Image Processing - Lecture-8

Digital Image Processing
Dr Mai Kamal-2022 1
Digital Iamge Processing

Course
By
Dr. Mai Kamal El Den
Lecturer, Computer Science Department,
Faculty of Computers and Artificial Intelligence,
Benha University
Dr Mai Kamal-2022 2
Digital Iamge Processing Course
Dr Mai Kamal-2022 3
Digital Iamge Processing

Course
Teams Link
https://teams.microsoft.com/l/team/19%3az6X9_t1JI66-
A0ZBSXtqyEeY4MaCpjrEIIm2ouxKVuU1%40thread.tacv2/conversations
?groupId=5376f567-69ce-4117-92c0-
21871798fba1&tenantId=ff4a48d6-4b5e-4fd3-8266-7eafc3e6e23e
Dr Mai Kamal-2022 4
Grading Formula
Weights of the Assessments

Final Term Examination 50
Final Oral Exam 10
Midterm Exam 15
Practical Exam 15
Semester Work 10
Dr Mai Kamal-2022 5
Text Book
R. Gonzalez, R. Woods.
Digital Ιmage Processing,
Prentice Hall, 2008.
Third Edition .
Dr Mai Kamal-2022 6
Digital Iamge Processing Course
Lecture 8
Image Enhancement in the frequency
Domain
Dr Mai Kamal DIP 2022
Chapter 4
Frequency Domain

Why Frequency Domain?
 Why Frequency Domain?
– Many times, image processing tasks can be best
performed in a domain other than the spatial
domain.
– Difficult problems in special domain , easy to
solve in the frequency domain by using Fourier
transform.
– The frequency domain is more convenient for
measurements of systems sensitivity to noise
and parameter variations.
Applications of Fourier
Transforms
 1-D Fourier transforms are used in Signal Processing

 2-D Fourier transforms are used in Image Processing
 3-D Fourier transforms are used in Computer Vision
 Applications of Fourier transforms in Image
processing: –
– Image enhancement,
– Image restoration,
– Image encoding / decoding,
– Image description

Introduction
Fourier Transform: a review
 The French mathematician Jean Baptiste Joseph
Fourier.
 Published Fourier series in 1822.
 Fourier’s ideas were met with skepticism.
 Fourier Series:
– Any periodical function can be
expressed as the sum of sines and/or
cosines of different frequencies,
each multiplied by a different
coefficient.
Introduction
 Basic ideas:
 A periodic function can
be represented by the
sum of sines/cosines
functions of different
frequencies, multiplied
by a different coefficient.
 This is known as
Fourier Transform

Introduction
 For an image, the output of the transformation is
the representation of the image in frequency
space, while the input image is the real space
equivalent.
 In the Fourier space image, each point
represents a particular frequency
contained in the real domain image.
Introduction
 A function expressed in either a Fourier Series
or transform can be reconstructed completely
via an inverse process with no loss of
information.
 This is one of the important characteristics of
these representations because they allow us to
work in the Fourier Domain and then return to
the original domain.

Frequency Filters – Fourier
Transform
 Fourier transform (FT)
– Decompose an image into its sine and
cosine components.
– Transform real space images into Fourier or
frequency space images.
– In a frequency space image, each point
represents a particular frequency contained
in the real domain image.

transform
 Fourier in Continuous Form:

What we need to work
with Image?
1. Fourier transform in Discrete Form.

2. Fourier transform in 2D not 1D.
The Discrete Fourier

Transform (DFT)
 The Discrete Fourier Transform of 𝑓 𝑥, 𝑦 ,
for 𝑥 = 0, 1, 2…𝑀-1 and 𝑦 = 0,1,2…𝑁-1,
denoted by 𝐹(𝑢, 𝑣), is given by the equation:
M 1 N 1
F (u, v)   f ( x, y)e  j 2 ( ux / M vy / N )
x 0 y 0
for 𝑢 = 0, 1, 2…𝑀-1 and 𝑣 = 0, 1, 2…𝑁-1.

The Inverse DFT
 It is really important to note that the
Fourier transform is completely
reversible The inverse DFT is given by:
M 1 N 1
1
f ( x, y )  
MN u 0 v 0
F (u, v)e j 2 ( ux / M vy / N )
 for x = 0, 1, 2…M-1 and y = 0, 1, 2…N-1

How to Perform Fourier

Transform ?

Comparison between Frequency
Domain & Spatial Domain
Filters
 Spatial (time) domain techniques are
operated directly on pixels of the image.
 Frequency domain techniques are based
on modifying the Fourier transform of an
image.
Comparison between Frequency

Domain FD& Spatial Domain SD
Filters
 Similar jobs can be done in the spatial and frequency
domains.
 Filtering in the spatial domain can be easier to
FD order𝒏 𝟒
SD order 𝒏𝟐 𝒍𝒐𝒈𝒏
understand.
 Filtering in the frequency domain can be much faster –
especially for large images.
 Easier to remove undesirable frequencies in the
frequency domain (i.e., frequency domain filtering).
 Faster to perform certain operations in the frequency
domain than in the spatial domain (e.g., convolution)
using the Fast Fourier Transform (FFT).
transform
 Frequency Filtering: consists of modifying the
FT of an input image and then finding the IFT
to get the output image.
 Mathematically its given by:
G(u,v)= H(u,v)F(u,v)
Steps of Frequency Domain

Filter
1. Multiply the input image by (−1) x+y

2. Compute F(u,v)
3. Multiply F(u,v) by a filter function
H(u,v)
4. Compute the inverse DFT
5. Obtain the real part
6. Multiply the real part by (−1) x+y

Fourier Transform: shift
 It is common to multiply input image
by (-1)x+y prior to computing the FT.
Shift
Fourier Transform: shift
 Example:

How to visualize DFT?
 Typically, we visualize the magnitude (spectrum):
|F(u,v)|
 Need to apply some “stretching” since the dynamic
range of |F(u,v)| is very large, e.g.:
original image |F(u,v)| |D(u,v)|

Subsequently
mapping the
values to [0, 255]
will further
improve
visualization!
without stretching after stretching


transform
– Example:
FT
(log)

Example
 For the following image shown:
(a) Compute the 2D-DFT for the marked pixels
𝑀−1 𝑁−1
𝑢𝑥 𝑣𝑦
−2𝜋𝑗( 𝑀 + 𝑁 )
𝐹 𝑢, 𝑣 = ෍ ෍ 𝑓(𝑥, 𝑦)𝑒
𝑥=0 𝑦=0
Example
 Answer:
M=4, N=4.
F(x,y)=f(0,0)=232.
F(x,y)=f(3,2)=1.
𝑀−1 𝑁−1
𝐹 0,0 = ෍ ෍ 𝑓(𝑥, 𝑦)𝑒 −2𝜋𝑗(0+0)

𝑥=0 𝑦=0
F(0,0)=(232+177+82+7+241+18+152+140+156+221+67
+3+100+45+1+103)=1745.

Example
 Answer:
M=4, N=4.
F(x,y)=f(0,0)=232.
F(x,y)=f(3,2)=1.
𝑀−1 𝑁−1
3𝑥 2𝑦
−2𝜋𝑗( + )
𝐹 3,2 = ෍ ෍ 𝑓(𝑥, 𝑦)𝑒 4 4
𝑥=0 𝑦=0
𝑭 𝟑, 𝟐
3𝑥 𝑦 3𝑥
−𝜋𝑗( +𝑦)
= σ𝑀−1 𝑁−1
𝑥=0 σ𝑦=0 𝑓(𝑥, 𝑦)𝑒
−2𝜋𝑗( 4 + 2 )
= σ𝑀−1 𝑁−1
𝑥=0 σ𝑦=0 𝑓(𝑥, 𝑦)𝑒 2
Example
 Answer:
3𝑥
−𝜋𝑗( +𝑦)
𝐹 3,2 =σ𝑀−1 σ𝑁−1
𝑥=0 𝑦=0 𝑓(𝑥, 𝑦)𝑒 2
3
F(3,2)=(232+177𝑒 −𝜋𝑗 +82𝑒 −2𝜋𝑗 +7𝑒 −3𝜋𝑗 +241𝑒 −𝜋𝑗(2) +18
5 7 9
−𝜋𝑗(2) −𝜋𝑗(2) −𝜋𝑗(2)
𝑒 +152𝑒 +140𝑒 +156𝑒 −𝜋𝑗(3) +221𝑒 −𝜋𝑗(4)
9 11 13
+67𝑒 −𝜋𝑗(5) +3𝑒 −𝜋𝑗(6) +100𝑒 −𝜋𝑗(2) +45𝑒 −𝜋𝑗( 2 ) +1𝑒 −𝜋𝑗( 2 )
15
−𝜋𝑗( 2 )
+103𝑒 )= 232 – 177 + 82 – 7 +….

Filters in Frequency Domain
Frequency Filters
 Relationship and comparison with spatial
filters
– Spatial Filtering:
g ( x, y)  h( x, y) f ( x, y)
– Frequency Filtering:
G(u, v)  H (u, v) F (u, v)
–
h( x, y)  H (u, v)
Correspondence between FD
and SD filtering:
 Special Domain (SD): filtering is given by discrete
convolution and Correlation.
 Frequency Domain (FD): Discrete convolution in FD

is equivalent to multiplication
 Filtering can be done in either domain.
SD VS FD filters
Low pass High pass
Average filters Laplace filters

transform
 Image in frequency domain consist of:

– white pixels – low frequencies (details in
image).
– black pixels– high frequencies (not details
in image).
Frequency Filters
Frequency
Filter
smoothing Sharping

Frequency Filters
 Frequencies in an image correspond to
the rate of change in pixel values
– High frequencies
• rapid changes of gray level values.
– Low frequencies
• slow changes of gray level values.
Frequency Filters
1. Lowpass filters
– attenuate high frequencies while “passing”
low frequencies.
– Remove high frequency and keep only low frequency.
2. Highpass filters
– attenuate low frequencies while “passing”
high frequencies.
– Remove low frequency and keep high frequency.

Low pass & High pass filters in
FD
 Low-pass filter: A filter that eliminating high frequencies
while passing low frequencies.- used for blurring
(smoothing).
 High-pass filter: A filter that eliminating low frequencies
while passing high frequencies. used for sharpening.
1) Low pass – image smoothing
 Image smoothing using Frequency Domain

filters
Noise is usually high frequency.
Hence noise removal is usually termed smoothing or
blurring.
Smoothing is achieved by lowpass filters (LPFs).
3 types of LPFs will be studied i.e. Ideal LPF,
Butterworth LPF and Gaussian LPF.
 Note/Caution: Image details and edges have high

frequency characteristics.

1) Low pass – image smoothing
 Smoothing filters:
1.1) Ideal low pass filters (ILPF)
1 if D(u, v)  D0
H (u, v)  
0 if D(u, v)  D0
1.2) Butterworth low pass filters (BLPF)

1
H (u, v) 
1  [ D(u, v) / D0 ]2 n
1.3) Gaussian low pass filters (GLPF)
 D 2 ( u ,v ) / 2 D0 2
H (u , v)  e
1.1) 2-D Ideal low pass filter

ILPF
 The simplest lowpass filter is ILPF, it (cuts off) all high
frequency components that are at distance greater than a
specified distance: D0 from the center. N
1 if D(u, v)  D0
H (u, v)   M
Where 0 if D(u, v)  D0
– Image size: MxN.
– The size of filter in FD have the same size of the image itself.
– D(u,v) is the distance from (u,v) to the center of the frequency rectangle
– Center of the frequency rectangle: (u,v) = (M/2, N/2)
– Distance to the center: D(u,v) = [(u – M/2)2 + (v – N/2)2]1/2

How does ILPF function works?
 Simply cut off all high frequency components that are a

specified distance 𝐷0 from the origin of the transform.
 Changing the distance changes the behaviour of the filter.
How does ILPF function works?
 The center of the circle contains the urgent information

or details.
 With the increase of D0 the image become better.
Radius total image power %
5 92.0
15 94.6
30 96.4
80 98
230 pixels 99.5

Ideal Low Pass Filter Results
Result of filtering
Original with ideal low
image pass filter of
Ringing is a radius 5
characteristics
of ideal Filter
Result of filtering
Result of filtering with ideal low
with ideal low pass filter of
pass filter of radius 30
radius 15
Result of filtering Result of filtering

with ideal low with ideal low
pass filter of pass filter of
radius 80 radius 230
• Ideal low pass filtering is not very practical but they can be implemented
on a computer to study their behavior.
Blurring and Ringing Feature of

ILPFs:
 The ILPF has sharp cutoffs or discontinuities which cause
ringing.
 The ILPF has a sink function behaviour in the Special
Domain.
 The center of lobe is the cause for blurring but the outer
smaller lobes cause ringing.
 We want to achieve blurring with little ringing.

1.2) Butterworth Low Pass
Filter - BLPF:
 ILPF transfer function has sharp discontinuities.
 BLPF transfer function does not have sharp
discontinuities in order to reduce ringing.
 The transfer function of a Butterworth lowpass

filter of order n with cutoff frequency at distance
D0 from the origin is defined as:
1
H (u, v) 
1  [ D(u, v) / D0 ]2 n
Butterworth low pass filter -

BLPF:
1
H (u, v) 
1  [ D(u, v) / D0 ]2 n
Note:
H(U,V) =0.50 (50% from its maximum value of 1) when D(u,v) =D0

Butterworth Lowpass Filter
(cont...)
Result of filtering
Original image with Butterworth
filter of order 2
and cutoff radius
Ringing is not visible 5
in any of these
images.
Result of filtering Result of filtering
with Butterworth with Butterworth
filter of order 2 and filter of order 2
cutoff radius 15 and cutoff radius
30
Result of filtering
Result of filtering
with Butterworth
with Butterworth
filter of order 2
filter of order 2
and cutoff radius
and cutoff radius
230
80
1.3) Gaussian Lowpass Filters -

GLPF
 The transfer function of a Gaussian lowpass filter
is defined as  D 2 ( u ,v ) / 2 D 2
H (u , v )  e 0
Note:
The IFT (INVERSE FOURIOUR) of Gaussian is also Gaussian.
Hence there is no ringing effect.

Gaussian low pass filter -
GLPF:
Note:
H(U,V) =0.667 when D(u,v) =D0
Gaussian Lowpass Filters

(cont…)
Result of
Original filtering with
image Gaussian filter
No ringing with cutoff
radius 5
Result of Result of filtering

filtering with with Gaussian
Gaussian filter filter with cutoff
with cutoff radius 30
radius 15
Result of Result of
filtering with filtering with
Gaussian filter Gaussian filter
with cutoff with cutoff
radius 85 radius 230
Lowpass Filters Compared
Result of Result of
ideal low Butterworth
pass filter of filter of order
radius 15 2 and cutoff
radius 15
Result of
filtering with
Gaussian
filter with
cutoff radius
15
Lowpass Filtering Examples

 A low pass Gaussian filter is used to
connect broken text
Machine recognition systems have

difficulty in reading broken characters.
GLPF with D0 = 80

Lowpass Filtering Examples
(cont…)
 Different lowpass Gaussian filters used
to remove blemishes in a photograph.
Sharping Frequency Filter

2)High pass - image Sharpening
 Blurring (smoothing) is achieved by attenuating the
HF(high frequency) components of DFT of an image.
(low pass filters).
 Sharpening is achieved by attenuating the LF (low

frequency) components of DFT of an image. Where
Edges and fine detail in images are associated with high
frequency components (high pass filters).
High pass filters – only pass the high frequencies, drop the
low ones. High pass frequencies are precisely the reverse
of low pass filters, so:
Hhp(u, v) = 1 – Hlp(u, v)
2.1) Ideal High Pass Filters

The ideal high pass filter is given as:
0 if D(u, v)  D0
H (u, v)  
1 if D(u, v)  D0
where D0 is the cut off distance as before

Ideal High Pass Filters (cont…)
Results of ideal high Results of ideal high Results of ideal high

pass filtering with D0 pass filtering with D0 pass filtering with D0
= 15 = 30 = 80
2.2) Butterworth High Pass

Filters
The Butterworth high pass filter is given
as: 1
H (u, v) 
1  [ D0 / D(u, v)]2 n
where n is the order and D0 is the cut off distance
as before

Butterworth High Pass Filters
(cont…)
Results of Results of
Butterworth Butterworth
high pass high pass
filtering of filtering of
order 2 with order 2 with
D0 = 15 D0 = 80
Results of Butterworth high pass

filtering of order 2 with D0 = 30
2.3) Gaussian High Pass Filters

The Gaussian high pass filter is given as:
 D 2 ( u ,v ) / 2 D0 2
H (u , v)  1  e
where D0 is the cut off distance as before

Gaussian High Pass Filters
(cont…)
Results of Results of
Gaussian Gaussian
high pass high pass
D0 = 15 D0 = 80
Results of Gaussian high pass

filtering with D0 = 30
Highpass Filter Comparison
BF represents a
transition
between the
sharpness
of the IF and the
smoothness of
the GF

Highpass Filter Comparison
Results of ideal Results of Butterworth Results of Gaussian high

high pass filtering high pass filtering of pass filtering with D0 =
with D0 = 15 order 2 with D0 = 15 15
Next Lecture
LEC.9:IMAGE RESTORATION
(NOISE REMOVAL)


Image Processing - Lecture-8

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Image Processing - Lecture-8

Uploaded by

Copyright:

Available Formats

Digital Image Processing

Digital Iamge Processing

Digital Iamge Processing

Weights of the Assessments

Dr Mai Kamal DIP 2022

Dr Mai Kamal DIP 2022

Dr Mai Kamal DIP 2022

 1-D Fourier transforms are used in Signal Processing

Dr Mai Kamal DIP 2022

Dr Mai Kamal DIP 2022

Dr Mai Kamal DIP 2022

Dr Mai Kamal DIP 2022

Dr Mai Kamal DIP 2022

Frequency Filters – Fourier

Dr Mai Kamal DIP 2022

1. Fourier transform in Discrete Form.

Dr Mai Kamal DIP 2022

The Discrete Fourier

for 𝑢 = 0, 1, 2…𝑀-1 and 𝑣 = 0, 1, 2…𝑁-1.

Dr Mai Kamal DIP 2022

 for x = 0, 1, 2…M-1 and y = 0, 1, 2…N-1

How to Perform Fourier

Dr Mai Kamal DIP 2022

Dr Mai Kamal DIP 2022

Comparison between Frequency

Dr Mai Kamal DIP 2022

Steps of Frequency Domain

1. Multiply the input image by (−1) x+y

Dr Mai Kamal DIP 2022

Dr Mai Kamal DIP 2022

Fourier Transform: shift

Dr Mai Kamal DIP 2022

original image |F(u,v)| |D(u,v)|

without stretching after stretching

Frequency Filters – Fourier

Dr Mai Kamal DIP 2022

Dr Mai Kamal DIP 2022

𝐹 0,0 = ෍ ෍ 𝑓(𝑥, 𝑦)𝑒 −2𝜋𝑗(0+0)

Dr Mai Kamal DIP 2022

Dr Mai Kamal DIP 2022

Dr Mai Kamal DIP 2022

Dr Mai Kamal DIP 2022

 Frequency Domain (FD): Discrete convolution in FD

 Filtering can be done in either domain.

Dr Mai Kamal DIP 2022

Average filters Laplace filters

Dr Mai Kamal DIP 2022

 Image in frequency domain consist of:

Dr Mai Kamal DIP 2022

Dr Mai Kamal DIP 2022

Dr Mai Kamal DIP 2022

Dr Mai Kamal DIP 2022

Dr Mai Kamal DIP 2022

1) Low pass – image smoothing

 Image smoothing using Frequency Domain

 Note/Caution: Image details and edges have high

Dr Mai Kamal DIP 2022

1.2) Butterworth low pass filters (BLPF)

Dr Mai Kamal DIP 2022

1.1) 2-D Ideal low pass filter

Dr Mai Kamal DIP 2022

 Simply cut off all high frequency components that are a