Chapter 4: Filtering in the

Frequency Domain
 Outline

► Fourier Transform

► Filtering in Fourier Transform Domain

4/7/2019 2
 Fourier Series and Fourier Transform: History
► Jean Baptiste Joseph Fourier, French mathematician and physicist
(03/21/1768-05/16/1830) http://en.wikipedia.org/wiki/Joseph_Fourier

Permanent
Orphaned: at nine
Secretary of the
French Academy of
Egyptian expedition Sciences: 1822
with Napoleon I:
1798 Théorie analytique
Governor of Lower de la chaleur :
Egypt 1822

(The Analytic
Theory of Heat)

4/7/2019 3
 Fourier Series and Fourier Transform: History

► Fourier Series
Any periodic function can be expressed as the sum of sines
and /or cosines of different frequencies, each multiplied by
a different coefficients

► Fourier Transform
Any function that is not periodic can be expressed as the
integral of sines and /or cosines multiplied by a weighing
function

4/7/2019 4
 Fourier Series: Example

4/7/2019 5
 Preliminary Concepts

j = −1, a complex number

C = R + jI
the conjugate
C * = R - jI

| C |= R 2 + I 2 and  = arctan( I / R )
C =| C | (cos  + j sin  )
Using Euler's formula,
C =| C | e j

4/7/2019 6
 Fourier Series

A function f (t ) of a continuous variable t that is periodic

with period, T , can be expressed as the sum of sines and
cosines multiplied by appropriate coefficients
 2 n

ce
j t
f (t ) = n
T

n =−

where
2 n
1 T /2 −j t
cn =  f (t )e T
dt for n = 0, 1, 2,...
T −T /2

4/7/2019 7
 Impulses and the Sifting Property (1)
A unit impulse of a continuous variable t located
at t =0, denoted  (t ), defined as
 if t = 0
 (t ) = 
0 if t  0
and is constrained also to satisfy the identity

 −
 (t )dt = 1


The sifting property  −

f (t ) (t − t0 )dt = f (t0 )

−
f (t ) (t ) dt = f (0)
4/7/2019 8
 Impulses and the Sifting Property (2)
A unit impulse of a discrete variable x located
at x =0, denoted  ( x), defined as
1 if x = 0
 ( x) = 
0 if x  0
and is constrained also to satisfy the identity


  ( x) = 1
x =−

The sifting property 
x =−
f ( x) ( x − x0 ) = f ( x0 )



x =−
f ( x) ( x) = f (0)
4/7/2019 9
 Impulses and the Sifting Property (3)

impulse train sT (t ),


sT (t ) =   (t − nT )
n =−

4/7/2019 10
 Fourier Transform: One Continuous Variable

The Fourier Transform of a continous function f (t )


F (  ) = { f (t )} =  f (t )e − j 2t dt
−

The Inverse Fourier Transform of F (  )


f (t ) =  {F (  )} =  F (  )e j 2t d 
−1
−

4/7/2019 11
 Fourier Transform: One Continuous Variable

 W /2
F ( ) =  f (t )e − j 2 t
dt =  Ae − j 2t dt
− −W /2

−A − j 2 t W /2 A
= e  = e jW − e − jW 
j 2 −W /2 j 2 W
sin(W )
= AW
(W )

4/7/2019 12
 Fourier Transform: Impulses

The Fourier transform of a unit impulse located at the origin:


F (  ) =   (t )e − j 2t dt
−

= e − j 2 0
=1
The Fourier transform of a unit impulse located at t = t0 :

F (  ) =   (t − t0 )e − j 2t dt
−

= e − j 2t0
=cos(2 t0 ) − j sin (2 t0 )

4/7/2019 13
 Fourier Transform: Impulse Trains

Impulse train sT (t ), sT (t ) =   (t − nT )
n =−

The Fourier series:

 2 n

c
j t
sT (t ) = n e T

n =−

where
2 n
1 T /2 −j t
cn =
T −T /2
sT (t )e T
dt

4/7/2019 14
 Fourier Transform: Impulse Trains

 j 21Tn t T /2  j 2−Tjn2tTn t− j 2t 1 T /2 j 2T−nj t2Tn−t j 2t

cn e= −=

T

T /2 − T −T −
sT (et )e e dt = dt = e(t )e e dt dt
/2

1 0 1
= e = n )t
 − j 2 (  −
T TT n
= e dt =  (  − )
− T
 2 n n  
2 n n
  ( j −T t ) 1=   ( j − t )e j 2t du
−
1 
sT (t ) = c n e T=   −
e T T
n =−2 nt T n =−
j
= e T

4/7/2019 15
 Fourier Transform: Impulse Trains
Let S (  ) denote the Fourier transform of the
periodic impulse train S T (t )
2 n
 1 

S (  ) = S T (t ) =   e
j t
T

 T n =− 
1   j 2Tn t 
=   e 
T  n =− 
1  n
= 
T n =−
 ( −
T
)

4/7/2019 16
 Fourier Transform and Convolution
The convolution of two functions is denoted
by the operator

f (t ) h(t ) =  f ( ) h(t −  ) d
−

h(t ) =   f ( )h(t −  )d e − j 2t dt


 
 f (t )
− 
 − 

f ( )  h (t −  )e − j 2t dt d

 
=
−  − 

= f ( )  H ( )e − j 2 d
−

=H ( )  f ( )e − j 2 d
−

=H ( ) F (  )
4/7/2019 17
 Fourier Transform and Convolution

Fourier Transform Pairs

f (t ) h(t )  H (  ) F (  )

f (t )h(t )  H (  ) F (  )

4/7/2019 18
 Fourier Transform of Sampled Functions

f (t ) = f (t ) sT (t )

= 
n =−
f (t ) (t − nT )

4/7/2019 19
 Fourier Transform of Sampled Functions

 
F ( ) =  f (t ) =  f (t ) sT (t ) = F (  ) ? S (  )

F (  ) = F1(  ) S (  ) = n F ( ) S (  −  )d
S ( ) = 
1T n =−

 (  −
 T
−
)
n
=  F ( )   (  −  − )d
T −
n =− T
 
1 n
=
T

n =−
−
F ( ) (  −  −
T
) d

1 n
=
T

n =−
F ( −
T
)

4/7/2019 20
 Question

The Fourier transform of the

sampled function (shown in the
following figure) is

1. Continuous

2. Discrete

4/7/2019 21
 Fourier Transform of Sampled Functions
► A bandlimited signal is a signal whose Fourier transform
is zero above a certain finite frequency. In other words, if
the Fourier transform has finite support then the signal is
said to be bandlimited.

An example of a simple bandlimited signal is a sinusoid of

the form,

x(t ) = sin(2 ft +  )

4/7/2019 22
 Fourier Transform of Sampled Functions

− max  max
Over-sampling
1
 2  max
F ( ) = T

1 n
T

n =−
F ( −
T
) Critically-sampling
1
= 2 max
T
under-sampling
1
 2 max
4/7/2019 T 23
 Nyquist–Shannon sampling theorem

► We can recover f (t ) from its sampled version if we can

isolate a copy of F ( ) from the periodic sequence of copies
of this function contained in F (  ) , the transform of the
sampled function f (t )

1
► Sufficient separation is guaranteed if  2max
T
Sampling theorem: A continuous, band-limited function
can be recovered completely from a set of its samples if
the samples are acquired at a rate exceeding twice the
highest frequency content of the function
4/7/2019 24
 Nyquist–Shannon sampling theorem

?

f (t ) =  F (  )e j 2t d 
−

4/7/2019 25
 Aliasing

If a band-limited function is sampled at a rate that is less

than twice its highest frequency?

The inverse transform will yield a corrupted function. This

effect is known as frequency aliasing or simply as
aliasing.

4/7/2019 26
 Aliasing

4/7/2019 27
 Aliasing

4/7/2019 28
 Function Reconstruction from Sampled Data

F ( ) = H ( ) F ( )

f (t ) =  −1
F (  )
= −1
H ( ) F (  )
= h (t ) f ( t )

f (t ) =  f (nT )sinc  (t − nT ) / nT 
n =−
4/7/2019 29
 The Discrete Fourier Transform (DFT) of One
Variable

M −1
F (  ) =  f ( x)e − j 2 x / M ,  =0,1,..., M − 1
x =0

M −1
1
f ( x) =
M
 F
 =0
(  ) e j 2 x / M
, x = 0,1, 2,..., M − 1

4/7/2019 30
 2-D Impulse and Sifting Property: Continuous
 if t = z = 0
The impulse  (t , z ),  (t , z ) = 
0 otherwise
 
and   − −
 (t , z )dtdz = 1

The sifting property

 
 
− −
f (t , z ) (t , z ) dtdz = f (0, 0)
and
 
 
− −
f (t , z ) (t − t0 , z − z0 ) dtdz = f (t0 , z0 )

4/7/2019 31
 2-D Impulse and Sifting Property: Discrete

1 if x = y = 0
The impulse  ( x, y),  ( x, y ) = 
0 otherwise

The sifting property

 


x =− y =−
f ( x, y ) ( x, y ) = f (0, 0)

and
 


x =− y =−
f ( x, y ) ( x − x0 , y − y0 ) = f ( x0 , y0 )

4/7/2019 32
 2-D Fourier Transform: Continuous

 
F (  , ) =   f (t , z )e − j 2 (  t + z )
dtdz
− −

and
 
f (t , z ) =   f (  , )e j 2 (  t + z )
d  d
− −

4/7/2019 33
 2-D Fourier Transform: Continuous
 
F (  , ) =   f (t , z )e − j 2 ( t + z ) dtdz
− −
T /2 Z /2
=  Ae − j 2 ( t + z ) dtdz
−T /2 − Z /2

 sin(T )   sin( T ) 
= ATZ     T 
  T  

4/7/2019 34
 2-D Sampling and 2-D Sampling Theorem

2 − D impulse train:
 
sT Z (t , z ) =    (t − mT , z − nZ )
m =− n =−

4/7/2019 35
 2-D Sampling and 2-D Sampling Theorem

Function f (t , z ) is said to be band-limited if its Fourier transform

is 0 outside a rectangle established by the intervals [-max ,max ]
and [- max , max ], that is
F (  , ) = 0 for |  | max and |  |  max

Two-dimensional sampling theorem:

A continuous, band-limited function f (t , z ) can be recovered with
no error from a set of its samples if the sampling intervals are
1 1
T< and Z<
2max 2 max
4/7/2019 36
 2-D Sampling and 2-D Sampling Theorem

4/7/2019 37
 Aliasing in Images: Example
In an image system, the
number of samples is fixed at
96x96 pixels. If we use this
system to digitize checkerboard
patterns …

Under-sampling

4/7/2019 38
 Aliasing in Images: Example

Re-sampling

4/7/2019 39
 Aliasing in Images: Example

Re-sampling

4/7/2019 40
 Moiré patterns

► Moiré patterns are often an undesired artifact of images

produced by various digital imaging and computer graphics
techniques
e. g., when scanning a halftone picture or ray tracing a
checkered plane. This cause of moiré is a special case of
aliasing, due to under-sampling a fine regular pattern

http://en.wikipedia.org/wiki/Moiré_pattern

4/7/2019 41
 Moiré patterns

4/7/2019 42
 Moiré patterns

4/7/2019 43
 Moiré patterns
A moiré pattern
formed by
incorrectly down-
sampling the
former image

4/7/2019 44
 2-D Discrete Fourier Transform and Its
Inverse
DFT:
M −1 N −1 − j 2 (  x / M + y / N )

F (  , ) =   f ( x, y )e
x =0 y =0

 = 0,1, 2,..., M − 1; = 0,1, 2,..., N − 1;

f ( x, y ) is a digital image of size M  N.

IDFT:
M −1 N −1 j 2 (  x / M + y / N )
1
f ( x, y ) =
MN
  F ( , )e
x =0 y =0

4/7/2019 45
 Properties of the 2-D DFT
relationships between spatial and frequency intervals

Let T and Z denote the separations between samples,

then the seperations between the corresponding discrete,
frequency domain variables are given by
1
 =
M T
1
and  =
N Z

4/7/2019 46
 Properties of the 2-D DFT
translation and rotation

f ( x, y )e j 2 ( 0 x / M + 0 y / N )  F (  − 0 , − 0 )
and
− j 2 (  x0 / M + y0 / N )
f ( x - x0 , y - y0 )  F (  , )e

Using the polar coordinates

x = r cos  y=rsin  = cos  = sin 
results in the following transform pair:
f (r , +  0 )  F ( ,  +  0 )
4/7/2019 47
 Properties of the 2-D DFT
periodicity

2 − D Fourier transform and its inverse are infinitely periodic

F (  , ) = F (  + k1M , ) = F (  , + k2 N ) = F (  + k1M , + k2 N )

f ( x, y ) = f ( x + k1M , y ) = f ( x, y + k2 N ) = f ( x + k1M , y + k 2 N )

f ( x)e j 2 ( 0 x / M )  F (  − 0 )
0 = M / 2, f ( x)(−1) x  F (  − M / 2)

f ( x, y )(−1) x + y  F (  − M / 2, − N / 2)

4/7/2019 48
 Properties of the 2-D DFT
periodicity

4/7/2019 49
 Properties of the 2-D DFT
Symmetry

4/7/2019 50
 Properties of the 2-D DFT
Fourier Spectrum and Phase Angle

2-D DFT in polar form

F (u, v) =| F (u, v) | e j (u ,v )
Fourier spectrum
1/2
| F (u, v) |=  R (u, v) + I (u, v) 
2 2

Power spectrum
P(u, v) =| F (u, v) |2 = R 2 (u, v) + I 2 (u, v)
Phase angle
 I (u, v) 
 (u,v)=arctan  
 R (u , v ) 

4/7/2019 51
4/7/2019 52
4/7/2019 53
 Example: Phase Angles

4/7/2019 54
Example: Phase Angles and The Reconstructed

4/7/2019 55
 2-D Convolution Theorem
1-D convolution
M −1
f ( x ) h( x ) =  f ( m)h( x − m)
m =0

2-D convolution
M −1 N −1
f ( x, y ) h ( x, y ) =   f (m, n)h( x − m, y − n)
m =0 n =0

x = 0,1,2,..., M − 1; y = 0,1,2,..., N − 1.
f ( x, y ) h( x, y )  F (u, v) H (u, v)
f ( x, y )h( x, y )  F (u, v) H (u, v)
4/7/2019 56
 An Example of Convolution

Mirroring h
about the
origin

Translating
the mirrored
function by x

Computing the
sum for each
x
4/7/2019 57
 An Example of Convolution

It causes the
wraparoun
d error

It can be
solved by
appending
zeros

4/7/2019 58
 Zero Padding

► Consider two functions f(x) and h(x) composed of A and B

samples, respectively

► Append zeros to both functions so that they have the same

length, denoted by P, then wraparound is avoided by
choosing

P ≥A+B-1

4/7/2019 59
 Zero Padding

► Let f(x,y) and h(x,y) be two image arrays of sizes A×B and
C×D pixels, respectively. Wraparound error in their
convolution can be avoided by padding these functions
with zeros

 f ( x, y ) 0  x  A -1 and 0  y  B -1
f p ( x, y ) = 
 0 A  x  P or B  y  Q

 h ( x, y ) 0  x  C -1 and 0  y  D -1
h p ( x, y ) = 
 0 C  x  P or D  y  Q

Here P  A + C − 1; Q  B + D − 1
4/7/2019 60
 Summary

4/7/2019 61
 Summary

4/7/2019 62
 Summary

4/7/2019 63
 Summary

4/7/2019 64
 The Basic Filtering in the Frequency Domain

Why is the spectrum at

almost ±45 degree stronger
than the spectrum at other
directions?

4/7/2019 65
 The Basic Filtering in the Frequency Domain

► Modifying the Fourier transform of an image

► Computing the inverse transform to obtain the processed

result

g ( x, y ) = −1{H (u , v) F (u , v)}

F (u, v) is the DFT of the input image

H (u , v) is a filter function.

4/7/2019 66
 The Basic Filtering in the Frequency Domain

► In a filter H(u,v) that is 0 at the center of the transform

and 1 elsewhere, what’s the output image?

4/7/2019 67
 The Basic Filtering in the Frequency Domain

4/7/2019 68
 The Basic Filtering in the Frequency Domain

4/7/2019 69
 Zero-Phase-Shift Filters
−1
g ( x, y ) =  {H (u, v) F (u, v)}
F (u, v) = R(u, v) + jI (u, v)
−1
g ( x, y) =   H (u, v) R(u, v) + jH (u, v) I (u, v) 
Filters affect the real and imaginary parts equally,
and thus no effect on the phase.
These filters are called zero-phase-shift filters

4/7/2019 70
 Examples: Nonzero-Phase-Shift Filters

Even small
Phasechanges
angle is in the phase angle can ishave
Phase angle
dramaticmultiplied
(usually by undesirable) effects on the
multiplied by filtered
output 0.5 0.25

4/7/2019 71
 Summary:
Steps for Filtering in the Frequency Domain
1. Given an input image f(x,y) of size MxN, obtain the
padding parameters P and Q. Typically, P = 2M and Q = 2N.

2. Form a padded image, fp(x,y) of size PxQ by

appending the necessary number of zeros to f(x,y)

3. Multiply fp(x,y) by (-1)x+y to center its transform

4. Compute the DFT, F(u,v) of the image from step 3

5. Generate a real, symmetric filter function*, H(u,v), of

size PxQ with center at coordinates (P/2, Q/2)
*generate from a given spatial filter, we pad the spatial filter, multiply the expadded
array by (-1)x+y, and compute the DFT of the result to obtain a centered H(u,v).
4/7/2019 72
 Summary:
Steps for Filtering in the Frequency Domain
6. Form the product G(u,v) = H(u,v)F(u,v) using array
multiplication

7. Obtain the processed image

 
g p ( x, y) = real −1 G(u, v) (−1) x + y

8. Obtain the final processed result, g(x,y), by

extracting the MxN region from the top, left quadrant
of gp(x,y)

4/7/2019 73
 An Example:
Steps for Filtering in the Frequency Domain

4/7/2019 74
 Correspondence Between Filtering in the
Spatial and Frequency Domains (1)

Let H(u) denote the 1-D frequency domain Gaussian filter

- u 2 /2 2
H (u) = Ae

The corresponding filter in the spatial domain

h( x) = 2 Ae −2 2 2 x 2

1. Both components are Gaussian and real

2. The functions behave reciprocally

4/7/2019 75
 Correspondence Between Filtering in the
Spatial and Frequency Domains (2)
Let H (u ) denote the difference of Gaussian filter
- u 2 /212 - u 2 /2 22
H (u ) = Ae − Be
with A  B and  1   2

The corresponding filter in the spatial domain

−2 212 x 2 −2 2 22 x 2
h( x) = 2 1 Ae − 2 2 Ae

High-pass filter or low-pass filter ?

4/7/2019 76
 Correspondence Between Filtering in the
Spatial and Frequency Domains (3)

4/7/2019 77
 Correspondence Between Filtering in the
Spatial and Frequency Domains: Example

600x600

4/7/2019 78
 Correspondence Between Filtering in the
Spatial and Frequency Domains: Example

4/7/2019 79
 Generate H(u,v)

 f ( x, y) 0  x  599 and 0  y  599

f p ( x, y ) = 
 0 600  x  602 or 600  y  602

h( x, y) 0  x  2 and 0  y  2
hp ( x, y) = 
0 3  x  602 or 3  y  602

Here P  A(600) + C (3) − 1 = 602;

Q  B(600) + D(3) − 1 = 602.

4/7/2019 80
 Generate H(u,v)

1. Multiply hp ( x, y) by (-1) x+ y to center the frequency domain filter

2. Compute the forward DFT of the result in (1)

3. Set the real part of the resulting DFT to 0 to account for

parasitic real parts

4. Multiply the result by (-1)u +v , which is implicit when h( x, y )

was moved to the center of hp ( x, y).

4/7/2019 81
 Image Smoothing Using Filter Domain Filters:
ILPF

Ideal Lowpass Filters (ILPF)

1 if D(u , v)  D0
H (u , v) = 
0 if D(u, v)  D0

D0 is a positive constant and D(u, v) is the distance between a point (u, v)

in the frequency domain and the center of the frequency rectangle
2 1/2
D(u, v) = (u − P / 2) + (v − Q / 2) 
2

4/7/2019 82
 Image Smoothing Using Filter Domain Filters:
ILPF

4/7/2019 83
 ILPF Filtering Example

4/7/2019 84
 ILPF
Filtering
Example

4/7/2019 85
 The Spatial Representation of ILPF

4/7/2019 86
 Image Smoothing Using Filter Domain Filters:
BLPF
Butterworth Lowpass Filters (BLPF) of order n and
with cutoff frequency D0
1
H (u , v ) =
1 +  D(u, v) / D0 
2n

4/7/2019 87
4/7/2019 88
 The Spatial Representation of BLPF

4/7/2019 89
 Image Smoothing Using Filter Domain Filters:
GLPF

Gaussian Lowpass Filters (GLPF) in two dimensions is given

− D 2 ( u , v )/2 2
H (u, v) = e

By letting  = D0
− D 2 ( u , v )/2 D02
H (u , v ) = e

4/7/2019 90
 Image Smoothing Using Filter Domain Filters:
GLPF

4/7/2019 91
4/7/2019 92
 Examples of smoothing by GLPF (1)

4/7/2019 93
 Examples of smoothing by GLPF (2)

4/7/2019 94
 Examples of smoothing by GLPF (3)

4/7/2019 95
 Image Sharpening Using Frequency Domain
Filters

A highpass filter is obtained from a given lowpass filter

using

H HP (u, v) = 1 − H LP (u, v)

A 2-D ideal highpass filter (IHPL) is defined as

0 if D(u, v)  D0
H ( u , v) = 
1 if D(u, v)  D0

4/7/2019 96
 Image Sharpening Using Frequency Domain
Filters

A 2-D Butterworth highpass filter (BHPL) is defined as

1
H ( u , v) =
1 +  D0 / D(u, v) 
2n

A 2-D Gaussian highpass filter (GHPL) is defined as

− D 2 ( u , v )/2 D02
H (u, v) = 1 − e

4/7/2019 97
4/7/2019 98
 The Spatial Representation of Highpass
Filters

4/7/2019 99
 Filtering Results by IHPF

4/7/2019 100
 Filtering Results by BHPF

4/7/2019 101
 Filtering Results by GHPF

4/7/2019 102
 Using Highpass Filtering and Threshold for
Image Enhancement

BHPF
(order 4 with a cutoff
frequency 50)

4/7/2019 103
 The Laplacian in the Frequency Domain

H (u, v) = −4 2 (u 2 + v 2 )

H (u, v) = −4 2 (u − P / 2) 2 + (v − Q / 2) 2 ) 

= −4 2 D 2 (u, v)
The Laplacian image
2 f ( x, y) = −1 H (u, v) F (u, v)

Enhancement is obtained
g ( x, y) = f ( x, y ) + c2 f ( x, y ) c = -1
4/7/2019 104
 The Laplacian in the Frequency Domain

The enhanced image

−1
g ( x, y ) =  F (u, v) − H (u, v) F (u, v)
= −1 1 − H (u, v)  F (u, v)

 
= −1 1 + 4 2 D 2 (u, v)  F (u, v)

4/7/2019 105
 The Laplacian in the Frequency Domain

4/7/2019 106
 Unsharp Masking, Highboost Filtering and
High-Frequency-Emphasis Fitering

g mask ( x, y ) = f ( x, y ) − f LP ( x, y )

f LP ( x, y) = −1  H LP (u, v) F (u, v)

Unsharp masking and highboost filtering

g ( x, y ) = f ( x, y ) + k * g mask ( x, y )

 
g ( x, y ) = −1 1 + k * 1 − H LP (u, v)  F (u, v)
= −1 1 + k * H HP (u, v)  F (u, v)

4/7/2019 107
 Unsharp Masking, Highboost Filtering and
High-Frequency-Emphasis Fitering

g ( x, y ) = −1  k1 + k2 * H HP (u, v)  F (u, v)

k1  0 and k2  0

4/7/2019 108
 Gaussian Filter
D0=40

High-Frequency-Emphasis Filtering
Gaussian Filter
K1=0.5, k2=0.75

4/7/2019 109
 Homomorphic Filtering

f ( x, y) = i( x, y)r ( x, y)
 f ( x, y) 
= i ( x, y )   r ( x, y )  ?

z( x, y) = ln f ( x, y) = ln i( x, y) + ln r ( x, y)

 z ( x, y) = ln f ( x, y) = ln i( x, y) + ln r ( x, y)

Z (u, v) = Fi (u, v) + Fr (u, v)

4/7/2019 110
 Homomorphic Filtering

S (u, v) = H (u, v) Z (u, v)

= H (u, v) Fi (u, v) + H (u, v) Fr (u, v)
s ( x, y ) = −1 S (u, v)
= −1  H (u, v) Fi (u, v) + H (u , v) Fr (u , v)
= −1  H (u, v) Fi (u, v) + −1 H (u , v) Fr (u , v)
= i '( x, y ) + r '( x, y )

g ( x, y) = e s ( x, y ) = ei '( x, y ) er '( x, y ) = i0 ( x, y )r0 ( x, y )

4/7/2019 111
 Homomorphic Filtering

The illumination component of an image generally is

characterized by slow spatial variations, while the
reflectance component tends to vary abruptly

These characteristics lead to associating the low

frequencies of the Fourier transform of the logarithm of an
image with illumination the high frequencies with
reflectance.
4/7/2019 112
 Homomorphic Filtering


H (u, v) = ( H −  L ) 1 − e
− c  D2 ( u ,v )/ D02 
   +
  L

Attenuate the contribution

made by illumination and
amplify the contribution made
by reflectance

4/7/2019 113
  L = 0.25
Homomorphic
H = 2
Filtering
c =1
D0 = 80

4/7/2019 114
 Homomorphic Filtering

4/7/2019 115
 Selective Filtering

Non-Selective Filters:
operate over the entire frequency rectangle

Selective Filters
operate over some part, not entire frequency rectangle
• bandreject or bandpass: process specific bands
• notch filters: process small regions of the frequency
rectangle

4/7/2019 116
 Selective Filtering:
Bandreject and Bandpass Filters

H BP (u, v) = 1 − H BR (u, v)

4/7/2019 117
 Selective Filtering:
Bandreject and Bandpass Filters

4/7/2019 118
 Selective Filtering:
Notch Filters
Zero-phase-shift filters must be symmetric about the origin.
A notch with center at (u0, v0) must have a corresponding
notch at location (-u0,-v0).

Notch reject filters are constructed as products of highpass

filters whose centers have been translated to the centers of
the notches.
Q
H NR (u, v) =  H k (u, v) H − k (u, v)
k =1

where H k (u, v) and H - k (u, v) are highpass filters whose centers are
at (uk , vk ) and (-uk , -vk ), respectively.
4/7/2019 119
 Selective Filtering:
Notch Filters
Q
H NR (u, v) =  H k (u, v) H − k (u, v)
k =1

where H k (u, v) and H - k (u, v) are highpass filters whose centers are
at (uk , vk ) and (-uk , -vk ), respectively.

A Butterworth notch reject filter of order n

3  1  1 
H NR (u, v) =   2n   2n 
k =1 1 +  D0 k / Dk (u , v )   1 +  D0 k / D− k (u , v )  
  
2 1/2
Dk (u, v) = (u − M / 2 − uk ) + (v − N / 2 − vk ) 
2

2 1/2
D− k (u, v) = (u − M / 2 + uk ) + (v − N / 2 + vk ) 
2

4/7/2019 120
Examples:
Notch
Filters (1)

A Butterworth notch
reject filter D0 =3
and n=4 for all
notch pairs

4/7/2019 121
 Examples:
Notch Filters
(2)

4/7/2019 122
4/7/2019 123