 − 10−5 )x 10 y 10 = 0.

FIG. 10. Shape of implicit polynomial in Example 8.

16 ÜNSALAN AND ERÇİL

Converting this implicit polynomial term by term proceeds as follows:

First term:

r 4 (cos4 θ − 1.9 sin2 θ cos2 θ + sin4 sin2 θ ) = 0.5

µ ¶1/4
0.5
h 1 (θ) = cos θ
(cos4 θ − 1.9 sin2 θ cos2 θ + sin4 sin2 θ )
µ ¶1/4
0.5
g1 (θ) = sin θ θ ∈ [0, 2π].
(cos4 θ − 1.9 sin2 θ cos2 θ + sin4 sin2 θ )

Second term:

r 8 (cos8 θ + sin8 θ ) = 10−5

µ ¶1/8
10−5
h 2 (θ) = cos θ
(cos8 θ + sin8 θ )
µ ¶1/8
10−5
g2 (θ) = sin θ θ ∈ [0, 2π].
(cos8 θ + sin8 θ )

Third term:

r 10 (cos10 θ + sin10 θ ) = 0
h 3 (θ ) = 0, g3 (θ ) = 0.

LEMMA 6 (Conversion for Higher Order Implicit Polynomials). Implicit polynomials

that are of order higher than 4 and cannot be written as sums of five homogeneous polyno-
mials can be converted to an approximate parametric form.
Proof. The conversion procedure is the same as in Lemmas 2–5. However, since there is
no analytical solution to an equation of degree higher than 4, the solution is approximated by
numerical techniques. A Fourier fitting technique is then applied to obtain the approximate
parametric shape. j
EXAMPLE 9. An implicit polynomial of the form f (x, y) = 0 is given to represent
an object plotted in Fig. 11. This implicit polynomial is converted to an approximated
parametric form and both of the representations are plotted in Fig. 12:

f (x, y) = 4663633x 2 y 2 − 51081695y 3 + 5603905960x y − 276518665x 3 + 408x 5 y

+ 1421498y 4 + 30547238788903 + 622833383233x + 72289297932y
− 9748431770y 2 + 4495262x 4 − 638483x y 3 − 18536463436x 2

FIG. 11. Implicit plot of an object.

 PARAMETRIC AND IMPLICIT FORM CONVERSION 17

FIG. 12. Parametric approximation to the shape and the original object.

− 3285437x 3 y + 45302709x 2 y + 31700x 5 − 693x 4 y 2 − 500x 6

− 159136086x y 2 − 2x y 5 − 119y 6 − 364x 2 y 4 + 349x 3 y 3 + 7017y 5
− 16672x 4 y + 37939x 3 y 2 − 5499x 2 y 3 + 8673x y 4 = 0.

In polar coordinate representation

f (r, θ) = (408 cos θ sin θ + 57 sin θ cos5 θ + 807 cos2 θ − 478 cos4 θ + 52 cos6 θ − 500
− 467 cos3 θ sin θ)r 6 + (31700 sin θ − 25461 sin θ cos2 θ + 2434 cos4 θ sin θ
− 4156 cos5 θ − 16672 cos θ + 27845 cos3 θ )r 5 + (1253127 cos4 θ
+ 2646954 cos3 θ sin θ − 4326891 cos2 θ − 3285437 cos θ sin θ + 4495262)r 4
+ (−96384404 cos3 θ − 276518665 sin θ + 45302709 cos θ
+ 117382579 sin θ cos2 θ )r 3 + (−18536463436 + 5603905960 cos θ sin θ
+ 8788031666 cos2 θ)r 2 + (72289297932 cos θ + 622833383233 sin θ )r
+ 30547238788903 = 0.

For each angle value θ, r (θ), which is sixth-order one-valued polynomial, is solved to obtain
the corresponding positive, real radius value, and Fourier fitting is applied to these points to
obtain the approximate parametric shape of the object. A fitted parametric curve is plotted
on top of the orignal object in Fig. 12.

3.2. Conversion in 3D
In this section, we will give results on converting an implicit polynomial in three variables
to a corresponding parametric form. The techniques used are similar to those in 2D except
that spherical coordinate representations will be used instead of polar representations. Proofs
of the following lemmas are similar to the proofs of lemmas given for 2D; hence they are
omitted.
LEMMA 7. A polynomial of the form
X
n
£ ¤ Xn X
n
£ ¤
am x m y (n−m) +bm x m z (n−m) +cm y m z (n−m) + dmt x m y t z (n−m−t)
m=0 m=1 t=1

X
k
£ ¤ Xk X
k
£ ¤
+ em x m y (k−m) + f m y m z (k−m) + gm y m z (k−m) + h mt x m y t z (k−m−t) =0 (70)
m=0 m=1 t=1

can be written as a patch of the form

18 ÜNSALAN AND ERÇİL

x(θ, ϕ)
v
u P
u k £ m (k−m) ¤ P k P k £ ¤
u em α β + f m α m γ (k−m) + gm β m γ (k−m) + h mt α m β t γ (k−m−t)
u m=0
= u− n £
m=1 t=1
¤ sin ϕ cos θ
n−k

t P ¤ P n P n £
am α y
m (n−m) + bm α γ
m (n−m) + cm β γ
m (n−m) + dmt α β γ
m t (n−m−t)
m=0 m=1 t=1
(71)
y(θ, ϕ)
v
u P
u k £ m (k−m) ¤ P k P k £ ¤
u em α β + f m α m γ (k−m) + gm β m γ (k−m) + h mt α m β t γ (k−m−t)
u
= u− n £
m=0 m=1 t=1
¤ sin ϕ sin θ
n−k

t P ¤ P n P n £
am α y
m (n−m) + bm α γ
m (n−m) + cm β γ
m (n−m) + dmt α β γ
m t (n−m−t)
m=0 m=1 t=1
(72)
z(θ, ϕ)
v
u P
u k £ m (k−m) ¤ P k P k £ ¤
u em α β + f m α m γ (k−m) + gm β m γ (k−m) + h mt α m β t γ (k−m−t)
u
= u− n £
m=0 m=1 t=1
¤ cos ϕ,
n−k

t P ¤ P n P n £
am α y
m (n−m) + bm α γ
m (n−m) + cm β γ
m (n−m) + dmt α β γ
m t (n−m−t)
m=0 m=1 t=1

(73)

where

α = sin ϕ cos θ (74)

β = sin ϕ sin θ (75)
γ = cos ϕ. (76)

Parameter values to obtain a real-valued patch specify parameter ranges.

EXAMPLE 10. Superquadratics in 3D can also be converted to parametric patch form by
Lemma 7. A general form of a superquadratic in 3D can be given as
"µ ¶ µ ¶2/ε2 #ε2 /ε1 µ ¶2/ε1
x 2/ε2 y z
+ h = 1, (77)
a1 a2 a3

obtaining R(θ, ϕ) as
,""µ ¶2/ε2 µ ¶2/ε2 #ε2 /ε1 µ ¶2/ε1 #ε1 /2
cos θ sin ϕ sin θ sin ϕ cos ϕ
R(θ, ϕ) = 1 + h .
a1 a2 a2

The parametric patch form for the general superquadratic in 3D is given as

x(θ, ϕ) = R(θ, ϕ) cos θ sin ϕ, y(θ, ϕ) = R(θ, ϕ) sin θ sin ϕ, z(θ, ϕ) = R(θ, ϕ) cos ϕ.

LEMMA 8. A polynomial of the form

" n−l
X2 X£ ¤
aml x m y (n−l−m) + bml x m z (n−l−m) + cml y m z (n−l−m)
l=0 m=0
#
X
n−l X
n−l
£ ¤
+ m t (n−l−m−t)
dmtl x y z =0 (78)
m=1 t=1
 PARAMETRIC AND IMPLICIT FORM CONVERSION 19

can be written as two patches of the forms

p
−p + p 2 − 4sq
x 1 (θ, ϕ) = sin ϕ cos θ (79)
2s
p
− p + p 2 − 4sq
y1 (θ, ϕ) = sin ϕ sin θ (80)
2s
p
− p + p 2 − 4sq
z 1 (θ, ϕ) = cos ϕ (81)
2s
p
− p − p 2 − 4sq
x2 (θ, ϕ) = sin ϕ cos θ (82)
2s
p
− p − p 2 − 4sq
y2 (θ, ϕ) = sin ϕ sin θ (83)
2s
p
− p − p 2 − 4sq
z 2 (θ, ϕ) = cos ϕ (84)
2s
Xn
£ ¤
s= am0 α m y (n−m) + bm0 α m γ (n−m) + cm0 β m γ (n−m)
m=0

X
n X
n
£ ¤
+ dmt0 α m β t γ (n−m−t) (85)
m=1 t=1

X
n−1
£ ¤
p= am1 α m y (n−1−m) + bm1 α m γ (n−1−m) + cm1 β m γ (n−1−m)
m=0

X
n−1 X
n−1
£ ¤
+ dmt1 α m β t γ (n−1−m−t) (86)
m=1 t=1

X
n−2
£ ¤
q= am2 α m y (n−2−m) + bm2 α m γ (n−2−m) + cm2 β m γ (n−2−m)
m=0

X
n−2 X
n−2
£ ¤
+ dmt2 α m β t γ (n−2−m−t) . (87)
m=1 t=1

α, β, γ are defined in Eqs. (74)–(76).

Parameter values to obtain real-valued patches specify parameter ranges.
LEMMA 9. A polynomial of the form
" n−1
X3 X£ ¤
aml x m y (n−l−m) + bml x m z (n−l−m) + cml y m z (n−l−m)
l=0 m=0
#
X
n−l X
n−l
£ ¤
+ dmtl x m y t z (n−l−m−t)
=0 (88)
m=1 t=1

can be written as three patches of the forms

x1 (θ, ϕ) = (A + B) sin ϕ cos θ (89)

y1 (θ, ϕ) = (A + B) sin ϕ sin θ (90)
z 1 (θ, ϕ) = (A + B) cos ϕ (91)
20 ÜNSALAN AND ERÇİL

µ ¶
A+B A − B√
x2 (θ, ϕ) = − + −3 sin ϕ cos θ (92)
2 2
µ ¶
A+B A − B√
y2 (θ, ϕ) = − + −3 sin ϕ sin θ (93)
2 2
µ ¶
A+B A − B√
z 2 (θ, ϕ) = − + −3 cos ϕ (94)
2 2
µ ¶
A+B A − B√
x3 (θ, ϕ) = − − −3 sin ϕ cos θ (95)
2 2
µ ¶
A+B A − B√
y3 (θ, ϕ) = − − −3 sin ϕ sin θ (96)
2 2
µ ¶
A+B A − B√
z 3 (θ, ϕ) = − − −3 cos ϕ, (97)
2 2
where
s r
3 b b2 a3
A = − + + (98)
2 4 27
s r
3 b b2 a3
B = − − + (99)
2 4 27
µ 2¶
1 q p
a = 3 − 2 (100)
3 s s
µ 3 ¶
1 p pq u
b = 2 3 − 9 2 + 27 . (101)
27 s s s
s, p, q are defined in Eqs. (85)–(87).

X
n−3
£ ¤
u= am3 α m y (n−3−m) + bm3 α m γ (n−3−m) + cm3 β m γ (n−3−m)
m=0

X
n−3 X
n−3
£ ¤
+ dmt3 α m β t γ (n−3−m−t) , (102)
m=1 t=1

and α, β, γ are defined in Eqs. (74)–(76).

Parameter values to obtain real-valued patches specify parameter ranges.
LEMMA 10. A polynomial of the form
" n−l
X4 X£ ¤
aml x m y (n−l−m) + bml x m z (n−l−m) + cml y m z (n−l−m)
l=0 m=0
#
X
n−l X
n−l
£ ¤
+ dmtl x m y t z (n−l−m−t)
=0 (103)
m=1 t=1

can be written as four patches of the forms

µ ¶
p R D
x1 (θ, ϕ) = − + + sin ϕ cos θ (104)
4s 2 2
 PARAMETRIC AND IMPLICIT FORM CONVERSION 21

µ ¶
p R D
y1 (θ, ϕ) = − + + sin ϕ sin θ (105)
4s 2 2
µ ¶
p R D
z 1 (θ, ϕ) = − + + cos ϕ (106)
4s 2 2
µ ¶
p R D
x2 (θ, ϕ) = − + − sin ϕ cos θ (107)
4s 2 2
µ ¶
p R D
y2 (θ, ϕ) = − + − sin ϕ sin θ (108)
4s 2 2
µ ¶
p R D
z 2 (θ, ϕ) = − + − cos ϕ (109)
4s 2 2
µ ¶
p R E
x3 (θ, ϕ) = − + + sin ϕ cos θ (110)
4s 2 2
µ ¶
p R E
y3 (θ, ϕ) = − + + sin ϕ sin θ (111)
4s 2 2
µ ¶
p R E
z 3 (θ, ϕ) = − + + cos ϕ (112)
4s 2 2
µ ¶
p R E
x4 (θ, ϕ) = − + − sin ϕ cos θ (113)
4s 2 2
µ ¶
p R E
y4 (θ, ϕ) = − + − sin ϕ sin θ (114)
4s 2 2
µ ¶
p R E
z 4 (θ, ϕ) = − + − cos ϕ. (115)
4s 2 2
Let o be any root of the equation
µ ¶
q pu p2 v qv u2
o3 − o2 + 2
− 4v o − 3 + 4 2 − 2 = 0; (116)
s s s s s
then
r
p q
R= − +o (117)
4s s
r
3 p2 q 4 pqs − 8us 2 − p 3
D= − R 2−2 + (118)
4s 2 s 4s 3 R
r
3 p2 q 4 pqs − 8us 2 − p 3
E= − R 2−2 − . (119)
4s 2 s 4s 3 R
s, p, q are defined in Eqs. (85)–(87); u is defined in Eq. (97); and
X
n−4
£ ¤
v= am4 α m y (n−4−m) + bm4 α m γ (n−4−m) + cm4 β m γ (n−4−m)
m=0

X
n−4 X
n−4
£ ¤
+ dmt4 α m β t γ (n−4−m−t) . (120)
m=1 t=1

α, β, γ are defined in Eqs. (74)–(76).

Parameter values to obtain real-valued patches specify parameter ranges.
22 ÜNSALAN AND ERÇİL

THEOREM 6 (Conversion in 3D). Exact parametric patch representations for implicit

polynomials up to degree 4 can be obtained by representing them in spherical coordinates
and solving the corresponding equation to obtain R(θ, ϕ).
Proof. See Lemmas 7–10. j

THEOREM 7 (Conversion from Intersecting Implicit Surface Representation to Parametric

Space Curve Representation). Let α be a space curve in <3 defined by the intersection of
two implicit surfaces as f (x, y, z) = 0, g(x, y, z) = 0. α can be converted to a parametric
space curve, α: (a, b) → <3 , with α(t) = (h(t), g(t), k(t)), where (a, b) is an open interval
in <, if an explicit relationship can be established between θ and ϕ.
Proof. The implicit polynomials f (x, y, z) = 0, g(x, y, z) = 0 can be represented in
spherical coordinates as f (R, θ, ϕ) = 0, g(R, θ, ϕ) = 0. For each of these equalities R can
be obtained by solving the R polynomial as in the previous lemmas.
Let R = F(θ, ϕ) and R = G(θ, ϕ) be obtained from these two implicit polynomials.
Since the space curve lies on both of the implicit surfaces, R values that are the same for these
two implicit surfaces represent this space curve. So we can write R = F(θ, ϕ) = G(θ, ϕ).
If an explicit relationship can be established between θ and ϕ, as θ = m(ϕ) or ϕ = n(θ ),
then a parametric space curve is obtained from the standard change of variable formulas
given in Eqs. (12)–(15) as follows:
Assume that there is a relationship s.t. ϕ = n(θ ). Then
h(θ) = R cos θ sin(n(θ)) g(θ ) = R sin θ sin(n(θ )) k(θ ) = R cos(n(θ)),
where
R = F(θ, n(θ )) or R = G(θ, n(θ )).
The same derivation can be done for θ = m(ϕ). j

Implicit polynomials which define the space curve may yield simpler solutions as in the
following example. In this example, an explicit relationship is not searched for since R can
be found such that a space curve is obtained easily.
EXAMPLE 11. Let the two implicit polynomials plotted in Fig. 8, x 2 + y 2 = 1 and z =
y , represent a space curve. Converting x 2 + y 2 = 1 into spherical form gives R 2 (cos2 θ +
5

sin2 θ) sin2 ϕ = 1. Solving the above equation for R,

1
R= ,
sin ϕ
we obtain the parametric space curve α: (0, 2π ) → <3 α(t) = (h(t), g(t), k(t)), h(t) = cos t,
g(t) = sin t, k(t) = sin5 t, which is plotted in red on the original implicit surfaces given in
Fig. 13. It is seen that α(t) actually lies in the intersection region.

4. INSPECTION BY IMPLICIT POLYNOMIALS IN 3D

In this section, the theorems introduced in the previous section are applied to a real-world
application, automated tolerance inspection by implicit polynomials in 3D. It is assumed
that the object to be inspected is represented as an implicit polynomial form. The following
theorem is given to inspect whether a given point is inside the tolerance values or not. More
information about this method can be found in [15].
 PARAMETRIC AND IMPLICIT FORM CONVERSION 23

THEOREM 8 (Inspection by Implicit Polynomial Representations). Given a tolerance

band ±d and an object having an implicit polynomial model f (x, y, z) = 0, to observe
whether a point (x1 , y1 , z 1 ) is inside tolerances, it is sufficient to check the existence of
a real-valued space curve defined by two implicit polynomials, f (x, y, z) = 0 and (x −
x1 )2 + (y − y1 )2 + (z − z 1 )2 = d 2 .
EXAMPLE 12. In this example, a simple model for water pipe in implicit polynomial
form, x 2 + y 2 = 1, is given. We will illustrate the technique for inspecting a single point
(0.8, 0.6, 0). The tolerance value is taken as ±0.1. The water pipe model and the point are
plotted in Fig. 14.
The space curve in implicit polynomial form for inspection is given as

x 2 + y 2 + z 2 = 0.12
(x − 0.8)2 + (y − 0.6)2 = 1.

Converting the space curve to parametric form gives

1.6 cos θ + 1.2 sin θ

R= or R = 0.
sin ϕ
h(θ) = (1.6 cos θ + 1.2 sin θ ) cos θ
g(θ) = (1.6 cos θ + 1.2 sin θ ) sin θ
p
k(θ) = 0.12 − (1.6 cos θ + 1.2 sin θ )2

θ ∈ (0.6888, 0.7205555) ∪ (0.6888 + π, 0.7205555 + π ).

There exists at least one real θ value such that the space curve exists. So the point is inside
the tolerances.

5. CONCLUSIONS

In this paper, a new method is introduced for conversions between parametric and im-
plicit forms based on polar/spherical coordinate representations of an object boundary. The
parametric-to-implicit conversions are carried out for star-shaped objects in 2D and 3D.
Theorems in the paper give methods to convert parametric representations to implicit rep-
resentations in a direct way. The only requirement is to obtain an invertible slope function.
The implicit-to-parametric conversions are carried out exactly for implicit polynomials up
to degree 4 or for polynomials that are sums of up to five homogeneous polynomials in
2D and 3D. The conversion is done approximately for higher degree polynomials in 2D
since there exists no method for solving polynomials of order greater than 4. The methods
are applied to an automated tolerance inspection problem for demonstration of a real-life
application of the techniques developed.

6. APPENDIX A

We can represent cos nθ and sin nθ for any integer n, from recursive calculations starting
from cos 2θ and sin 2θ as:
24 ÜNSALAN AND ERÇİL

For cosine terms:

cos 2θ = cos2 θ − sin2 θ (1)

cos 3θ = cos θ − 3 sin θ cos θ
3 2
(2)
cos 4θ = cos4 θ − 6 sin2 θ cos2 θ + sin4 θ (3)
...

For sine terms:

sin 2θ = 2 sin θ cos θ (4)

sin 3θ = 3 sin θ cos θ − sin θ
2 3
(5)
sin 4θ = 4 sin θ cos3 θ − 4 cos θ sin3 θ (6)
...

7. APPENDIX B

We construct an implicit polynomial form in 2D for n = 3:

X
3
r = a0 + an cos nθ + bn sin nθ
n=1
r = a0 + a1 cos θ + b1 sin θ + a2 cos 2θ + b2 sin 2θ + a3 cos 3θ + b3 sin 3θ
r = a0 + a1 cos θ + b1 sin θ + a2 (cos2 θ − sin 2 θ ) + b2 2 sin θ cos θ
+ a3 (cos3 θ − 3 cos θ sin2 θ + b3 (3 sin θ cos2 θ − sin3 θ )
a 1 x + b1 y a2 x 2 + b2 2x y − a2 y 2 a3 x 3 + b3 3x 2 y − a3 3x y 2 − b3 y 3
r = a0 + + +
r r2 r3
(x + y) ⊗ C1 (x + y) ⊗ C2
2
(x + y) ⊗ C3
3
r = a0 + + + .
r r2 r3

The final implicit function form is obtained as

X
3
Kn
(x 2 + y 2 )1/2 = ,
n=0
(x 2 + y 2 )n/2

where K n = (x + y)n ⊗ Cn and K 0 = a0 . The above equation can be converted to implicit

polynomial form as

K1 K2 K3
(x 2 + y 2 )1/2 = K 0 + + 2 +
(x 2 + y 2 )1/2 (x + y 2 ) (x 2 + y 2 )3/2
((x 2 + y 2 )2 − K 1 (x 2 + y 2 ) − K 3 )2 = (x 2 + y 2 )(K 0 (x 2 + y 2 ) + K 2 )2 .

ACKNOWLEDGMENTS

This study is partially supported by NSF Grant NSF—465, Tubitak Grant MISAG—116, Bogaziçi University
Research Grant 99A0301, and EUREKA 1770 Project.
 PARAMETRIC AND IMPLICIT FORM CONVERSION 25

REFERENCES

1. W. H. Beyer, CRC Standard Mathematical Tables, CRC Press, Boca Raton, FL, 1979.
2. J. Bloomenthal, Ed., Introduction to Implicit Surfaces, Morgan Kaufmann, San Francisco, 1995.
3. H. Çivi, C. Christopher, and A. Erçil, The Classical Theory of Invariants and Object Recognition using
Algebraic Curves and Surfaces, Technical Report FBE–IE–06/97–07, Boğaziçi University, 1997.
4. M. Hebert, J. Ponce, T. Boult, and A. Gross, Eds., Object Representation in Computer Vision, Lecture Notes
in Computer Science. Springer-Verlag, Berlin/New York, 1995.
5. C. Hoffman, Geometric and Solid Modeling: An Introduction, Morgan Kauffman, San Francisco, 1989.
6. C. Hoffman, Implicit curves and surfaces in computer-aided geometric design, IEEE Comput. Graph. Appl.
13(1), 1993, 79–88.
7. D. Manocha and J. F. Canny, Implicit representation of rational parametric surfaces, J. Symbol. Comput. 13,
1992, 485–510.
8. G. M. Nielson, GAGD’s top ten: What to watch, IEEE Comput. Graph. Appl. 1993, 35–37.
9. G. A. Salmon, Treatise on the Algebraic Geometry of Three Dimensions (R. Rogers, Ed.), Vols. 1 and 2,
Chelsea, New York, 1914. [Reprint]
10. T. W. Sederberg, Techniques for cubic algebraic surfaces, Part one, IEEE Comput. Graph. Appl. 10(4), 1990,
14–25.
11. T. W. Sederberg, Techniques for cubic algebraic surfaces, Part two, IEEE Comput. Graph. Appl. 1990, 12–21.
12. T. W. Sederberg, D. C. Anderson, and R. N. Goldman, Implicit representation of parametric curves and
surfaces, Comput. Vision Graph. Image Process. 1984, 72–84.
13. J. Subrahmonia, D. Cooper, and D. Keren, Practical Reliable Bayesian Recognition of 2D and 3D Objects
Using Implicit Polynomials and Algebraic Invariants, Technical Report LEMS–107, Division of Engineering,
Brown University, June 1992.
14. G. Taubin and D. B. Cooper, Recognition and Positioning of 3D Piecewise Algebraic Objects Using Euclidean
Invariants, IBM Research Division, T.J. Watson Research Center, 1990.
15. C. Unsalan and A. Ercil, Automated tolerance inspection by implicit polynomials, in Proceedings of ICIAP
99 10th International Conference on Image Analysis and Processing Image-Based Emerging Applications,
Venice, 1999, pp. 1218–1225.
16. M. Yalniz, Algebraic Invariants for Implicit Polynomials. Case Study: Patterns in Turkish Hand-Woven
Carpets, M.S. thesis, Bogaziçi University, Turkey.