Image Processing - Lesson 5

Fourier Transform - Part I

Introduction to Fourier Transform

Image Transforms
Basis to Basis
Fourier Basis Functions
Fourier Coefficients

Fourier Transform - 1D
Fourier Transform - 2D

The Fourier Transform

Jean Baptiste Joseph Fourier

Efficient Data
Representation
Data can be represented in many
ways.
There is a great advantage using an
appropriate representation.
It is often appropriate to view images
as combinations of waves.

How can we enhance such an image?

Solution: Image Representation

= 3

+5

+ 10

+ 23

+3

+ ...

+1

+5

+ ...

The inverse Fourier Transform

For linear-systems we saw that it is
convenient to represent a signal f(x)
as a sum of scaled and shifted
sinusoids.

f ( x ) = F ( ) e i 2x d

How is this done?

Transforms: Change of Basis

Standard Basis
Grayscale Image
X Coordinate

New Basis
Fourier Image
Frequency Coordinate

Standard Basis:

[ a1 a2 a3 a4 ] =
a1 [ 1 0 0 0 ] + a2 [ 0 1 0 0 ] + a3 [ 0 0 1 0 ] + a4 [ 0 0 0 1 ]

Hadamard Transform:

[ 2 1 0 1] =
= 1 [ 1 1 1 1 ] + 1/2 [ 1 1 -1 -1 ] - 1/2 [ -1 1 1 -1 ] + 0 [ -1 1 -1 1 ]
= [ 1 1/2 -1/2 0 ] Hadamard

1. Basis Functions.
2. Method for finding the image given the transform coefficients.
3. Method for finding the transform coefficients given the image.

Hadamard Basis Functions - 1D

Wave Number
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

N = 16

Finding the transform coefficients

Signal:

X = [ 2 1 0 1]

New Basis:

T0 = [ 1 1 1 1 ]

standard

T1 = [ 1 1 -1 -1 ]
T2 = [ -1 1 1 -1 ]
T3 = [ -1 1 -1 1 ]
New Coefficients:
a0 = <X,T0 > =< [ 2 1 0 1 ] , [ 1 1 1 1 ] > /4 = 1
a1 = <X,T1 > =< [ 2 1 0 1 ] , [ 1 1 -1 -1 ] > /4 = 1/2
a2 = <X,T2 > =< [ 2 1 0 1 ] , [ -1 1 1 -1 ] > /4 = -1/2
a3 = <X,T3 > =< [ 2 1 0 1 ] , [ -1 1 -1 1 ] > /4 = 0

Signal:

X = [ 1 1/2 -1/2 0 ]

new

Y Frequency

Y Coordinate

Transforms: Change of Basis - 2D

Grayscale
Image

X Coordinate

Fourier
Image

X Frequency

Standard Basis:

a11 a12
a21 a22

[ ]

= a11

1 0
0 0

[ ] [ ] [ ]

+ a12

0 1
0 0
0 0
+ a21
+ a22
0 0
1 0
0 1

Hadamard Transform:
2

[ ]

[ ] [ ] [ ] [ ]
[ ]

= 1

1 1
1 -1
-1 -1
-1 1
+ 1/2
- 1/2
+ 0
1 1
1 -1
1 1
1 -1

1 1/2
-1/2 0

Hadamard

1. Basis Functions.
2. Method for finding the image given the transform coefficients.
3. Method for finding the transform coefficients given the image.

Standard Basis:

] [ ] [ ] [ ] [ ]
1 0
0 0

= 2

+ 1

0 1
+ 1
0 0

0 0
+ 0
1 0

0 0
0 1

[ ] [ ]
1 0
0 0

0 1
0 0

[ ] [ ]
0 0
1 0

coefficients

0 0
0 1

Standard

Basis Elements
Hadamard Transform:
1

1/2

-1/2

] [ ]
= 1

1 1/2
-1/2 0

coefficients

[ ] [ ] [ ]

1 1
1 -1
-1 -1
-1 1
+ 1/2
- 1/2
+ 0
1 1
1 -1
1 1
1 -1

[ ] [ ]
1 1
1 1

Hadamard

1 -1
1 -1

[ ] [ ]
-1 -1
1 1

-1 1
1 -1

Hadamard

Basis Elements

Hadamard Basis Functions

size = 8x8

Black = +1 White = -1

For continuous images/signals f(x):

1) The number of Basis Elements Bi is .

a
B
(x)
di
i
i

f ( x) =

2) The dot product:

<f(x),Bi(x)> =

f(x) B i (x) dx

Fourier Transform
Basis Functions are sines and cosines

sin(x)

cos(2x)

sin(4x)

The transform coefficients determine the amplitude:

a sin(2x)

2a sin(2x)

-a sin(2x)

Every function equals a sum of sines and cosines

3 sin(x)

+ 1 sin(3x)

B
A+B

+ 0.8 sin(5x)

C
A+B+C

+ 0.4 sin(7x)

D
A+B+C+D

The Fourier Transform

The inverse Fourier Transform composes a
signal f(x) given F( )

f (x) =

F ( ) e i 2 x d

The Fourier Transform finds the F( )

given the signal f(x):

F ( ) = f(x)e
x

i2 x

dx

 F() is the Fourier transform of f(x):

~
F { f ( x )} = F ( )
f(x) is the inverse Fourier transform
of F():
~
F 1{F ( )} = f ( x )

f(x) and F() are a Fourier

transform pair.

 The Fourier transform F() is a

function over the complex numbers:

F ( ) = R e

R tells us how much of frequency

is needed.
tells us the shift of the Sine wave
with frequency .

Alternatively:

F ( ) = a + ib
a tells us how much of cos with
frequency is needed.
b tells us how much of sin with
frequency is needed.

 R - is the amplitude of F().

- is the phase of F().
|R|2=F* () F() - is the power
spectrum of F() .
If a signal f(x) has a lot of fine details
F() will be high for high .
If the signal f(x) is "smooth" F() will
be low for high .

Why do we need representation in

the frequency domain?

Problem in
Frequency
Space

Relatively
easy solution

Inverse
Fourier
Transform

Fourier
Transform

Original
Problem

Solution in
Frequency
Space

Difficult
solution

Solution of
Original
Problem

Examples:
The Delta Function:
Let

f ( x) = (x )
F ( ) =

i 2 x

(
x
)

e
=1

f(x)

Fourier

The Constant Function:

Let

f ( x ) =1

F() = e

i 2 x

= ()

f(x)

Fourier

Real

Imag

The Cosine wave:

f ( x) = cos(2 0 x )

Let

1 i 20 x i 20 x i 2x
F ( )= e
+e
e
dx =
2

1
= [ ( 0 )+ ( + 0 )]
2
f(x)

Fourier

Real

-0

-0

Imag

The Window Function (rect):

1 if x < 1
rect 1 ( x ) =
2
2
0 otherwise

Let
0 .5

F ( ) = e
0.5

i 2x

sin ( )
dx =
= sinc ( )

f(x)

x
-0.5

0.5

Fourier

Proof:

f(x) = rect1/2(x) =

-1/2

|x| 1/2

otherwise

1/2

1/ 2

1/ 2

F(u) = f (x)e2iuxdx= e2iuxdx

1
2iux 1/ 2
=
e
1/ 2
2iu
1
=
eiu eiu

[
2iu

1
=
[cos(u) i sin(u) cos(u) i sin(u)]
2iu
sin(u)
=
u

= sinc(u)
F(u)

sinc(u)

The Gaussian:
Let

f ( x) = e

x 2

F ( )=e

f(x)

x
Fourier

The bed of nails function:

f(x)

ck(x)
x

k
Fourier

C1/k()

1/k

Fourier Transform - 2D
Given a continuous real function f(x,y),
its Fourier transform F(u,v) is defined as:

~
F{f (x,y)}= F(u,v) = f (x,y)e 2i(ux+vy)dxdy

The Inverse Fourier Transform:

~ 1
F {F(u,v)}= f(x,y) = F(u,v)e2 i(ux+vy)dudv

F(u,v) = a(u,v) + ib(u,v) =|F(u,v)|ei(u,v)

Phase =

(u,v) = tg-1(b(u,v)/a(u,v))

Spectrum (Amplitude) = |F(u,v)| = a2(u,v) + b2(u,v)

Power Spectrum =

|F(u,v)|2 =

a2(u,v) + b2(u,v)

Fourier Wave Functions - 2D

F(u,v) is the coefficient of the sine wave e2i(ux+vy)
y

1
u2 + v2

1
v

1/u

lines of
(real) value 1

e2i(ux+vy) = cos(2(ux+vy)) + isin(2(ux+vy))

The ratio u
v determines the Direction.
The size of u,v determines the Frequency.

u=0

direction of waves

v=0
direction of waves

u=-2, v=2

u=-1, v=2

u=0, v=2

u=1, v=2

u=2, v=2

u=-2, v=1

u=-1, v=1

u=0, v=1

u=1, v=1

u=2, v=1

U
u=-2, v=0

u=-1, v=0

u=0, v=0

u=1, v=0

u=2, v=0

u=-2, v=-1

u=-1, v=-1

u=0, v=-1

u=1, v=-1

u=2, v=-1

u=-2, v=-2

u=-1, v=-2

u=0, v=-2

u=1, v=-2

u=2, v=-2

Fourier Transform 2D - Example

2D Function

2D Fourier Transform

Fourier Transform 2D - Example

f(x,y)
1
1/2
1/2

f(x,y) = rect(x,y) =

1 |x| 1/2, |y| 1/2

0 otherwise

F(u,v) = sinc(u) sinc(v) = sinc(u,v)

|F(u,v)|

Proof of Fourier of Rect = sinc in 2D

F (u ) =

f ( x, y )e 2i ( ux + vy ) dxdy =

2i ( ux + vy )
e
dxdy

1 / 2 1 / 2

1/2

1 / 2 1/ 2

e 2 iux dx

1 / 2

1/2

e 2 ivy dy

1 / 2

sin( u ) sin( v )
=
= Sinc ( u , v )
u
v

Fourier Transform Examples

Image Domain

Frequency Domain