EE 7730

2D Fourier Transform
Summary of Lecture 2
 We talked about the digital image properties, including spatial
resolution and grayscale resolution.
 We reviewed linear systems and related concepts, including
shift invariance, causality, convolution, etc.

Bahadir K. Gunturk EE 7730 - Image Analysis I 2

Fourier Transform
 What is ahead?
 1D Fourier Transform of continuous signals
 2D Fourier Transform of continuous signals
 2D Fourier Transform of discrete signals
 2D Discrete Fourier Transform (DFT)

Bahadir K. Gunturk EE 7730 - Image Analysis I 3

Fourier Transform: Concept
■ A signal can be
represented as a weighted
sum of sinusoids.

■ Fourier Transform is a
change of basis, where the
basis functions consist of
sines and cosines.

Bahadir K. Gunturk EE 7730 - Image Analysis I 4

Fourier Transform
 Cosine/sine signals are easy to define and interpret.
 However, it turns out that the analysis and manipulation of
sinusoidal signals is greatly simplified by dealing with related
signals called complex exponential signals.

 A complex number: z = x + j*y

 A complex exponential signal: r*exp(j*a) =r*cos(a) + j*r*sin(a)

Bahadir K. Gunturk EE 7730 - Image Analysis I 5

Fourier Transform: 1D Cont. Signals

■ Fourier Transform of a 1D continuous signal


F (u )  

f ( x)e  j 2 ux dx

“Euler’s formula” e  j 2 ux  cos  2 ux   j sin  2 ux 

■ Inverse Fourier Transform


f ( x)  

F (u )e j 2 ux du

Bahadir K. Gunturk EE 7730 - Image Analysis I 6

Fourier Transform: 2D Cont. Signals

■ Fourier Transform of a 2D continuous signal

 
F (u , v)  
 
f ( x, y )e  j 2 (ux  vy ) dxdy

■ Inverse Fourier Transform

 
f ( x, y )  
 
F (u , v)e j 2 (ux  vy ) dudv

■ F and f are two different representations of the same signal.

f F
Bahadir K. Gunturk EE 7730 - Image Analysis I 7
Examples

Magnitude: “how much” of each component

Phase: “where” the frequency component in
the image

Bahadir K. Gunturk EE 7730 - Image Analysis I 8

Examples

Bahadir K. Gunturk EE 7730 - Image Analysis I 9

Fourier Transform: Properties

■ Linearity af ( x, y )  bg ( x, y )  aF (u, v)  bG (u, v)

 j 2 ( ux0  vy0 )
■ Shifting f ( x  x0 , y  x0 )  e F (u , v)
■ Modulation e j 2 (u0 x  v0 y ) f ( x, y )  F (u  u0 , v  v0 )
■ Convolution f ( x, y ) * g ( x, y )  F (u , v)G (u , v)
■ Multiplication f ( x, y ) g ( x, y )  F (u , v) * G (u, v)
■ Separable functions f ( x, y )  f ( x ) f ( y )  F (u , v )  F (u ) F (v )

Bahadir K. Gunturk EE 7730 - Image Analysis I 10

Fourier Transform: Properties

■ Separability
 
F (u , v)  
 
f ( x, y )e  j 2 (ux  vy ) dxdy



  j 2 vy
    f ( x , y )e  j 2 ux
dx  e dy
   

 

F (u , y )e  j 2 vy dy

2D Fourier Transform can be implemented as a

sequence of 1D Fourier Transform operations.

Bahadir K. Gunturk EE 7730 - Image Analysis I 11

Fourier Transform: Properties

■ Energy conservation

   

 
2 2
f ( x, y ) dxdy  F (u , v) dudv
   

Bahadir K. Gunturk EE 7730 - Image Analysis I 12

Fourier Transform: Properties

■ Remember the impulse function (Dirac delta function)

definition 

  ( x  x ) f ( x)dx  f ( x )

0 0

■ Fourier Transform of the impulse function

 
F  ( x, y )      ( x , y ) e  j 2 ( ux  vy )
dxdy  1
 

Bahadir K. Gunturk EE 7730 - Image Analysis I 13

Fourier Transform: Properties

■ Fourier Transform of 1
 
F 1    e  j 2 ( ux  vy )
dxdy   (u , v)
 

Take the inverse Fourier Transform of the impulse function

 
F 1  (u , v)      (u , v ) e j 2 ( ux  vy )
dudv e j 2 (0 x  v 0)
1
 

Bahadir K. Gunturk EE 7730 - Image Analysis I 14

Fourier Transform: 2D Discrete
Signals
■ Fourier Transform of a 2D discrete signal is defined as
 
F (u , v)  
m  n 
f [m, n]e  j 2 (um  vn )

1 1
where  u, v 
2 2

■ Inverse Fourier Transform

1/ 2 1/ 2
f [m, n]   
1/ 2 1/ 2
F (u, v)e j 2 (um  vn ) dudv

Bahadir K. Gunturk EE 7730 - Image Analysis I 15

Fourier Transform: Properties

■ Periodicity: Fourier Transform of a discrete signal is periodic

with period 1.
 
F (u  k , v  l )  
m  n 
f [m, n]e
 j 2  ( u  k ) m  ( v  l ) n 

 
1 1
  f [m, n]e
 j 2 um  vn   j 2 km  j 2 ln
e e
Arbitrary m  n 

integers  
 
m  n 
f [m, n]e  j 2 (um  vn )

 F (u , v)

Bahadir K. Gunturk EE 7730 - Image Analysis I 16

Fourier Transform: Properties

■ Linearity, shifting, modulation, convolution, multiplication,

separability, energy conservation properties also exist for the
2D Fourier Transform of discrete signals.

Bahadir K. Gunturk EE 7730 - Image Analysis I 17

Fourier Transform: Properties

■ Linearity af [m, n]  bg[m, n]  aF (u , v)  bG (u , v)

■ Shifting f [m  m0 , n  n0 ]  e  j 2 (um0  vn0 ) F (u , v)

j 2 ( u0 m  v0 n )
■ Modulation e f [m, n]  F (u  u0 , v  v0 )

■ Convolution f [m, n]* g[m, n]  F (u , v)G (u , v)

■ Multiplication f [m, n]g[m, n]  F (u , v) * G (u, v)

■ Separable functions f [m, n]  f [m] f [n]  F (u , v)  F (u ) F (v)

   

 
2 2
■ Energy conservation f [m, n]  F (u , v) dudv
m  n   

Bahadir K. Gunturk EE 7730 - Image Analysis I 18

Fourier Transform: Properties

■ Define Kronecker delta function

1, for m  0 and n  0 

 [m, n]   
0, otherwise 

■ Fourier Transform of the Kronecker delta function

 
F (u , v)   
m  n 

 [ m, n ]e
 j 2 um  vn 

  e
 j 2  u 0  v 0 
 1

Bahadir K. Gunturk EE 7730 - Image Analysis I 19

 Fourier Transform: Properties

■ Fourier Transform of 1
   
f (m, n)  1  F (u , v)    1e
m  n 
 j 2 um  vn 
     (u  k , v  l )
 k  l 

To prove: Take the inverse Fourier Transform of the Dirac

delta function and use the fact that the Fourier Transform has
to be periodic with period 1.

Bahadir K. Gunturk EE 7730 - Image Analysis I 20

Impulse Train

■ Define a comb function (impulse train) as follows

 
combM , N [m, n]     [m  kM , n  lN ]
k  l 

where M and N are integers

comb2 [n]
1

n
Bahadir K. Gunturk EE 7730 - Image Analysis I 21
Impulse Train
 
combM , N [m, n]     [m  kM , n  lN ]
k  l 
 
combM , N ( x, y )      x  kM , y  lN 
k  l 

 Fourier Transform of an impulse train is also an impulse train:

   
1  k l 
   m  kM , n  lN      u  ,v  
k  l  MN k  l   M N

combM , N [m, n] comb 1 1 (u, v)

,
M N

Bahadir K. Gunturk EE 7730 - Image Analysis I 22

Impulse Train

comb2 [n] 1
comb1 (u )
2 2

1 1
2

n u
1
2

Bahadir K. Gunturk EE 7730 - Image Analysis I 23

Impulse Train
 
combM , N ( x, y )      x  kM , y  lN 
k  l 

 In the case of continuous signals:

   
1  k l 
 
k  l 
  x  kM , y  lN  
MN
   u  ,v  
k  l   M N

combM , N ( x, y ) comb 1 1 (u, v)

,
M N

Bahadir K. Gunturk EE 7730 - Image Analysis I 24

Impulse Train

1
comb2 ( x) comb1 (u )
2 2

1 1
2

x u
2 1
2

Bahadir K. Gunturk EE 7730 - Image Analysis I 25

Sampling
f ( x) F (u )

x u
combM ( x) comb 1 (u )
M

x u
1
M
M

F (u ) * comb 1 (u )
f ( x)combM ( x) M

x u
Bahadir K. Gunturk EE 7730 - Image Analysis I 26
Sampling
f ( x) F (u )

x u
W W

F (u ) * comb 1 (u )
f ( x)combM ( x) M

x W
u
M
1
M
1
No aliasing if  2W
M

Bahadir K. Gunturk EE 7730 - Image Analysis I 27

Sampling
F (u ) * comb 1 (u )
f ( x)combM ( x) M

x W
u
M
1
M 1
2M

If there is no aliasing, the original signal

can be recovered from its samples by
low-pass filtering.

Bahadir K. Gunturk EE 7730 - Image Analysis I 28

Sampling
f ( x) F (u )

x u
W W

F (u ) * comb 1 (u )
M

f ( x)combM ( x)

W
u

1
Aliased M

Bahadir K. Gunturk EE 7730 - Image Analysis I 29

Sampling
f ( x) F (u )
Anti-aliasing
filter
x u
W W

1
2M
f ( x ) * h( x )
u
W W

 f ( x) * h( x) combM ( x)
u
1
M

Bahadir K. Gunturk EE 7730 - Image Analysis I 30

Sampling
■ Without anti-aliasing filter:

f ( x)combM ( x)

W
u
1
■ With anti-aliasing filter: M

 f ( x) * h( x) combM ( x)
u

1
M

Bahadir K. Gunturk EE 7730 - Image Analysis I 31

 Anti-Aliasing

a=imread(‘barbara.tif’);

Bahadir K. Gunturk EE 7730 - Image Analysis I 32

 Anti-Aliasing

a=imread(‘barbara.tif’);
b=imresize(a,0.25);
c=imresize(b,4);

Bahadir K. Gunturk EE 7730 - Image Analysis I 33

 Anti-Aliasing

a=imread(‘barbara.tif’);
b=imresize(a,0.25);
c=imresize(b,4);

H=zeros(512,512);
H(256-64:256+64, 256-64:256+64)=1;

Da=fft2(a);
Da=fftshift(Da);
Dd=Da.*H;
Dd=fftshift(Dd);
d=real(ifft2(Dd));

Bahadir K. Gunturk EE 7730 - Image Analysis I 34

Sampling
y v
f ( x, y ) F (u , v)
Wv

x u
Wu

comb 1 1 (u , v)
y v ,
M N
combM , N ( x, y )

x u
1
N N

M
1
M

Bahadir K. Gunturk EE 7730 - Image Analysis I 35

Sampling
v

Wv

f ( x, y )combM , N ( x, y ) u
1
N

1 Wu
M

1 1
No aliasing if  2Wu and  2Wv
M N

Bahadir K. Gunturk EE 7730 - Image Analysis I 36

Interpolation
v
1
2N

u
1
N

1
2M
1
M

Ideal reconstruction
filter:  1 1
 MN , for u  and v 
H (u , v)   2M 2N
 0, otherwise

Bahadir K. Gunturk EE 7730 - Image Analysis I 37

Ideal Reconstruction Filter
1 1
  2 N 2M
h ( x, y )  
 
H (u , v)e j 2 (ux  vy ) dudv   
1 1
MNe j 2 (ux  vy ) dudv
1 1 2 N 2M
2M 2N
 
1
Me j 2 ux du 
1
Ne j 2 vy dv
2M 2N

1  j 2 x 21M  1  j 2 y 2 N 
1 1 1
 j 2 x  j 2 y
M e e 2M
 N  e  e 2N

j 2 x   j 2 y  
    1 jx  jx
sin  x  sin  y sin( x)  e  e 
 M  N  2j
 
x y
M N

Bahadir K. Gunturk EE 7730 - Image Analysis I 38