10.

More on Transformations

Inverse Transformations

If a matrix causes a particular transformation on a point, then the inverse matrix reverses the
transformation. Therefore, the inverse matrix of the transformation allows us to reverse any
translation, stretching, rotation or skewing.

Example 1. The differential scaling matrix S has inverse S-1,

S=
[ ]
3 0
0 1
−1
has inverse S =
1/3 0
0 1
. [ ]
Applying S to the point , for instance, gives the new point

[ ][ ] [ ]
3 0 ¿2
0 1 ¿4
= ¿6 .
¿4

So applying S-1 to gets back to .

[ 1/30 01 ][ ¿¿ 64 ]=[ ¿¿ 24 ].
Recalling the “house” in chapter 7, S-1 “squashes” the stretched house back into the original.

Homogeneous Matrices

The homogeneous matrix of a transformation adds a dimension in order to make translation a

matrix operation like all the others. So a 3D vector of form becomes .

Then, for example, the translation matrix for a move 8 along x, 6 along y and 4 up z is

[ ]
1 0 0 8
0 1 0 6
.
0 0 1 4
0 0 0 1
This hides the old transformation vector in the top right-hand corner. Similarly, the
homogeneous matrix for rotation by 100 around the z axis is

[ ]
∘ ∘
cos 1 0 0 −sin 1 0 0 0 0
sin 1 0 0∘ cos 1 0 0∘ 0 0.
0 0 1 0
0 0 0 1

Generally, then, homogeneous matrices look like the following.

Inverses of Homogeneous Matrices

The pattern to follow is ;

Where
 The bulk of the matrix is the inverse of the 3 3 translation matrix
 The translation vector has been transformed using this inverse, then all its elements
inverted (multiplied by -1).

Concatenation

Now that all the operations are in the same form- matrix form- we can concatenate different
operations into the same matrix. That is to say, we can find a single matrix that
simultaneously translates, reflects, scales, rotates and/or shears an object. Indeed we can
concatenate any number of matrices, given a single matrix that saves a lot of computer
memory.

Example 2. Suppose we wished to first translate our shoebox as above, then to rotate it
through 100 around the z axis. The concatenated matrix consists of the rotation matrix
multiplied by the translation matrix.

[ ][ ]
∘ ∘
cos 1 0 0 −sin 1 0 0 0 0 1 0 0 8
sin 1 0 0∘ cos 1 0 0∘ 0 0 0 1 0 6
0 0 1 0 0 0 1 4
0 0 0 1 0 0 0 1

[ ]
0 ( 8 cos 1 0 0 −6 sin 1 0 0 )
∘ ∘ ∘ ∘
cos 10 0 −sin 10 0
0 ( 8 sin 10 0 +6 cos 1 0 0 )
∘ ∘ ∘ ∘
¿ sin1 0 0 cos 1 0 0
0 0 1 4
0 0 0 1

Note that the multiplication is in reverse order to the operations; translate then rotate, but
rotation matrix times translation.

L is convenient for demonstration. Operate on L’s vector- remember L is now 4D, - with
the translate then rotate matrix;

[ ][ ]
0 ( 8 cos 1 0 0 −6 sin1 0 0 )
∘ ∘ ∘ ∘
cos 1 0 0 −sin 1 0 0 5
0 ( 8 sin 1 0 0 +6 cos 1 0 0 ) 4
∘ ∘ ∘ ∘
sin 1 0 0 cos 1 0 0
0 0 1 4 0
0 0 0 1 1

[ ]
−12.1055
11.0660
¿ =LTR.
4
1

It’s important to remember that we’re using the term “moved” loosely when we talk about
transforming a point. It’s not like the point L really moves through the intervening space on

the screen, until it stops at . Rather, a new point is found at , as though L

had moved.
Euler Angles

Consider three different rotation matrices,

Rx which rotates angle p around the x axis,
Ry which rotates angle h around the y axis,
Rz which rotates angle r around the z axis.
The angles r, p and h are the Euler angles (named after Leonhard Euler, an 18th century
mathematician). If we concatenate the three matrices
Rx Ry Rz , the result is a matrix which describes the change in orientation of a 3D object in all
three axes.

Example 3. Returning to the shoe box,

rotations through Euler angles of r = 60, h = 72 and p = 30 might be rendered with the

familiar matrices. Taking point A as the example in each case, applying Rz to A = gives

. (All other points are likewise transformed, rotating the shoebox 60 around the z
axis.)

[ ][ ] [ ]
cos 6 0∘ −sin 6 0∘ 0 0 1 −4.6962
sin 6 0∘ cos 6 0∘ 0 0 6 = 3.8660
0 0 1 0 8 8
0 0 0 1 1 1
Then applying Ry to gives (rotating the shoebox 72 around the y axis).

[ ][ ] [ ]
cos 7 2∘ 0 sin 7 2∘ 0 −4.6962 6.1573
0 1 0 0 3.8660
= 3.8660
−sin 7 2∘ 0 cos 7 2∘ 0 8 6.9384
0 0 0 1 1 1

Then applying Rx to gives (rotating the shoebox 30 around the x axis).

[ ][ ] [ ]
1 0 0 0 6.1573 6.1573
0 cos 3 0 −sin 3 0∘
∘
0 3.8660 −0.1211
∘ ∘
=
0 sin 3 0 cos 3 0 0 6.9384 7.9419
0 0 0 1 1 1

But concatenating the three matrices gives a single matrix in which the Euler angles can be
inserted to jump straight from the first orientation (in which A is ) to the last (where A is
 ). We’ll call this the E matrix, which for the above angles would be

[ ]
cos 6 0 cos 7 2 −cos 7 2sin 6 0 sin 7 2 0
sin 3 0 sin7 2 cos 6 0+cos 3 0 sin 6 0 −sin 3 0 sin 7 2 sin6 0+cos 3 0 cos 6 0 −sin 3 0 cos 7 2 0
−cos 3 0 sin 7 2 cos 6 0+sin 3 0 sin 6 0 cos 3 0 sin 7 2 sin 6 0+sin 3 0 cos 6 0 cos 3 0 cos 7 2 0
0 0 0 1

This would have the general form

[ ]
cos r cos h −cos h sin r sin h 0
sin p sin hcos r +cos p sin r −sin p sin h sin r + cos p cos r −sin p cos h 0
−cos p sin h cos r +sin p sin r cos p sin h sin r +sin p cos r cos p cos h 0
0 0 0 1 []

in which it’s only necessary to decide what values of r, p and h you require.

Example 4. Consider a die, seen from faces 1, 2 and 3. What faces would show after
successive Euler angles of r = h = p = 90?

If we define this as a cube with vertices A to H as follows,

[ ] [] [ ] [ ] [ ] [ ] [ ] [ ]
1 1 −1 −1 1 1 −1 −1
−1 1 1 −1 −1 1 1 −1
A= , B= , C= , D= , E= , F= , G= , H= .
1 1 1 1 −1 −1 −1 −1
1 1 1 1 1 1 1 1

then the cube lies with its centre at the origin, and successive rotations won’t displace the
cube.
In order to turn the die first 90 around the z axis, then 90 around the x axis, and finally 90
around the y axis, the familiar three could be applied in turn to each point. In a program, we
might choose to do this to simulate the die rolling.

So faces 1, 3 and 5 show. But we could more easily place the angles r = h = p = 90 into the
single E matrix as above. Because sin90 = 0 and cos90 = 1, the matrix E simplifies to

[ ]
0 0 1 0
0 −1 0 0
,
1 0 0 0
0 0 0 1

which goes straight to the final orientation.