Date post: | 04-Apr-2018 |
Category: |
Documents |
Upload: | umerahmed300 |
View: | 236 times |
Download: | 0 times |
of 47
7/30/2019 Chap4 Frequency Domain
1/47
Chapter 4:
Frequency Domain Processing
Image Transformations
1
7/30/2019 Chap4 Frequency Domain
2/47
7/30/2019 Chap4 Frequency Domain
3/47
Introduction
Although we discuss other transforms in some
detail in this chapter, we emphasize the Fourier
transform because of its wide range of
applications in image processing problems.
3
7/30/2019 Chap4 Frequency Domain
4/47
Fourier Transform (1-D)
uXuXsin
AXuF
euXsinu
Adxux2jexpxfuF uXj
4
7/30/2019 Chap4 Frequency Domain
5/47
Fourier Transform (2-D)
vYvYsin
uX
uXsinAXYv,uFdydxvyux2jexpy,xf)v,u(F
5
7/30/2019 Chap4 Frequency Domain
6/47
Discrete Fourier Transform
In the two-variable case the discrete Fourier transform pair is
6
7/30/2019 Chap4 Frequency Domain
7/47
Discrete Fourier Transform
When images are sampled in a squared array, i.e. M=N,
we can write
7
7/30/2019 Chap4 Frequency Domain
8/47
Discrete
Fourier
Transform
Examples
At all of theseexamples, the Fourier
spectrum is shifted
from top left corner to
the center of the
frequency square.
8
7/30/2019 Chap4 Frequency Domain
9/47
Discrete
Fourier
Transform
Examples
At all of theseexamples, the Fourier
spectrum is shifted
from top left corner to
the center of the
frequency square.
9
7/30/2019 Chap4 Frequency Domain
10/47
Discrete
Fourier
Transform
Display
At all of theseexamples, the Fourier
spectrum is shifted
from top left corner to
the center of the
frequency square.
10
7/30/2019 Chap4 Frequency Domain
11/47
Discrete Fourier Transform
Example
Main Image (Gray Level) DFT of Main image
(Fourier spectrum)
Logarithmic scaled
of the Fourier spectrum
11
7/30/2019 Chap4 Frequency Domain
12/47
Discrete Fourier Transform
(Properties)
Separability
1N
0v
1N
0u
1N
0y
1N
0x
N/vy2jexpv,uFN/ux2jexpN
1y,xf
N/vy2jexpy,xfN/ux2jexpN
1v,uF
The discrete Fourier transform pair can be expressed in the seperable forms:
12
7/30/2019 Chap4 Frequency Domain
13/47
Discrete Fourier Transform
(Properties)
Translation
N/vyux2jexpv,uFyy,xxfand
vv,uuFN/yvxu2jexpy,xf
0000
0000
The translation properties of the
Fourier transform pair are :
13
7/30/2019 Chap4 Frequency Domain
14/47
Discrete Fourier Transform
(Properties)
Periodicity
The discrete Fourier transform and its
inverse areperiodic with period N; that is,
F(u,v)=F(u+N,v)=F(u,v+N)=F(u+N,v+N)
If f(x,y) is real, the Fourier transform also
exhibits conjugate symmetry:
F(u,v)=F*(-u,-v)
Or, more interestingly:
|F(u,v)|=|F(-u,-v)|
14
7/30/2019 Chap4 Frequency Domain
15/47
Discrete Fourier Transform
(Properties)
Rotation
sinvcosu
sinrycosrx
If we introduce the polar coordinates
Then we can write:
00 ,F,rf
In other words, rotating F(u,v)
rotates f(x,y) by the same angle.
15
7/30/2019 Chap4 Frequency Domain
16/47
Discrete Fourier Transform
(Properties)
Convolution
v,uG*v,uFy,xgy,xfand
v,uGv,uF)y,x(g*y,xf
The convolution theorem in
two dimensions is expressed
by the relations :
Note :
ddy,xg,fy,xg*y,xf
16
7/30/2019 Chap4 Frequency Domain
17/47
Discrete Fourier Transform
(Properties)
Correlation
dxgfxgxf
*
The correlation of two continuous
functions f(x) and g(x) is defined
by the relation
So we can write:
v,uGv,uFy,xgy,xf
and
v,uGv,uFy,xgy,xf
*
*
17
7/30/2019 Chap4 Frequency Domain
18/47
Discrete
Fourier
Transform
Sampling
(Properties)
1-D
f(x) : a given function
F(u): Fourier Transform of f(x)
which is band-limited
s(x) : sampling function
S(u): Fourier Transform of s(x)
G(u): window for recovery of the
main function F(u) and f(x).
Recovered f(x) from sampled data
18
7/30/2019 Chap4 Frequency Domain
19/47
Discrete
Fourier
Transform
Sampling
0000 y,xfdydxyy,xxy,xf
(Properties)
2-D
The sampling process for 2-Dfunctions can be formulated
mathematically by making use
of the 2-D impulse function
(x,y), which is defined as
A 2-D sampling function is
consisted of a train of impulses
separated x units in the x
direction and y units in the y
direction as shown in the figure.
19
7/30/2019 Chap4 Frequency Domain
20/47
Discrete
Fourier
Transform
Sampling
0
1
v,uG
(Properties)
2-D
If f(x,y) is band limited (that is, its
Fourier transform vanishes outsidesome finite region R) the result of
covolving S(u,v) and F(u,v) might
look like the case in the case shown
in the figure. The function shown is
periodic in two dimensions.
(u,v) inside one of the rectangles
enclosing R
elsewhere
The inverse Fourier transform of
G(u,v)[S(u,v)*F(u,v)] yields f(x,y).
20
7/30/2019 Chap4 Frequency Domain
21/47
The Fast Fourier Transform (FFT) Algorithm21
7/30/2019 Chap4 Frequency Domain
22/47
Other Seperable Image Transforms
AFAT
BTBF
AB
BAFABBTB
1
1N
0x
1N
0y
v,u,y,xgy,xfv,uT
Where F is the NN image matrix,
A is an NN symmetric transformation matrix
T is the resulting N
N transform.
If the kernel g(x,y,u,v) is seperable and symmetric,
also may be expressed in matrix form:
And for inverse transform we have:
22
7/30/2019 Chap4 Frequency Domain
23/47
Discrete Cosine Transform23
7/30/2019 Chap4 Frequency Domain
24/47
Discrete Cosine Transform (DCT)
Each block consists
of 44 elements,
corresponding to x
and y varying from 0to 3. The highest
value is shown in
white. Other values
are shown in grays,
with darker meaning
smaller.
24
7/30/2019 Chap4 Frequency Domain
25/47
Discrete Cosine Transform
Example
Main Image (Gray Level) DCT of Main image
(Cosine spectrum)
Logarithmic scaled
of the Cosine spectrum
25
7/30/2019 Chap4 Frequency Domain
26/47
DCT and Fourier Transform26
7/30/2019 Chap4 Frequency Domain
27/47
DCT Example27
7/30/2019 Chap4 Frequency Domain
28/47
Blockwise DCT Example28
7/30/2019 Chap4 Frequency Domain
29/47
Comparison Of Various Transforms29
7/30/2019 Chap4 Frequency Domain
30/47
Comparison Of Various Transforms30
7/30/2019 Chap4 Frequency Domain
31/47
The DFT and Image Processing
To filter an image in the frequency domain:1. Compute F(u,v) the DFT of the image
2. Multiply F(u,v) by a filter functionH(u,v)
3. Compute the inverse DFT of the result
ImagestakenfromGonzalez&Woods,DigitalImageProcessing(2002)
32
7/30/2019 Chap4 Frequency Domain
32/47
Some Basic Frequency Domain Filters
ImagestakenfromGonzalez&Woods,DigitalImageProcessing(2002) Low Pass Filter
High Pass Filter
7/30/2019 Chap4 Frequency Domain
33/47
Smoothing Frequency Domain Filters
Smoothing is achieved in the frequency domain by
dropping out the high frequency components
The basic model for filtering is:
G(u,v) = H(u,v)F(u,v)where F(u,v) is the Fourier transform of the imagebeing filtered andH(u,v) is the filter transformfunction
Low pass filters only pass the low frequencies,drop the high ones
7/30/2019 Chap4 Frequency Domain
34/47
Ideal Low Pass Filter
Simply cut off all high frequency components thatare a specified distance D0 from the origin of the
transform
changing the distance changes the behaviour of
the filter
ImagestakenfromGonzalez&Woods,DigitalImageProcessing(2002)
7/30/2019 Chap4 Frequency Domain
35/47
Ideal Low Pass Filter (cont)
The transfer function for the ideal low pass filter
can be given as:
whereD(u,v) is given as:
0
0
),(if0
),(if1
),( DvuD
DvuD
vuH
2/122 ])2/()2/[(),( NvMuvuD
7/30/2019 Chap4 Frequency Domain
36/47
Ideal Low Pass Filter (cont)
Above we show an image, its Fourier spectrum
and a series of ideal low pass filters of radius 5,
15, 30, 80 and 230 superimposed on top of it
ImagestakenfromGonzalez&Woods,DigitalImageProcessing(2002)
7/30/2019 Chap4 Frequency Domain
37/47
Ideal Low Pass Filter (cont)
ImagestakenfromGonzalez&Woods,DigitalImageProcessing(2002)
Original
image
Result of filtering
with ideal low pass
filter of radius 5
Result of filtering
with ideal low pass
filter of radius 30
Result of filtering
with ideal low pass
filter of radius 230
Result of filtering
with ideal low pass
filter of radius 80
Result of filtering
with ideal low pass
filter of radius 15
7/30/2019 Chap4 Frequency Domain
38/47
Gaussian Lowpass Filters
The transfer function of a Gaussian lowpass
filter is defined as:
ImagestakenfromGonzalez&Woods,DigitalImageProcessing(2002)
20
2 2/),(
),(
DvuDevuH
7/30/2019 Chap4 Frequency Domain
39/47
Gaussian Lowpass Filters (cont)
ImagestakenfromGonzalez&Woods,DigitalImageProcessing(2002)
Original
image
Result of filtering
with Gaussian
filter with cutoff
radius 5
Result of filtering
with Gaussian filter
with cutoff radius 30
Result of filtering
with Gaussian
filter with cutoff
radius 230
Result of
filtering with
Gaussian filter
with cutoff
radius 85
Result of filtering
with Gaussian
filter with cutoff
radius 15
7/30/2019 Chap4 Frequency Domain
40/47
Lowpass Filtering Examples
A low pass Gaussian filter is used to connect
broken text
ImagestakenfromGonzalez&Woods,Digital
ImageProcessing(2002)
7/30/2019 Chap4 Frequency Domain
41/47
Sharpening in the Frequency Domain
Edges and fine detail in images are associated
with high frequency components
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) = 1Hlp(u, v)
7/30/2019 Chap4 Frequency Domain
42/47
Ideal High Pass Filters
The ideal high pass filter is given as:
where D0 is the cut off distance as before
0
0
),(if1
),(if0),(
DvuD
DvuDvuH
ImagestakenfromGonzalez&Woods,Digital
ImageProcessing(2002)
7/30/2019 Chap4 Frequency Domain
43/47
Ideal High Pass Filters (cont)
Results of ideal
high pass filtering
withD0 = 15
Results of ideal
high pass filtering
withD0 = 30
Results of ideal
high pass filtering
withD0 = 80
ImagestakenfromGonzalez&Woods,Digital
ImageProcessing(2002)
7/30/2019 Chap4 Frequency Domain
44/47
Gaussian High Pass Filters
The Gaussian high pass filter is given as:
whereD0 is the cut off distance as before
20
2 2/),(1),(
DvuDevuH
ImagestakenfromGonzalez&Woods,Digital
ImageProcessing(2002)
7/30/2019 Chap4 Frequency Domain
45/47
Fast Fourier Transform
The reason that Fourier based techniques have
become so popular is the development of the
Fast Fourier Transform (FFT) algorithm
Allows the Fourier transform to be carried out in
a reasonable amount of time
Reduces the amount of time required to
perform a Fourier transform by a factor of 100
600 times!
7/30/2019 Chap4 Frequency Domain
46/47
Frequency Domain Filtering & Spatial Domain
Filtering
Similar jobs can be done in the spatial and
frequency domains
Filtering in the spatial domain can be easier to
understand
Filtering in the frequency domain can be much
faster especially for large images
7/30/2019 Chap4 Frequency Domain
47/47
References
Digital image processing by Rafael C.
Gonzalez and Richard E. Woods
Digital image processing by Jain
Image communication I by Bernd Girod
Lecture 3, DCS339/AMCM053 by Pengwei
Hao, University of London
47