The Central Idea
Euclidean geometry treats length, angle, and parallelism as fundamental. Projective geometry asks a different question:
What geometric relationships remain unchanged under projection?
Imagine shining light from a point through a geometric figure onto a screen. The image can be stretched, tilted, or compressed, yet certain structural relationships survive.
In particular, projective transformations preserve incidence: if a point lies on a line before the transformation, its image lies on the transformed line afterward.
Projection as a Geometric Operation
Suppose a point source of light is placed at a point called the center of projection. Each point of an object determines a ray from the center. Where that ray meets an image plane, it creates the projected point.
This is the geometry behind perspective drawing, cameras, and many imaging systems.
Projection usually does not preserve Euclidean distance or angle. A square may project to a general quadrilateral, and equal lengths may become unequal.
What Projective Geometry Preserves
The most basic preserved relationships involve incidence.
- Points remain points.
- Lines remain lines, except in degenerate cases outside the transformation domain.
- Collinear points remain collinear.
- Concurrent lines remain concurrent.
- A point lying on a line remains incident with the transformed line.
- Conic sections are carried to conic sections under nonsingular projective transformations.
- The cross-ratio of four collinear points is preserved.
These invariants replace familiar Euclidean measurements as the primary objects of study.
What Projective Geometry Does Not Generally Preserve
A projective transformation does not ordinarily preserve:
- length;
- distance;
- angle;
- area;
- midpoints;
- ratios of ordinary lengths;
- parallelism.
This is why a perspective image may look highly distorted while still retaining a precise projective structure.
Parallel Lines and the Projective View
In the Euclidean plane, distinct parallel lines never meet. Projective geometry modifies the plane by adjoining additional points called points at infinity or ideal points.
Every family of Euclidean parallel lines with the same direction is assigned one common ideal point.
Thus, in the projective plane:
Any two distinct projective lines meet in exactly one point.
For formerly parallel Euclidean lines, that point is at infinity.
Why Points at Infinity Are Useful
Points at infinity eliminate the special-case distinction between intersecting and parallel lines.
Instead of saying
- nonparallel lines meet, but
- parallel lines do not,
projective geometry uses one unified rule:
Every pair of distinct lines meets.
This simplification is one of the main structural advantages of the projective viewpoint.
The Line at Infinity
The ideal points corresponding to all possible directions in the Euclidean plane form a single projective line called the line at infinity.
Adding this line to the Euclidean plane produces the real projective plane, denoted
\[ \mathbb{RP}^2. \]Ordinary Euclidean points form an affine part of the projective plane, while the line at infinity supplies the missing directional points.
Perspective and Vanishing Points
In a perspective drawing, parallel railway tracks appear to approach one another and meet at a vanishing point.
Euclidean geometry interprets this as an optical effect. Projective geometry treats the vanishing point as the image of the common ideal point associated with that direction.
Different families of parallel lines generally have different vanishing points.
The Horizon as a Projective Object
In many perspective images, the ideal points of directions parallel to a ground plane appear along a visible horizon line.
The horizon is therefore closely related to the image of the ground plane's line at infinity.
This connection is central in perspective drawing, photogrammetry, computer vision, and camera calibration.
Homogeneous Coordinates
Projective geometry is especially convenient when points are represented by homogeneous coordinates.
A projective point is represented by a nonzero triple
\[ (X:Y:Z), \]where multiplying every coordinate by the same nonzero scalar does not change the point:
\[ (X:Y:Z)=(\lambda X:\lambda Y:\lambda Z), \qquad \lambda\ne 0. \]The coordinates therefore describe a ratio rather than one unique ordered triple.
From Cartesian to Homogeneous Coordinates
An ordinary Cartesian point
\[ (x,y) \]can be embedded into the projective plane as
\[ \boxed{(x:y:1)}. \]For example,
\[ (3,-2)\longleftrightarrow(3:-2:1). \]Because homogeneous coordinates are defined only up to nonzero scale,
\[ (3:-2:1)=(6:-4:2)=(-3:2:-1). \]Returning to Cartesian Coordinates
If a projective point has
\[ Z\ne 0, \]divide by the third coordinate to obtain the representative
\[ \left(\frac{X}{Z}:\frac{Y}{Z}:1\right). \]Its ordinary Cartesian coordinates are therefore
\[ \boxed{x=\frac{X}{Z},\qquad y=\frac{Y}{Z}}. \]Points at Infinity in Homogeneous Coordinates
Points at infinity are exactly the projective points with
\[ Z=0. \]Thus a point at infinity has the form
\[ (X:Y:0), \]with \\(X\\) and \\(Y\\) not both zero.
These coordinates encode direction rather than an ordinary finite location.
Example: Direction as a Point at Infinity
Consider lines with direction vector
\[ (2,1). \]Their common ideal point is represented by
\[ \boxed{(2:1:0)}. \]Any Euclidean line parallel to this direction reaches the same projective point at infinity.
Lines in Homogeneous Coordinates
A projective line can be written
\[ aX+bY+cZ=0, \]where \\(a,b,c\\) are not all zero.
As with point coordinates, multiplying all three line coefficients by a common nonzero scalar gives the same line.
When \\(Z=1\\), this becomes the familiar affine equation
\[ ax+by+c=0. \]The Line at Infinity in Coordinates
The line at infinity consists of all points satisfying
\[ \boxed{Z=0}. \]In the line equation
\[ aX+bY+cZ=0, \]the line at infinity is therefore represented by
\[ (a,b,c)=(0,0,1). \]Incidence as an Algebraic Condition
Let a point be represented by
\[ P=(X:Y:Z) \]and a line by coefficients
\[ \ell=(a:b:c). \]The point lies on the line exactly when
\[ \boxed{aX+bY+cZ=0}. \]This compact equation expresses the central incidence relation of projective geometry.
The Cross Product View
Homogeneous coordinates allow joins and intersections to be computed using three-dimensional cross products.
If \\(P\\) and \\(Q\\) are distinct projective points, the line through them is
\[ \boxed{\ell=P\times Q}. \]If \\(\ell\\) and \\(m\\) are distinct projective lines, their intersection is
\[ \boxed{P=\ell\times m}. \]The same algebraic operation therefore handles both joining points and intersecting lines.
Example: Line Through Two Points
Let
\[ P=(1:2:1),\qquad Q=(4:3:1). \]Their cross product is
\[ P\times Q= \begin{vmatrix} \mathbf{i}&\mathbf{j}&\mathbf{k}\\ 1&2&1\\ 4&3&1 \end{vmatrix} =(-1,3,-5). \]Therefore the projective line has equation
\[ -X+3Y-5Z=0. \]In affine coordinates \\(Z=1\\), this is
\[ \boxed{-x+3y-5=0}. \]Example: Intersection of Parallel Euclidean Lines
Consider
\[ y=2x+1 \]and
\[ y=2x-3. \]Their homogeneous equations are
\[ 2X-Y+Z=0 \]and
\[ 2X-Y-3Z=0. \]Subtracting the equations gives \\(4Z=0\\), so \\(Z=0\\). Then
\[ 2X-Y=0. \]Their projective intersection is therefore
\[ \boxed{(1:2:0)}. \]The Euclidean lines are parallel, but the projective lines meet at the ideal point representing slope \\(2\\).
Projective Transformations
A projective transformation of the projective plane can be represented by an invertible \\(3\times3\\) matrix.
If
\[ \mathbf{x}=\begin{bmatrix}X\\Y\\Z\end{bmatrix}, \]then a projective transformation has the form
\[ \boxed{\mathbf{x}'=H\mathbf{x}}, \]where \\(H\\) is nonsingular.
Since homogeneous coordinates are defined up to scale, multiplying \\(H\\) itself by a nonzero scalar produces the same projective transformation.
Homographies
A projective transformation represented by a nonsingular matrix is often called a homography, especially in computer vision and imaging.
Homographies map projective lines to projective lines and preserve incidence. They can represent perspective distortion between two views of the same plane.
Affine Transformations Inside Projective Geometry
Affine transformations—translations, rotations, shears, and nonsingular scalings—form a special subclass of projective transformations.
An affine transformation preserves the line at infinity. Consequently, it preserves parallelism in the ordinary Euclidean part of the plane.
A general projective transformation need not preserve the line at infinity, so parallel lines may become intersecting finite lines.
Euclidean Geometry Inside Affine Geometry
Euclidean transformations are more restrictive still. They preserve distances and angles as well as incidence and parallelism.
This creates a useful hierarchy:
Euclidean geometry preserves the most metric structure, affine geometry preserves parallel and linear structure, and projective geometry preserves the more fundamental incidence structure.
A Hierarchy of Geometric Invariants
Moving from Euclidean to affine to projective geometry allows progressively more transformations.
- Euclidean transformations preserve distance and angle.
- Similarity transformations preserve angle and ratios of corresponding lengths.
- Affine transformations preserve parallelism and ratios along a common line.
- Projective transformations preserve incidence and cross-ratio.
The larger the transformation group, the fewer properties remain invariant.
Projective Geometry and Conic Sections
Circles, ellipses, parabolas, and hyperbolas look different in Euclidean geometry because metric properties distinguish them.
Projectively, however, all nondegenerate real conics are equivalent: a suitable projective transformation can carry one nondegenerate conic to another.
This is one reason projective geometry provides a natural unifying language for conic sections.
A General Conic Equation
In homogeneous coordinates, a conic can be written as a homogeneous quadratic equation
\[ aX^2+bXY+cY^2+dXZ+eYZ+fZ^2=0. \]Every term has total degree \\(2\\), which makes the equation compatible with homogeneous scaling.
Why Homogeneity Matters
If a projective point is represented by \\(\mathbf{x}\\), then \\(\lambda\mathbf{x}\\) represents the same point for every nonzero \\(\lambda\\).
A homogeneous polynomial of degree \\(d\\) satisfies
\[ F(\lambda\mathbf{x})=\lambda^dF(\mathbf{x}). \]Therefore
\[ F(\mathbf{x})=0 \]is independent of which nonzero scalar representative is chosen.
The Cross-Ratio
Ordinary ratios of distances are generally not preserved by projective transformations. A more subtle quantity called the cross-ratio is preserved.
For four distinct collinear points with suitable affine coordinates \\(a,b,c,d\\), one common convention is
\[ \boxed{(A,B;C,D)= \frac{(c-a)(d-b)}{(c-b)(d-a)}}. \]Different texts may arrange the four points differently, producing one of several algebraically related values. The ordering must therefore be stated.
Example: Computing a Cross-Ratio
Let four points on a line have coordinates
\[ a=0,\qquad b=2,\qquad c=3,\qquad d=6. \]Using
\[ (A,B;C,D)=\frac{(c-a)(d-b)}{(c-b)(d-a)}, \]we obtain
\[ \frac{(3-0)(6-2)}{(3-2)(6-0)} =\frac{12}{6} =2. \]Thus
\[ \boxed{(A,B;C,D)=2}. \]Why the Cross-Ratio Matters
Perspective projection can change ordinary spacing along a line. Points that were equally spaced need not remain equally spaced in an image.
The cross-ratio survives this distortion. It provides a numerical invariant even when ordinary Euclidean distances are unreliable.
This makes the cross-ratio important in classical projective geometry, reconstruction, and computer vision.
Harmonic Division
A particularly important projective configuration occurs when four collinear points have cross-ratio \\(-1\\), under a standard ordering convention.
Such a quadruple is called a harmonic range.
\[ (A,B;C,D)=-1. \]Harmonic configurations appear naturally in classical constructions and provide an especially symmetric projective relationship.
Duality
One of the most striking features of plane projective geometry is duality.
Many valid statements remain valid when the roles of points and lines are exchanged, together with the corresponding incidence language.
For example:
- Two distinct points determine one line.
- Two distinct lines determine one point of intersection.
These statements are dual versions of one another.
Point-Line Symmetry
In homogeneous coordinates, a point may be represented by a triple
\[ (X:Y:Z), \]while a line is represented by a triple
\[ (a:b:c). \]Their incidence condition
\[ aX+bY+cZ=0 \]is symmetric enough to make the duality between points and lines visible algebraically.
Desargues' Theorem
A classical theorem states, roughly, that if two triangles are perspective from a point, then corresponding sides meet in three collinear points.
More precisely, suppose the lines joining corresponding vertices of two triangles are concurrent. Then, under the usual nondegeneracy assumptions, the three intersection points of corresponding sides are collinear.
Desargues' theorem illustrates how projective geometry organizes incidence relationships independently of metric measurement.
Perspective from a Point
Two triangles are said to be perspective from a point when the lines connecting corresponding vertices pass through one common point.
If
\[ AA',\qquad BB',\qquad CC' \]are concurrent, the triangles \\(ABC\\) and \\(A'B'C'\\) are perspective from that concurrency point.
Perspective from a Line
The dual configuration asks whether intersections of corresponding sides lie on one common line.
If
\[ AB\cap A'B',\qquad BC\cap B'C',\qquad CA\cap C'A' \]are collinear, the two triangles are perspective from a line.
Desargues' theorem connects these two forms of perspectivity.
Pappus' Theorem
Another foundational incidence theorem is Pappus' theorem. In one standard form, begin with three points on one line and three points on another line. Construct intersections from certain pairs of connecting lines. The three resulting intersection points are collinear.
The theorem contains no distances or angles. Its content is entirely about points, lines, intersections, and collinearity—the natural language of projective geometry.
Projective Geometry Is Not Merely Perspective Drawing
Perspective drawing is an important motivation, but projective geometry is a mathematical theory in its own right.
It studies abstract projective spaces, incidence structures, homogeneous coordinates, transformations, conics, duality, and invariants such as the cross-ratio.
Artistic perspective is one concrete application of this broader structure.
Projective Space in Higher Dimensions
The same idea extends beyond the projective plane.
Real projective \\(n\\)-space is denoted
\[ \mathbb{RP}^n. \]Its points are equivalence classes of nonzero vectors in \\(\mathbb{R}^{n+1}\\), where vectors differing by a nonzero scalar are identified.
Thus
\[ (X_0:\cdots:X_n) \]represents a projective point.
A Projective Point as a Line Through the Origin
Homogeneous coordinates have a useful geometric interpretation.
Every nonzero vector
\[ (X,Y,Z)\in\mathbb{R}^3 \]lies on a unique line through the origin. All nonzero scalar multiples of that vector lie on the same line.
Therefore, a point of \\(\mathbb{RP}^2\\) can be viewed as a line through the origin in \\(\mathbb{R}^3\\).
Why \\(P\\) and \\(-P\\) Represent the Same Projective Point
Since homogeneous coordinates identify all nonzero scalar multiples,
\[ (X:Y:Z)=(-X:-Y:-Z). \]This identification distinguishes the projective plane from an ordinary sphere of directions. Opposite vectors determine the same line through the origin and therefore the same projective point.
The Projective Plane Is Not an Ordinary Plane with a Distant Border
The phrase point at infinity can be misleading if interpreted too literally.
The line at infinity is not a physical boundary located infinitely far away. It is part of the mathematical completion that allows directions to be represented as genuine projective points.
Under a projective transformation, a point that is finite in one affine coordinate chart can move to infinity, while a formerly ideal point can become finite.
Affine Charts
To recover familiar coordinates from projective space, choose a region where one homogeneous coordinate is nonzero.
The standard affine chart
\[ Z\ne0 \]is normalized by setting \\(Z=1\\). This produces ordinary coordinates
\[ (x,y)=\left(\frac{X}{Z},\frac{Y}{Z}\right). \]Other affine charts can be obtained by choosing \\(X\ne0\\) or \\(Y\ne0\\) instead.
No Single Affine Chart Covers the Whole Projective Plane
The chart \\(Z\ne0\\) excludes the line \\(Z=0\\). Similarly, other charts omit other projective lines.
Several overlapping coordinate charts are needed to describe the entire projective plane without singularities in the coordinates.
This idea anticipates the broader use of coordinate charts in geometry and manifold theory.
One-Dimensional Projective Geometry
The real projective line is denoted
\[ \mathbb{RP}^1. \]Its points are homogeneous pairs
\[ (X:Z), \]not both zero, modulo nonzero scalar multiplication.
Ordinary real numbers appear as
\[ (x:1), \]and one additional point
\[ (1:0) \]acts as the point at infinity.
Fractional Linear Transformations
On the projective line, projective transformations take the form
\[ \boxed{x' = \frac{ax+b}{cx+d}}, \]where
\[ ad-bc\ne0. \]Such transformations are called fractional linear, linear fractional, or Möbius transformations, depending on context and the underlying number system.
Why a Denominator Can Vanish
If
\[ cx+d=0, \]the affine formula appears undefined. Projectively, the corresponding point is sent to the point at infinity.
The homogeneous formulation removes the need to treat this as a mysterious failure of the transformation.
Matrix Form on the Projective Line
Let
\[ \begin{bmatrix}X'\\Z'\end{bmatrix} = \begin{bmatrix}a&b\\c&d\end{bmatrix} \begin{bmatrix}X\\Z\end{bmatrix}. \]If \\(Z\ne0\\) and \\(Z'\ne0\\), then with \\(x=X/Z\\),
\[ x'=\frac{X'}{Z'} =\frac{ax+b}{cx+d}. \]The fractional formula is therefore simply the affine form of a linear map acting on homogeneous coordinates.
Composition of Projective Transformations
If one projective transformation is represented by \\(H_1\\) and another by \\(H_2\\), applying them successively gives
\[ \mathbf{x}'=H_2H_1\mathbf{x}. \]Thus composition corresponds to matrix multiplication.
Because each transformation matrix is nonsingular, every projective transformation has an inverse represented by the inverse matrix.
Projective Transformations Form a Group
Projective transformations can be composed, have inverses, possess an identity transformation, and satisfy associativity under composition.
In the plane, the projective linear group is commonly written
\[ \operatorname{PGL}(3,\mathbb{R}). \]Matrices differing only by a nonzero scalar represent the same projective transformation.
Degrees of Freedom of a Planar Homography
A general \\(3\times3\\) matrix has nine entries. Because an overall nonzero scalar does not change the represented projective transformation, only eight independent ratios matter.
A planar homography therefore has eight projective degrees of freedom.
This observation helps explain why four point correspondences in general position can determine a planar homography: each correspondence supplies two independent coordinate constraints.
Four Points in General Position
In the projective plane, four points are in general position when no three are collinear.
Given two ordered sets of four points in general position, there is a unique projective transformation carrying one ordered quadruple to the other.
This is a powerful expression of how flexible projective transformations are.
Why a Square Can Become a General Quadrilateral
A square is characterized in Euclidean geometry by equal sides and right angles. Neither property is projectively invariant.
Under a projective transformation, the four vertices can map to the vertices of a general quadrilateral, provided the configuration remains nondegenerate.
What survives is the incidence pattern of vertices and sides, not the square's metric regularity.
Perspective Correction
Suppose a rectangular sign is photographed from an oblique angle. In the image it may appear as a trapezoid or general quadrilateral.
If the four image corners correspond to the four corners of the original rectangle, a homography can often be used to rectify the image so the sign appears front-facing.
This is a practical application of projective geometry in document scanning, image processing, and computer vision.
Camera Geometry
An ideal pinhole camera performs a central projection from three-dimensional space onto a two-dimensional image plane.
Homogeneous coordinates make this projection particularly natural because depth-dependent scale can be absorbed into the projective coordinate factor.
For this reason, projective geometry is foundational in modern multi-view geometry and 3D reconstruction.
Projective Geometry in Computer Graphics
Three-dimensional graphics pipelines commonly represent points using homogeneous coordinates such as
\[ (X,Y,Z,W). \]Translation, rotation, scaling, and perspective projection can then be organized through matrix multiplication in a unified coordinate framework.
The final perspective division by \\(W\\) converts back to ordinary display coordinates.
Projective Geometry in Architecture and Art
Renaissance perspective systems formalized how parallel spatial directions appear to converge in a planar picture.
Projective concepts help describe:
- one-point perspective;
- two-point perspective;
- three-point perspective;
- vanishing points;
- horizon lines;
- perspective grids;
- reconstruction from drawings and photographs.
Projective Geometry and Measurement
Projective geometry deliberately gives up ordinary metric measurement, but metric information can sometimes be recovered when additional structure is known.
For example, camera calibration, known parallel directions, known right angles, or known reference lengths can supply constraints that allow a projectively distorted image to be rectified into an affine or Euclidean interpretation.
This layered recovery of structure is important in computer vision.
Projective Completion
The process of adjoining ideal elements so that formerly exceptional cases become ordinary ones is called projective completion.
Completing the Euclidean or affine plane by adding the line at infinity turns parallel line pairs into ordinary intersecting projective lines.
This is a recurring mathematical strategy: enlarge a space so that its rules become more uniform.
Degenerate Configurations Still Matter
Projective formulas often remain meaningful when a Euclidean configuration approaches a limiting or apparently exceptional case.
For example, as an intersection point of two lines moves farther and farther away while the lines become nearly parallel, projective geometry naturally interprets the limiting intersection as a point at infinity.
This ability to absorb limiting cases is one of the theory's major strengths.
Projective Geometry Versus Spherical Geometry
Spherical geometry changes Euclidean geometry by changing the space and its metric behavior. Distances, geodesics, and angle sums are central.
Projective geometry changes emphasis in a different way. It studies incidence and projection while largely discarding metric quantities such as length and angle.
Both are alternatives to a purely Euclidean viewpoint, but they alter different parts of the geometric structure.
Projective Geometry Versus Hyperbolic Geometry
Hyperbolic geometry has an intrinsic metric, negative curvature, and a characteristic parallel structure. Projective geometry does not use the same metric assumptions and treats every pair of distinct projective lines as intersecting.
Nevertheless, projective methods can be used to construct and study models of hyperbolic geometry. This illustrates how one geometric framework can provide tools for another.
Projective Geometry Versus Affine Geometry
Affine geometry preserves parallelism, collinearity, and ratios of lengths measured along the same line.
Projective geometry preserves collinearity and incidence but allows the line at infinity to move. Consequently, parallelism and ordinary affine ratios need not survive.
The projective category is therefore broader than the affine category.
Common Error: Thinking Points at Infinity Are Infinitely Far Cartesian Points
A point at infinity is not an ordinary Cartesian point with enormously large coordinates.
It is a projective point represented by
\[ (X:Y:0). \]It encodes a direction in a chosen affine chart.
Common Error: Assuming Parallelism Is Projectively Preserved
General projective transformations do not preserve Euclidean parallelism.
A family of parallel lines may transform into a family of lines meeting at a finite point because their common ideal point can be mapped to an ordinary point.
Parallelism is preserved by affine transformations, not by every projective transformation.
Common Error: Treating Homogeneous Coordinates as Unique
The triples
\[ (1:2:3),\qquad(2:4:6),\qquad(-1:-2:-3) \]all represent the same projective point.
Homogeneous coordinates are equivalence classes under multiplication by a nonzero scalar.
Common Error: Dividing by \\(Z\\) When \\(Z=0\\)
Cartesian recovery formulas
\[ x=\frac{X}{Z},\qquad y=\frac{Y}{Z} \]apply only when \\(Z\ne0\\).
If \\(Z=0\\), the point is on the line at infinity in that affine chart and has no finite Cartesian coordinate pair there.
Common Error: Assuming Projective Transformations Preserve Shape
Projective transformations preserve line structure and incidence, not Euclidean shape.
A rectangle can become a nonrectangular quadrilateral. A circle can become another conic. Angles and side lengths can change.
The correct question is not whether the figure looks the same, but which structural relations remain invariant.
Common Error: Using Ordinary Distance Ratios as Projective Invariants
A perspective transformation can change ordinary ratios of distances along a line.
The projectively preserved quantity for four collinear points is the cross-ratio, not a single ordinary length ratio.
Common Error: Confusing Projective and Topological Equivalence
Projective transformations are not arbitrary continuous deformations. They are highly structured transformations induced by invertible linear maps in homogeneous coordinates.
Topology preserves different properties and permits a much broader class of deformations.
A Reliable Procedure for Converting to Homogeneous Coordinates
- Begin with the affine point \\((x,y)\\).
- Append a homogeneous coordinate equal to \\(1\\).
- Write the projective point as \\((x:y:1)\\).
- Remember that every nonzero scalar multiple represents the same point.
A Reliable Procedure for Returning to Affine Coordinates
- Start with \\((X:Y:Z)\\).
- Check whether \\(Z\ne0\\).
- If so, divide all coordinates by \\(Z\\).
- Read the affine coordinates as \\((X/Z,Y/Z)\\).
- If \\(Z=0\\), recognize that the point lies at infinity in this chart.
A Reliable Procedure for Finding the Line Through Two Points
- Write both points as homogeneous three-component vectors.
- Compute their cross product.
- Interpret the resulting triple \\((a,b,c)\\) as line coefficients.
- Write the line as \\(aX+bY+cZ=0\\).
- Substitute both original points to verify incidence.
A Reliable Procedure for Intersecting Two Lines
- Write each line as a homogeneous coefficient vector.
- Compute the cross product of the two line vectors.
- The resulting vector represents their projective intersection point.
- If its third coordinate is nonzero, convert to affine coordinates.
- If its third coordinate is zero, the lines meet at infinity in that chart.
A Reliable Procedure for Applying a Homography
- Convert the point to homogeneous coordinates.
- Multiply by the nonsingular homography matrix \\(H\\).
- Keep the resulting vector in homogeneous form while doing projective calculations.
- If finite affine coordinates are needed, divide by the final homogeneous coordinate.
- Check whether that final coordinate is zero before dividing.
Example: Applying a Projective Matrix
Consider
\[ H= \begin{bmatrix} 1&0&2\\ 0&1&1\\ 0&0&1 \end{bmatrix} \]and the affine point \\((3,4)\\), represented by
\[ \mathbf{x}=\begin{bmatrix}3\\4\\1\end{bmatrix}. \]Then
\[ H\mathbf{x} = \begin{bmatrix} 5\\5\\1 \end{bmatrix}. \]The image point is therefore
\[ \boxed{(5,5)}. \]In this example the projective matrix happens to represent an affine translation.
Example: A Genuine Perspective Effect
Consider
\[ H= \begin{bmatrix} 1&0&0\\ 0&1&0\\ 1&0&1 \end{bmatrix}. \]For \\((x,y)\\), use \\((x,y,1)^T\\):
\[ H \begin{bmatrix}x\\y\\1\end{bmatrix} = \begin{bmatrix}x\\y\\x+1\end{bmatrix}. \]When \\(x\ne-1\\), the affine image is
\[ \boxed{ x'=\frac{x}{x+1},\qquad y'=\frac{y}{x+1} }. \]The denominator depends on position, which creates the nonlinear-looking perspective distortion in affine coordinates.
Example: A Finite Point Sent to Infinity
Using the previous transformation, points with
\[ x=-1 \]produce transformed homogeneous coordinates with third component \\(0\\):
\[ (-1,y,1) \longmapsto (-1,y,0). \]Thus the entire affine line \\(x=-1\\) is mapped to the line at infinity.
This example demonstrates that being “at infinity” depends on the chosen affine chart rather than representing an absolute distant region.
Example: Detecting Collinearity with a Determinant
Three projective points
\[ P=(X_1:Y_1:Z_1),\quad Q=(X_2:Y_2:Z_2),\quad R=(X_3:Y_3:Z_3) \]are collinear exactly when
\[ \boxed{ \det \begin{bmatrix} X_1&Y_1&Z_1\\ X_2&Y_2&Z_2\\ X_3&Y_3&Z_3 \end{bmatrix}=0 }. \]This determinant criterion is an algebraic form of an incidence statement.
Example: Checking Three Affine Points
Consider
\[ (0,1),\qquad(2,5),\qquad(3,7). \]In homogeneous form, test
\[ \det \begin{bmatrix} 0&1&1\\ 2&5&1\\ 3&7&1 \end{bmatrix}. \]The determinant is
\[ 0(5-7)-1(2-3)+1(14-15)=1-1=0. \]Therefore the points are collinear.
Projective Thinking as Structural Thinking
Projective geometry teaches a broad mathematical lesson: which properties matter depends on which transformations are allowed.
If transformations are permitted to distort length and angle, then those quantities cannot serve as invariants. More structural properties—incidence, collinearity, concurrency, and cross-ratio—take their place.
Geometry is therefore not only the study of shapes. It is also the study of properties that remain stable under chosen families of transformations.
Practice 1: Homogeneous Coordinates
Convert the affine point
\[ (-4,7) \]to homogeneous coordinates.
Practice 2: Equivalent Representatives
Determine whether
\[ (2:-6:4) \]and
\[ (-1:3:-2) \]represent the same projective point.
Practice 3: Return to Affine Coordinates
Convert
\[ (12:-8:4) \]to affine coordinates.
Practice 4: Identify a Point at Infinity
Which of the following represents a point at infinity in the standard \\(Z=1\\) affine chart?
\[ A=(2:5:1),\qquad B=(4:-3:0),\qquad C=(0:0:7). \]Practice 5: Find a Line
Find the projective line through
\[ P=(1:0:1),\qquad Q=(0:1:1). \]Practice 6: Parallel Lines
Find the projective intersection of the affine lines
\[ y=3x+2 \]and
\[ y=3x-5. \]Practice 7: Cross-Ratio
Using
\[ (A,B;C,D)=\frac{(c-a)(d-b)}{(c-b)(d-a)}, \]compute the cross-ratio for
\[ a=0,\qquad b=1,\qquad c=2,\qquad d=4. \]Practice 8: Apply a Homography
Let
\[ H= \begin{bmatrix} 2&0&0\\ 0&3&0\\ 0&0&1 \end{bmatrix}. \]Find the affine image of \\((2,-1)\\).
Practice 9: Preserved or Not?
For each property, state whether a general projective transformation necessarily preserves it:
- collinearity;
- distance;
- incidence;
- parallelism;
- cross-ratio.
Practice 10: Explain the Geometry
Two railway rails are physically parallel but appear to meet in a photograph. Explain the phenomenon using projective geometry rather than saying the lines “really intersect” in Euclidean space.
Practice Answers
Practice 1
Append the homogeneous coordinate \\(1\\):
\[ \boxed{(-4:7:1)}. \]Practice 2
Multiply \\((-1:3:-2)\\) by \\(-2\\):
\[ (-2)(-1:3:-2)=(2:-6:4). \]Therefore the two triples represent the same projective point.
Practice 3
Divide by \\(Z=4\\):
\[ (12:-8:4)=(3:-2:1). \]Therefore the affine coordinates are
\[ \boxed{(3,-2)}. \]Practice 4
A point at infinity has \\(Z=0\\). Therefore
\[ \boxed{B=(4:-3:0)}. \]Practice 5
Compute
\[ P\times Q = \begin{vmatrix} \mathbf{i}&\mathbf{j}&\mathbf{k}\\ 1&0&1\\ 0&1&1 \end{vmatrix} =(-1,-1,1). \]Therefore the line is
\[ -X-Y+Z=0, \]or equivalently
\[ \boxed{X+Y-Z=0}. \]Practice 6
The two lines have the same direction vector \\((1,3)\\), so they meet at the ideal point
\[ \boxed{(1:3:0)}. \]Practice 7
\[ (A,B;C,D) =\frac{(2-0)(4-1)}{(2-1)(4-0)} =\frac{6}{4} =\boxed{\frac{3}{2}}. \]Practice 8
Write \\((2,-1)\\) as \\((2,-1,1)^T\\):
\[ H \begin{bmatrix}2\\-1\\1\end{bmatrix} = \begin{bmatrix}4\\-3\\1\end{bmatrix}. \]Thus the affine image is
\[ \boxed{(4,-3)}. \]Practice 9
- Collinearity: preserved.
- Distance: not generally preserved.
- Incidence: preserved.
- Parallelism: not generally preserved.
- Cross-ratio: preserved for four collinear points with fixed ordering.
Practice 10
The physical rails belong to one Euclidean family of parallel lines and therefore share an ideal point. Perspective projection maps that ideal point to a vanishing point in the image. The rails remain parallel in the physical Euclidean model while their projected image lines meet at the corresponding vanishing point.
What to Remember
- Projective geometry emphasizes incidence rather than distance and angle.
- Collinearity and concurrency are projective invariants.
- General projective transformations do not preserve distance, angle, area, or parallelism.
- The affine plane is completed projectively by adding a line at infinity.
- Parallel Euclidean lines meet at a common ideal point in the projective plane.
- Homogeneous coordinates represent a point by \\((X:Y:Z)\\) up to nonzero scale.
- Finite affine points can be represented as \\((x:y:1)\\).
- Points with \\(Z=0\\) lie at infinity in the standard affine chart.
- A projective line has equation \\(aX+bY+cZ=0\\).
- The line through two projective points can be found with a cross product.
- The intersection of two projective lines can also be found with a cross product.
- A planar projective transformation is represented by a nonsingular \\(3\times3\\) matrix up to nonzero scale.
- Planar homographies preserve lines and incidence.
- Affine and Euclidean transformations are special cases within projective geometry.
- Nondegenerate conics are projectively related.
- The cross-ratio is a central numerical invariant of four collinear points.
- Point-line duality is a fundamental structural symmetry of the projective plane.
- Perspective drawing, camera geometry, graphics, and computer vision are major applications.
- “Infinity” in projective geometry is a structural completion, not a distant Cartesian boundary.
Where We Go Next
Projective geometry showed that a geometry can be organized around incidence and transformation rather than ordinary Euclidean measurement. The next lesson changes the metric itself.
In Taxicab and Alternative Metrics, we will examine what happens when the rule for distance is changed. Familiar ideas such as circles, shortest paths, and equidistance can then take surprisingly different forms, even when the underlying set of points remains the ordinary plane.