 Types

• Scaling
• Rotation
• Shear
• Translation
 Mathematical
representation
 sx

Horizontal shift proportional to horizontal position

 sy
1

Vertical shift proportional to vertical position

µ
µ
 µ

(this is a counterclockwise rotation; reverse signs of sines to get a clockwise one)

µ
rizontal displacement proportional to vertical position
(Shear factor h is positive for the figure above)
# Shifting Right # Shifting Up

for j in range(WIDTH): for j in range(WIDTH):

for i in range(HEIGHT): for i in range(HEIGHT):
if (i < HEIGHT-20): if (j < WIDTH - 20 and j >
img[j][i] = img[j][i+20] 20):
plt.imshow(img) img[j][i] = img[j+20][i]
plt.show() else:
img[j][i] = 0
plt.imshow(img)
plt.show()
Also used in Image Augmentation to increase
our dataset size to train robust Convolutional
Network models
 import cv2
 import numpy as np

 img = cv2.imread('messi5.jpg',0)
 rows,cols = img.shape

 M = np.float32([[1,0,100],[0,1,50]])
 dst = cv2.warpAffine(img,M,(cols,rows))

 cv2.imshow('img',dst)
 cv2.waitKey(0)
 cv2.destroyAllWindows()
 replace 2d points with 3d points, last
coordinate 1

 for a 3d point (x,y,w) the corresponding

2d point is (x/w,y/w) if w is not zero
 each 2d point (x,y) corresponds to a
line in 3d; all points on this line can be
written as [kx,ky,k] for some k.
 From homogeneous to 2d: [x,y,w]
becomes [x/w,y/w]
 From 2d to homogeneous: [x,y] becomes
[kx,ky,k] (can pick any nonzero k!)
 Full-generality 3 x 3 homogeneous
transformation is called a homography
 Full-generality 3 x 3 homogeneous
transformation is called a homography

Translation components
 Full-generality 3 x 3 homogeneous
transformation is called a homography

Scale/rotation components
 Full-generality 3 x 3 homogeneous
transformation is called a homography

Shear/rotation components
 Full-generality 3 x 3 homogeneous
transformation is called a homography

Homogeneous scaling factor

 Full-generality 3 x 3 homogeneous
transformation is called a homography

When these are zero (as they have been so far), H

is an affine transformation
 Methods for object modeling
transformation in three dimensions
are extended from two dimensional
methods by including consideration
for the z coordinate.
 Generalize from 2D by including z coordinate
 Straightforward for translation and scale,
rotation more difficult
 Homogeneous coordinates: 4 components
 Transformation matrices: 4×4 elements
 We will consider points as column vectors.
Thus, a typical point with coordinates (x, y, z)
is represented as:

 x
 y
 
 z 
 A 3D point P is represented in
homogeneous coordinates by a 4-dim. Vect:

x
 y
P 
z
 
1 
 We don't lose anything
x
 The main advantage: it is easier to  y
compose translation and rotation  
 Everything is matrix multiplication z
 
1 
 Right Hand  Left Hand coordinate
coordinate system: system:
 In homogeneous coordinates, 3D
transformations are represented by 4×4
matrixes:
a b c t
x

d e f t 
 y

g h i t
z
 
0 0 0 1
3D Translation
 P is translated to P' by:
 x  1 0 0 tx  x
 y  0 1 0 t y   y 
   
 z  0 0 1 tz   z
     
1 0 0 0 1  1 

P  T  P
 An object is translated in 3D dimensional by
transforming each of the defining points of the
objects .
 An Object represented as a set of polygon surfaces, is translated
by translate each vertex of each surface and redraw the polygon
facets in the new position.

 x ' , y ' , z '

T  t x , t y , t z 
 x, y , z 

 Inverse Translation:
T (t , t , t )  T (t ,t ,t )
1
x y z x y z
3D Rotation
 In general, rotations are specified by a
rotation axis and an angle. In
two-dimensions there is only one
choice of a rotation axis that leaves
points in the plane.
 The easiest rotation axes are those that parallel to the
coordinate axis.
 Positive rotation angles produce counterclockwise rotations
about a coordinate axix, if we are looking along the positive half
of the axis toward the coordinate origin.
Coordinate Axis
Rotations
 Z-axis rotation: For z axis same as 2D rotation:
 x' cos   sin  0 0  x 
 y '  sin  cos  0 0  y 
  
 z'  0 0 1 0  z 
     
1  0 0 0 1  1 

P   R z (θ )  P
 X-axis rotation:
 x '  1 0 0 0  x 
 y ' 0 cos   sin  0  y 
   
 z '  0 sin  cos  0  z 
     
 1  0 0 0 1  1 

P   R x ( )  P
 Y-axis rotation:
 x'  cos  0 sin  0  x 
 y '  0 1 0 0  y 
   
 z '   sin  0 cos  0  z 
     
1  0 0 0 1  1 

P   R ( )  P
y
Other way to look at
rotation
Roll, Pitch, Yaw
 Imagine three lines running through an
airplane and intersecting at right
angles at the airplane's center of
gravity.
 Roll: rotation around the front-to-back
axis.
 PITCH: Rotation around the side-to-side
axis
 YAW: Rotation around the vertical axis.
An Example of the
Airplane
 Consider the following example. An airplane is oriented
such that its nose is pointing in the positive z direction,
its right wing is pointing in the positive x direction, its
cockpit is pointing in the positive y direction. We want
to transform the airplane so that it heads in the direction
given by the vector DOF (direction of flight), is centre
at P, and is not banked.
3D Scaling
 About origin: Changes the size of the object and
repositions the object relative to the coordinate origin.
 x  s x 0 0 0  x 
 y   0 sy 0 0  y 
  
 z   0 0 sz 0  z 
     
1 0 0 0 1  1 
P  S  P
 Composite
3D Transformations
 Multiply matrices
 Rightmost term in matrix product is the first
transformation to be applied
Rotation axis parallel with coordinate axis (Example x axis):

P  T 1
 R x (θ )  T  P
An arbitrary axis (with the rotation axis projected onto the Z
axis):

R ( )  T 1  R x 1 ( )  R y1 (  )  R z ( )  R y (  )  R x ( )  T
 To compose transformations, multiply
the matrices:
• composition of a rotation and a translation:
M = RT
 all transformations can be expressed
as matrices
• even transformations that are not
translations, rotations and scaling
 Rotation about a point Q:
• translate Q to origin (TQ),
• rotate about origin (R)
• translate back to Q (- TQ).

P’=(-TQ)RTQ P