Projective Geometry

We are all familiar with Euclidean geometry and with the fact that it describes our three
dimensional world so well. In Euclidean geometry, the sides of objects have lengths, intersecting
lines determine angles between them, and two lines are said to be parallel if they lie in the same
plane and never meet. Moreover, these properties do not change when the Euclidean
transformations (translation and rotation) are applied. Since Euclidean geometry describes our
world so well, it is at first tempting to think that it is the only type of geometry. (Indeed, the
word geometry means measurement of the earth.") However, when we consider the imaging
process of a camera, it becomes clear that Euclidean geometry is insufficient: Lengths and angles
are no longer preserved, and parallel lines may intersect
Euclidean geometry is actually a subset of what is known as projective geometry. In fact,
there are two geometries between them: similarity and affine. To see the relationships between
these different geometries, consult Figure 1. Projective geometry models well the imaging
process of a camera because it allows a much larger class of transformations than just
translations and rotations, a class which includes perspective projections. Of course, the
drawback is that fewer measures are preserved | certainly not lengths, angles, or parallelism.
Projective transformations preserve type (that is, points remain points and lines remain lines),
incidence (that is, whether a point lies on a line), and a measure known as the cross ratio, which
will be described in section 2.4.
Projective geometry exists in any number of dimensions, just like Euclidean geometry. For
example the projective line, which we denote by P1 , is analogous to a one-dimensional
Euclidean world; the projective plane, P2 , corresponds to the Euclidean plane; and projective
space, P3 , is related to three-dimensional Euclidean space. The imaging process is a projection
from P3 to P2 , from three-dimensional space to the two-dimensional image plane. Because it is
easier to grasp the ma jor concepts in a lower-dimensional space, we will spend the bulk of our
effort, indeed all of section 2, studying P 2 , the projective plane. That section presents many
concepts which are useful in understanding the image plane and which have analogous concepts
in P3 . The final section then briefly discusses the relevance of projective geometry to computer
vision, including discussions of the image formation equations and the Essential and
Fundamental matrices.
From the ancient beginnings of geometry until well into the nineteenth century it was almost
universally accepted that the geometry of the space we live in is the only geometry conceivable
by man. This point of view was most eloquently formulated by the German philosopher
Immanuel Kant (1724-1804). Ironically, shortly after Kant's death the discovery of non-
Euc1idean geometry by GauB, Lobachevski, and Bolyai made his position untenable. Today, we
study in mathematics not just one geometry, or two geometries, but an infinity of geometries.
This means: When you start learning geometry (the subject), you are immediately offered
geometries (structures) in plural .There is not a unique geometry, but many geometries, and all
have equal rights (even though some might be more interesting than others).
Figure 01: The four diff erent geometries, the transformati ons allowed in each, and the
measures that remain invariant under those transformati ons.

The purpose of this monograph will be to provide a readable introduction to the field of
projective geometry and a handy reference for some of the more important equations. The first-time
reader may and some of the examples and derivations excessively detailed, but this thoroughness should
prove helpful for reading the more advanced texts, where the details are often omitted.

Historical Background
Projective geometry has its origins in the early Italian Renaissance, particularly in the
architectural drawings of Filippo Brunelleschi (1377–1446) and Leon Battista Alberti (1404–72),
who invented the method of perspective drawing. By this method, as shown in the figure, the eye
of the painter is connected to
points on the landscape (the
horizontal reality plane, RP) by
so-called sight lines. The
intersection of these sight lines
with the vertical picture plane
(PP) generates the drawing.
Thus, the reality plane is
projected onto the picture
plane, hence the
name projective geometry.
Although some isolated
properties concerning projections were known in antiquity, particularly in the study of optics, it
was not until the 17th century that mathematicians returned to the subject. The French
mathematicians Girard Desargues (1591–1661) and Blaise Pascal (1623–62) took the first
significant steps by examining what properties of figures were preserved (or invariant) under
perspective mappings. The subject’s real importance, however, became clear only after 1800 in
the works of several other French mathematicians, notably Jean-Victor Poncelet (1788–1867). In
general, by ignoring geometric measurements such as distances and angles, projective geometry
enables a clearer understanding of some more generic properties of geometric objects. Such
insights have since been incorporated in many more advanced areas of mathematics.
Foundations
It is typical of geometry that we study not only one type of object (such as points), but
different types of objects (such as points and lines, points, lines, and planes, etc.) and their
relationships. We first define a very general notion of geometry which can be used to
describe all possible geometries. Definition. A geometry is a pair G = en, I), where n is a set
and I a relation on n that is symmetric and reflexive; this means the following. - If (x, y) E I
then also (y, x) E I. - (x, x) E I for all x E n. What is the idea behind this definition? The idea
is that the set n contains all geometrically relevant objects and I describes their being
'incident'. Let us consider some examples. Before doing this we note the follO\ving. In many
situations the 'natural' incidence relation is set-theoretical inclusion. This relation is not
symmetric, but can easily be made symmetric by defining that two elements are incident if
one is contained in the other. Usually, we shall not mention this explicitly.
Projective geometry is concerned with properties of incidence—properties which are invariant
under stretching, translation, or rotation of the plane. Thus in the axiomatic development of the
theory, the notions of distance and angle will play no part.
Basic Definitions
Definition 1.1. A geometry is denoted G = (Ω, I), where Ω is a set and I a relation which is
both symmetric and reflexive. The relation on a geometry is called an incidence relation. For
example, consider the tradional Euclidean geometry. In this geometry, the objects of the set Ω
are points and lines. A point is incident to a line if it lies on that line, and two lines are incident if
they have all points in common - only when they are the same line. There is often this same
natural division of the elements of Ω into different kinds such as the points and lines.
Definition 1.2. Suppose G = (Ω, I) is a geometry. Then a flag of G is a set of elements of Ω
which are mutually incident. If there is no element outside of the flag, F, which can be added and
also be a flag, then F is called maximal.
Definition 1.3. A geometry G = (Ω, I) has rank r if it can be partitioned into sets Ω1, . . . , Ωr
such that every maximal flag contains exactly one element of each set. The elements of Ωi are
called elements of type i.
Lemma 1.4. Remark: Let G be a geometry of rank r. Then no two distinct elements of the
same type are incident.
 Proof. Suppose not. Then there exist two distinct elements of the same type which are
incident. Then these elements, by definition form a flag. Now, these elements must be elements
of some maximal flag, F. But then F has two elements of the same type, but this is a
contradiction because G is a geometry of rank r. Thus, as we saw with the Euclidean geometry,
two lines are incident if and only if they are truly the same line. Often for geometries of rank 2
the types of elements are termed points and lines. This is the case for the projective spaces which
are the focus of this paper.
A geometry G = (Q, I) of rank 2 is often called an incidence structure. In this case one calls
the elements of type I points and the elements of type 2 blocks. If G is an incidence structure
with point set fJ' and block set ffi then one also writes G = (fP, ffi, I).
Henceforth, let G = (P, L, I) be a geometry of rank two with elements of P termed points, and
those of L termed lines. There are many different such geometries which satisfy the following
axioms, all of which are types of projective geometries.
Axiom 1 (Line Axiom). For every two distinct
points there is one distinct line incident to them.

Axiom 2 (Veblen-Young). If there are points A, B, C,

D such that AB intersects CD, then AC intersects BD. That
is to say, any two lines of a ’plane’ meet.

1. The Veblen-Young axiom is a truly ingenious way of saying that any two lines of a plane
meet - before one knows what a plane is.
2. Some people call the Veblen-Young
axiom the axiom of Pasch, because the German
geometer Moritz Pasch (1843-1930) used a
similar picture.

Axiom 3. Any line is incident with at least three points.

A projective space is a geometry P = (H', .£, I) of rank 2 that satisfies the axioms 1,2, and 3.
A projective space P is called nondegenerate if it also satisfies the following axiom 4.
 Axiom 4. There are at least two lines. In projective geometries, the above axioms imply that
there are no ’parallel’ lines. That is, there are no lines lying in the same plane which do not
intersect. The following lemma is derived easily from these axioms.
Lemma 2.1. Any two distinct lines are incident with at most one common point.
Proof. Suppose g and h are two distinct lines, but share more than one common point. By
Axiom 1, two distinct points cannot both be incident with two distinct points, so g = h.
The above axioms are used to define the following general structures.
Definition. A projective plane is a
nondegenerate projective space in which axiom 2
is replaced by the stronger axiom 2':

Definition 2.3. A projective plane is a

nondegenerate projective space with Axiom 2 replaced by the stronger statement: Any two lines
have at least one point in common. It is not too difficult to show that projective planes are indeed
two dimensional as expected, although the notion of dimension for a geometry is defined further
into the paper. A projective plane is therefore what one might naturally consider it to be. It is a
plane, according to the usual conception of such, in which all lines meet as is expected from the
term projective.
Structure of projective geometry
Definition. A subset U of the point set is called linear if for any two points in U all points on
the line from one to the other are also in U.
It is often useful to consider all the points on a give line, so we denote this by letting (g) be
the set of points incident with the line g. Just as in linear algebra, the notion of a subspace is
remarkably useful. For geometries, it is quite natural to consider a subset of the points of the
geometry as a subspace. However, to make this well defined, we must ensure that the same
incidence structure makes sense. Thus we have the following definition.
Definition 3.2. A space P(U) = (U, L0 , I0 ) is a (linear) subspace of P, where L 0 is the set
of lines contained in U and I 0 is the induced incidence. Also, the span of subset X is defined as:
hX i = ∩{U | X ⊆ U, a linear set} Then X spans hX i. From these definitions we can finally
formally define the notion of a plane, which we already have an intuitive conception of.
Definition 3.3. A set of points is collinear if all points are incident with common line;
otherwise, it is called noncollinear. A plane is the span of a set of three noncollinear points. The
following Theorems and Lemmas should look familiar from linear algebra. Their proofs are not
significantly different from their respective counterparts, and thus they will be given without
proof as a reference for the rest of the paper.
Theorem 3.4. A set B of points of P is a basis if and only if it is a minimal spanning set. An
important theorem regarding the basis holds here as well. Every independent set can be
completed to form a basis of the whole space. The straightforward proof, which is along similar
lines as that of the corresponding proof from linear algebra, is not given here.
Theorem 3.5 (Basis Extension Theorem). Let P a finitely generated projective space. Then
all bases of P have the same number of elements, and any independent set can be extened to a
basis. The basis of a geometry is a fundamental property of a specific structure. Thus there is a
name related to the number of elements which are in such a basis.
Definition 3.6. Suppose P is a finitely generated projective space. Then the dimension of P
is one less than the number of elements in a basis. Likewise, the subspaces of a space also have
dimension, and some of these subspaces are classified accordingly.
Definition 3.7. Let P have diminsion d. Then subspaces of dimension 2 are called planes,
and subspaces of dimension d − 1 are called hyperplanes. Finally, we give a very important
theorem from linear algebra which appears all over mathematics. The proof is not given here, but
it is not too difficult and again not far from its linear algebra counterpart.
Theorem 3.8 (Dimension Formula). Suppose U and W are subspaces of P. Then
dim(hU,Wi) = dim(U) + dim(W) − dim(U ∩ W). 4.
The Projective Plane
For any field F, the projective plane P2 (F) is the set of equivalence classes of nonzero points
in F 3 , where the equivalence relation is given by:
(x, y, z) ∼ (rx, ry, rz)
For any nonzero r ∈ F. Let F2 be the ordinary plane (defined relative to the field F.) There is
an injective map from F 2 into P 2 (F) given by:
(x, y) → [(x, y, 1)],
The equivalence class of the point (x, y, 1). In this way, we think of F 2 as a subset of P 2 (F).
A set S ⊂ F 3 is called a cone if it has the following property: For all v ∈ S and all nonzero r ∈ F,
we have rv ∈ S. Given a cone S, we define the projectivization [S] ⊂ P 2 (F) to be the set of
points [v] such that v ∈ S.
Lines
A line in the projective plane is the set of equivalence classes of points in a 2-
dimensional F-subspace of F 3 . In other words, a line is the set of equivalence classes which
solve the equation ax + by + cz = 0 for some a, b, c ∈ F. That is, a line is the projectivization of a
plane through the origin. The set of lines in P 2 (F) is often known as the dual projective plane.
Think about it: Each line is specified by a triple (a, b, c), where at least one entry is nonzero, and
the two triples (a, b, c) and (ra, rb, rc) give rise to the same lines. Note that P 2 (F) − F 2 is the line
consisting of solutions to z = 0. This particular line is known as the line at infinity and we
sometimes write it as L∞.
Projective Transformations
 A linear isomorphism from F3 to itself respects equivalence classes, and therefore induces a map
from P2 (F) to itself. This map is called a projective transformation. A projective transformation is always
a bijection which maps lines to lines. In case F = R or F = C, the projective transformations are
continuous. The set of projective transformations forms a group, often known as the projective group.

Homogeneous Polynomials

Given a triple I = (a1, a2, a3), we defineX l =x a1 x a1 x a3 .

1 2 3

(1) Here a1, a2, a3 are non-negative integers. We define |I| = a 1 + a2 + a3. We say that a
homogeneous polynomial of degree d (in 3 variables) over the field F is a polynomial of
the form
∑ c I X I , cI ∈ F
|I|=d
(2) The variables here are x1, x2, x3. Sometimes it is convenient to use the variables x, y, z
in place of x1, x2, x3.
Homogenization and Dehomogenization
A degree d polynomial in 3 variables has a homogenization, where we just pad the polynomial
with suitable powers of the third variable to get something that is homogeneous. An example
should suffice to explain this.
x5 + 3x2 y2 + x2 y − 5 =⇒ x5 + 3x2 y2 z + x 2 yz2 − 5z 5 .
 Conversely, every homogeneous polynomial of degree d in 3 variables has a dehomogenization,
obtained by setting the third variable to 1. The operations of homogenization and
dehomogenization are obviously inverses of each other.
Projective and Affine Curves
Let P be a homogeneous polynomial of degree d. If v ∈ F3 and r ∈ F, we have
P(rv) = rdP(v).
Therefore, when r 6= 0, we have P(rv) = 0 if and only if P(v) = 0. In other words, the solution
P = 0 is a cone in F3 . Because of this fact, the following definition makes sense.
VP = {[v]| P(v) = 0} ⊂ P 2(F)}.
This VP is just the projectivization of the solution set P = 0. The set VP is known as a
projective curve. A projective curve is a kind of completion of the solution set to a polynomial.
Suppose that p(x, y) is a degree d polynomial in 2 variables and P(x, y, z) is the homogenization.
Let Vp = {(x, y)| p(x, y) = 0}. The set Vp is known as an affine curve. Since F2 is naturally a
subset of P2 (F), in the way described above, we have the inclusion
Vp ⊂ F 2 ⊂ P 2 (F).
Nonsingular Curves
It makes sense to take the formal partial derivatives of a polynomial over any field. In
particular, the gradient

makes sense. We say that a singular point of P is a point v ≠ 0

such that P(v) = 0 and ∇P(v) = 0 : If r ∈ F is nonzero, then v is a singular point if and only if rv is
a singular point. The polynomial P is called nonsingular if it has no singular points. The
projective curve VP is called nonsingular if P is nonsingular.
When it comes time to discuss elliptic curves, we will always work with nonsingular ones.
The Tangent Line
Let P be a nonsingular projective curve and let [v] ∈ P 2 (F) be a point. The tangent line to P at
[v] is defined to be the line determined by the equation

In case F = R you can think about this geometrically. In R3 ,

the tangent plane to the level set P(x, y, z) = 0 at the point (x0, y0, z0) is given by the equation

Here we are assuming that P(x0, y0, z0) = 0. In

case P is a homogeneous polynomial, we have
 Therefore, in this case, the equation of the tangent plane
simplifies to
(x, y, z) · ∇P = 0
So, in R3 the plane Π0 given is a good approximation along the line through (x0, y0, z0) to the
level set P(x, y, z) = 0. Both sets are cones, and so the projectivization of the tangent plane (the
tangent line) is a good approximation of the projectivization of the polynomial level set (the
projective curve).

A Case of Bezout’s Theorem

Homogeneous Polynomials in Two Variables
A field is F algebraically closed if every polynomial over F has all its roots in F. The results
here work for any algebraically closed field, but for convenience we’ll take F = C, the field of
complex numbers. The Fundamental Theorem of Algebra says that C is algebraically closed.
Example: Let A(x, y) be a homogeneous polynomial of degree n in 2 variables over C. Prove
that A(x, y) factors into linear factors
Here ci , di ∈ C.
Multiplicity
Example above has implications for homogeneous polynomials in 3 variables. If P(x, y, z) is
such a polynomial, we can write

Where B has lower degree. Assuming that A is nontrivial, we can factor A as in Example.
This gives

Bezout’s Theorem
Suppose now that L is an arbitrary line in P 2(C) and that p ∈ L ∩ VP . One possibility is
that L ⊂ VP . Suppose that this doesn’t happen. Then we choose a projective transformation T
such that T(L) = L∞, and we define the multiplicity of p to be the multiplicity of Tp = L∞ ∩
T(VC).
Lemma 3.1 The multiplicity of p is well-defined.
Theorem 3.2 (Bezout) Suppose that P is a homogeneous polynomial of degree n and VP
is the corresponding projective curve. Let L be any line that is not contained in VP . Then L ∩ VP
consists of n points, counting multiplicity. In particular, if VP contains no lines, then every line
intersects VP in n points, counting multiplicity.
The general case of Bezout’s Theorem says that a projective curve of degree d 1 and a
projective curve of degree d2, having no common components, intersect in exactly d1d2 points,
when these points are counted with multiplicity. The proof of this result, as well as a good
definition of multiplicity that works in any algebraically closed field, is harder

Parallel Lines And The Projection Of Infinity

A theorem from Euclid’s Elements (c. 300 BC) states that if a line is drawn through a triangle
such that it is parallel to one side (see the figure), then the line will divide the other two sides
proportionately; that is, the ratio of segments on each side will be equal. This is known as the
proportional segments theorem, or the fundamental theorem of similarity, and for triangle ABC,
shown in the diagram, with line segment DE parallel to side AB, the theorem corresponds to the
mathematical expression
CD/DA = CE/EB.

Now consider the effect produced by projecting these line segments onto another plane as shown
in the figure. The first thing to note is that the projected line segments A′B′ and D′E′ are not
parallel; i.e., angles are not preserved. From the point of view of the projection, the parallel lines
AB and DE appear to converge at the horizon, or at infinity, whose projection in the picture
plane is labeled Ω. (It was Desargues who first introduced a single point at infinity to represent
the projected intersection of parallel lines. Furthermore, he collected all the points along the
horizon in one line at infinity.) With the introduction of Ω, the projected figure corresponds to a
theorem discovered by Menelaus of Alexandria in the 1st century AD:
C′D′/D′A′ = C′E′/E′B′ ∙ ΩB′/ΩA′.
Since the factor ΩB′/ΩA′ corrects for the projective
distortion in lengths, Menelaus’s theorem can be seen
as a projective variant of the proportional segments
theorem.

Projective Invariants
With Desargues’s provision of infinitely distant points for parallels, the reality plane and the
projective plane are essentially interchangeable—that is, ignoring distances and directions
(angles), which are not preserved in the projection. Other properties are preserved, however. For
instance, two different points have a unique connecting line, and two different lines have a
unique point of intersection. Although almost nothing else seems to be invariant under projective
mappings, one should note that lines are mapped onto lines. This means that if three points are
collinear (share a common line), then the
same will be true for their projections.
Thus, collinearity is another invariant
property. Similarly, if three lines meet in a
common point, so will their projections.
The following theorem is of fundamental
importance for projective geometry. In its
first variant, by Pappus of Alexandria (fl.
AD 320) as shown in the figure, it only
uses collinearity:

Let the distinct points  A,  B,  C  and  D,  E,  F  be on two different lines. Then the
three intersection points—
x  of  AE  and  BD,  y  of  AF  and  CD, and  z  of  BF  and  CE
—are collinear.
The second variant, by Pascal, as shown in the figure, uses
certain properties of circles:
If the distinct points A, B, C, D, E, and F are on one circle, then
the three intersection points x, y, and z (defined as above) are
collinear.
There is one more important invariant under projective mappings, known as the cross ratio (see
the figure). Given four distinct collinear points A, B, C, and D, the cross ratio is defined as
CRat(A, B, C, D) = AC/BC ∙ BD/AD.
It may also be written as the quotient of two ratios:
CRat(A, B, C, D) = AC/BC : AD/BD.

The latter formulation reveals the cross ratio as a ratio of ratios of distances. And while neither
distance nor the ratio of distance is preserved under projection, Pappus first proved the startling
fact that the cross ratio was invariant—that is,
CRat(A, B, C, D) = CRat(A′, B′, C′, D′).
However, this result remained a mere curiosity until its real significance became gradually clear
in the 19th century as mappings became more and more important for transforming problems
from one mathematical domain to another.
Projective Conic Sections
Conic sections can be regarded as plane sections of a right
circular cone (see the figure). By regarding a plane
perpendicular to the cone’s axis as the reality plane (RP), a
“cutting” plane as the picture plane (PP), and the cone’s apex as
the projective “eye,” each conic
section can be seen to correspond
to a projective image of a circle
(see the figure). Depending on the
orientation of the cutting plane,
the image of the circle will be a
circle, an ellipse, a parabola, or a hyperbola.
A plane Ω passing through the apex and parallel to PP defines the line at infinity in the
projective plane PP. The situation of Ω relative to RP determines the conic section in PP: If Ω
intersects RP outside the base circle (the circle formed by the intersection of the cone and RP),
the image of the circle will be an ellipse (as shown in the figure). If Ω is tangent to the base
circle (in effect, tangent to the cone), the image will be a parabola. If Ω intersects the base circle
(thus, cutting the circle in two), a hyperbola will result.
Pascal’s theorem, quoted above, also follows easily for any conic section from its special case
for the circle. Start by selecting six points on a conic section and project them back onto the base
circle. As given earlier, the three relevant intersection points for six points on the circle will be
collinear. Now project all nine points back to the conic section. Since collinear points (the three
intersection points from the circle) are mapped onto collinear points, the theorem holds for any
conic section. In this way the projective point of view unites the three different types of conics.
Similarly, more complicated curves and surfaces in higher-dimensional spaces can be unified
through projections. For example, Isaac Newton (1643–1727) showed that all plane curves
defined by polynomials in x and y of degree 3 (the highest power of the variables is 3) can be
obtained as projective images of just five types of polynomials.
Projective Geometry Applied to Computer Vision

Projective geometry is a mathematical framework in which to view computer vision in general,

and especially image formation in particular. The main areas of application are those in which image
formation and/or invariant descriptions between images are important, such as camera calibration,
stereo, ob ject recognition, scene reconstruction, mosaicing, image synthesis, and the analysis of
shadows. This latter application arises from the fact that the composition of two perspective projections
is not necessarily a perspective projection but is definitely a pro jective transformation; that is,
projective transformations form a group, whereas perspective projections do not. Many areas of
computer vision have little to do with projective geometry, such as texture analysis,color,segmentation,
and edge detection. And even in a field such as motion analysis, projective geometry offers little help
when the rigidity assumption is lost because the relationship between projection rays in successive
images cannot be described by such simple and elegant mathematics.

[1] H. S. M. Coxeter. Projective Geometry. Toronto: University of Toronto, second edition, 1974.

[2] O. Faugeras. Three-Dimensional Computer Vision. Cambridge, MA: MIT Press, 1993.

[3] L. Guibas. Lecture notes for CS348a: Computer Graphics { Mathematical Foundations. Stanford
University, Autumn 1996.

[4] Q.-T. Luong and O. D. Faugeras. Fundamental matrix: Theory, algorithms, and stability analysis.
International Journal of Computer Vision, 17(1):43{75, 1996.

[5] J. L. Mundy and A. Zisserman. Geometric Invariance in Computer Vision. Cambridge, MA: MIT Press,
1992.
[6] Z. Zhang and G. Xu. Epipolar Geometry in Stereo, Motion and Object Recognition: Aed Approach.
Kluwer Academic Publishers, 1996.

[7] Benno Artmann,Professor emeritus of mathematics at the University of Goettingen, Goettingen,

Germany. Author of Euclid: The Creation of
Mathematics.https://www.britannica.com/science/projective-geometry/Projective-conic-sections