Image Processing and Computer Graphics

Transformations and
Homogeneous Coordinates
Matthias Teschner

Computer Science Department

University of Freiburg
Motivation
 transformations are used
 to position, reshape, and animate
objects, lights, and the virtual camera
 to orthographically or perspectivly project
three-dimensional geometry onto a plane
 transformations are represented with 4x4 matrices
 transformations are applied to vertices and normals
 vertices (positions) and normals (directions) are
represented with 4D vectors

University of Freiburg – Computer Science Department – Computer Graphics - 2

Outline
 transformations in the rendering pipeline
 motivations for the homogeneous notation
 homogeneous notation
 basic transformations in homogeneous notation
 compositing transformations
 summary

University of Freiburg – Computer Science Department – Computer Graphics - 3

Vertex Processing
 modelview transform
 (lighting)
 projection transform
 (clipping)
 viewport transform

University of Freiburg – Computer Science Department – Computer Graphics - 4

 Modelview Transform

Inverse
M1
M3 M2
V-1

V M4
[Akenine-Moeller et al.: Real-time Rendering]

 M1, M2, M3, M4, V are matrices representing transformations

 M1, M2, M3, M4 are model transforms to place the objects in the scene
 V places and orientates the camera in space
 V-1 transforms the camera to the origin looking along the negative z-axis
 model and view transforms are combined in the modelview transform
 the modelview transform V-1M1..4 is applied to the objects
University of Freiburg – Computer Science Department – Computer Graphics - 5
Projection Transform
 P transforms the view volume to the canonical view volume
 the view volume depends on the camera properties
 orthographic projection  cuboid
 perspective projection  pyramidal frustum

Porthographic Pperspective
[Song Ho Ahn]
 canonical view volume is a cube from (-1,-1,-1) to (1,1,1)
 view volume is specified by near, far, left, right, bottom, top
University of Freiburg – Computer Science Department – Computer Graphics - 6
Viewport Transform /
Screen Mapping
 projected primitive coordinates (xp, yp, zp) are
transformed to screen coordinates (xs, ys)
 screen coordinates together with depth value are
window coordinates (xs, ys, zw)

[Akenine-Moeller et al.: Real-time Rendering]

University of Freiburg – Computer Science Department – Computer Graphics - 7
Vertex Transforms
object space

 modelview transform
eye space / camera space

 projection transform
normalized
device coordinates
 viewport transform
window space

University of Freiburg – Computer Science Department – Computer Graphics - 8

Vertex Transforms
local object space
 model transform
world coordinate space
 inverse view transform
eye space / camera space
 projection transform
normalized
device coordinates
 viewport transform
window space
University of Freiburg – Computer Science Department – Computer Graphics - 9
Outline
 transformations in the rendering pipeline
 motivations for the homogeneous notation
 homogeneous notation
 basic transformations in homogeneous notation
 compositing transformations
 summary

University of Freiburg – Computer Science Department – Computer Graphics - 10

Some Transformations
 congruent transformations
(Euclidean transformations)
 preserve shape and size
 translation, rotation, reflection
 similarity transformations
 preserve shape
 translation, rotation, reflection, scale

University of Freiburg – Computer Science Department – Computer Graphics - 11

Affine Transformations
 preserve collinearity
 points on a line are transformed to points on a line
 preserve proportions
 ratios of distances between points are preserved
 preserve parallelism
 parallel lines are transformed to parallel lines
 angles and lengths are not preserved
 translation, rotation, reflection, scale, shear are affine
 orthographic projection is a combination of affine transf.
 perspective projection is not affine

University of Freiburg – Computer Science Department – Computer Graphics - 12

Affine Transformations
 affine transformations of a 3D point p:
 affine transformations preserve affine combinations
for
 e.g., a line can be transformed
by transforming its control points

University of Freiburg – Computer Science Department – Computer Graphics - 13

Affine Transformations
 affine transformations of a 3D point p

 the 3x3 matrix A represents scale and rotation

 the 3D vector t represents translation
 using homogeneous coordinates,
all affine transformations are represented
with one matrix-vector multiplication

University of Freiburg – Computer Science Department – Computer Graphics - 14

Points and Vectors
 the rendering pipeline transforms vertices,
normals, colors, texture coordinates
 points (e.g. vertices) specify a location in space
 vectors (e.g. normals) specify a direction
 relations between points and vectors
 point - point = vector
 point + vector = point
 vector + vector = vector
 point + point = not defined


University of Freiburg – Computer Science Department – Computer Graphics - 15

Points and Vectors
 transformations can have different
effects on points and vectors
 translation
 translation of a point moves the point to a different position
 translation of a vector does not change the vector
 using homogeneous coordinates,
transformations of vectors and points
can be handled in a unified way

University of Freiburg – Computer Science Department – Computer Graphics - 16

Outline
 transformations in the rendering pipeline
 motivations for the homogeneous notation
 homogeneous notation
 basic transformations in homogeneous notation
 compositing transformations
 summary

University of Freiburg – Computer Science Department – Computer Graphics - 17

Homogeneous Coordinates of Points
 with w ≠ 0 are the homogeneous
coordinates of the 3D point
 represents the same point
for all  with  ≠ 0
 examples
 (2, 3, 4, 1)  (2, 3, 4)
 (2, 4, 6, 1)  (2, 4, 6)
 (4, 8, 12, 2)  (2, 4, 6)

University of Freiburg – Computer Science Department – Computer Graphics - 18

1D Illustration

University of Freiburg – Computer Science Department – Computer Graphics - 19

Homogeneous Coordinates of Vectors
 for varying w , a point is scaled and the
points represent a line in 3D space
 the direction of this line is characterized by
 for w  0, the point moves to infinity
in the direction
 is a point at infinity in the direction of
 is a vector in the direction of

University of Freiburg – Computer Science Department – Computer Graphics - 20

1D Illustration

University of Freiburg – Computer Science Department – Computer Graphics - 21

Points and Vectors
 if points are represented in the homogeneous (norma-
lized) form, point - vector relations can be represented
 vector + vector = vector

 point + vector = point

 point - point = vector

University of Freiburg – Computer Science Department – Computer Graphics - 22

Homogeneous Representation
of Linear Transformations

 if the transform of results in , then

 the transform of results in

University of Freiburg – Computer Science Department – Computer Graphics - 23

Affine Transformations and Projections

 general form

 mii represent rotation, scale

 ti represent translation
 pi represent projection
 w is analog to the fourth component
for points / vectors
University of Freiburg – Computer Science Department – Computer Graphics - 24
Homogeneous Coordinates - Summary

 with w ≠ 0 are the homogeneous

coordinates of the 3D point
 is a point at infinity in the direction of
 is a vector in the direction of
 is a transformation,
representing rotation, scale,
translation, projection

University of Freiburg – Computer Science Department – Computer Graphics - 25

Outline
 transformations in the rendering pipeline
 motivations for the homogeneous notation
 homogeneous notation
 basic transformations in homogeneous notation
 compositing transformations
 summary

University of Freiburg – Computer Science Department – Computer Graphics - 26

Translation
 of a point

 of a vector

 inverse (T-1 "undoes" the transform T)

University of Freiburg – Computer Science Department – Computer Graphics - 27

Rotation
 positive (anticlockwise)
rotation with angle 
around the x-, y-, z-axis

University of Freiburg – Computer Science Department – Computer Graphics - 28

Inverse Rotation

 the inverse of a rotation matrix

corresponds to its transpose
University of Freiburg – Computer Science Department – Computer Graphics - 29
Mirroring / Reflection
 mirroring with respect to x =0, y =0, z =0 plane
 changes the sign of the x -, y -, z - component

 the inverse of a reflection

corresponds to its transpose

University of Freiburg – Computer Science Department – Computer Graphics - 30

Orthogonal Matrices
 rotation and reflection matrices are orthogonal

 are orthogonal  is orthogonal

 rotation: reflection:
 length of a vector does not change
 angles are preserved

University of Freiburg – Computer Science Department – Computer Graphics - 31

Scale
 scaling x -, y - , z - components of a point or vector

 inverse
 uniform scaling: sx = sy = sz = s

University of Freiburg – Computer Science Department – Computer Graphics - 32

Shear
 one component is offset
with respect to another component
 six basic shear modes in 3D
 e.g., shear of x with respect to z

 inverse

University of Freiburg – Computer Science Department – Computer Graphics - 33

Basis Transform - Translation
 two coordinate systems

University of Freiburg – Computer Science Department – Computer Graphics - 34

Basis Transform - Translation
 the coordinates of with respect to are given
by
 the coordinates of a point in the transformed
basis correspond to the coordinates of point in the
untransformed basis transformed by the inverse
basis transform
 translating the origin by t corresponds
to translating the object by -t
 also: rotating the basis vectors by an angle corresponds
to rotating the object by the same negative angle

University of Freiburg – Computer Science Department – Computer Graphics - 35

Basis Transform - Rotation
 two coordinate systems

University of Freiburg – Computer Science Department – Computer Graphics - 36

Basis Transform - Rotation
 the coordinates of with respect to are given by

 b1, b2, b3 are the basis vectors of C2 with respect to C1

 b1, b2, b3 are orthonormal, therefore
the basis transform is a rotation
 rotating the basis vectors by an angle corresponds
to rotating the object by the same negative angle
University of Freiburg – Computer Science Department – Computer Graphics - 37
Basis Transform - Application
 the view transform can be seen as a basis transform
 objects are placed with respect to
a (global) coordinate system
 the camera is also positioned at and oriented at
given by viewing direction and up-vector
 after the view transform, all objects are represented in
the eye or camera coordinate system
 placing and orienting the camera corresponds to the
application of the inverse transform to the objects
 rotating the camera by R and translating it by T,
corresponds to translating the objects by T-1 and
rotating them by R -1
University of Freiburg – Computer Science Department – Computer Graphics - 38
Planes and Normals
 planes can be represented by a surface normal n and
a point r. All points p with form a plane.

 the transformed points

are on the plane represented by

 if a surface is transformed by A, its homogeneous notation

(including the surface normal) is transformed by (A-1)T
University of Freiburg – Computer Science Department – Computer Graphics - 39
Outline
 transformations in the rendering pipeline
 motivations for the homogeneous notation
 homogeneous notation
 basic transformations in homogeneous notation
 compositing transformations
 summary

University of Freiburg – Computer Science Department – Computer Graphics - 40

Compositing Transformations
 composition is achieved by matrix multiplication
 a translation T applied to p, followed by a rotation R

 a rotation R applied to p, followed by a translation T

 note that generally

 the order of composed transformations matters

University of Freiburg – Computer Science Department – Computer Graphics - 41

Examples
 rotation around a line through t parallel
to the x-, y-, z- axis

 scale with respect to an arbitrary axis

 e.g., b1, b2, b3 represent an orthonormal basis,

then scaling along these vectors can be done by

University of Freiburg – Computer Science Department – Computer Graphics - 42

Rigid-Body Transform


with R being a rotation and t being a

translation is a combined transformation
 inverse

 in Euclidean coordinates
 the inverse transform
 therefore

University of Freiburg – Computer Science Department – Computer Graphics - 43

Outline
 transformations in the rendering pipeline
 motivations for the homogeneous notation
 homogeneous notation
 basic transformations in homogeneous notation
 compositing transformations
 summary

University of Freiburg – Computer Science Department – Computer Graphics - 44

Summary
 usage of the homogeneous notation is motivated
by a unified processing of affine transformations,
perspective projections, points, and vectors
 all transformations of points and vectors are
represented by a matrix-vector multiplication
 "undoing" a transformation is represented by
its inverse
 compositing of transformations is accomplished
by matrix multiplication

University of Freiburg – Computer Science Department – Computer Graphics - 45