digital image processing - sjtumin.sjtu.edu.cn/files/courses/dip16/lecture04.pdf · 2018-04-09 ·...
TRANSCRIPT
Digital Image Processing
Hongkai Xiong
熊红凯http://ivm.sjtu.edu.cn
电子工程系上海交通大学
2016
http://ivm.sjtu.edu.cn
Topic
• Simple review of Fourier transform
• Filtering in frequency domain
http://ivm.sjtu.edu.cn
Simple review of Fourier transform
http://ivm.sjtu.edu.cn
The Fourier Transform
• The Fourier transform converts a signal from spatial domain to the frequency domain.
• The Continuous Fourier Transform of a one-dimensional continuous function f(t) is defined as
f(t) F(s)Fourier
Transform
dtetfsFtf stj 2)()()}({F
http://ivm.sjtu.edu.cn
• The inverse Fourier transform
dsesFsF stj 21 )()}({F
F(s) f(t)Inverse Fourier
Transform
http://ivm.sjtu.edu.cn
• If is a sequence of length N, then its discrete Fourier transform (DFT) is given by
• And the inverse DFT is given by
Where are indices.
1 2
0
1, 0,1, 1
nN j iN
n i
i
F f e n NN
1 2
0
, 0,1, , 1iN j nN
i n
n
f F e i N
}{ if
1,0 Nni
The Discrete Fourier Transform
http://ivm.sjtu.edu.cn
Properties of the Fourier Transform
• The Addition Theorem
▫ If and , then
▫ And take it as an axiom that for any real number c
▫ This implies that Fourier transform is a linear transform.
)()}({ sFtf F )()}({ sGtg F
)()()}()({ sGsFtgtf F
)()}({ scFtcf F
http://ivm.sjtu.edu.cn
The Addition Theorem
• Addition-5 0 5
0
2
4
f(t)
-2 -1 0 1 2
-200
0
200
F(s
)
-5 0 50
2
4
g(t
)
-2 -1 0 1 2
-200
0
200
G(s
)
-5 0 50
2
4
6
f(t)
+g
(t)
-2 -1 0 1 2
-200
0
200F
(s)+
G(s
)
http://ivm.sjtu.edu.cn
The Shift Theorem
• Time shift:
• Frequency shift:
When ,
)()}({ 2 sFeatf asj F
02 /
0{ ( ) } ( )j s t N
f t e F s s
F
0 / 2s N
{ ( )( 1) } ( / 2)tf t F s N Fback
http://ivm.sjtu.edu.cn
The Convolution Theorem
• The Fourier transform of the convolution of two functions is the product of the Fourier transforms of these two functions
)()()}()({ sGsFtgtf F
http://ivm.sjtu.edu.cn
The Two-Dimensional DFT
• The discrete Fourier transform of an image of size M*N is given by
• Inverse Fourier transform
1 1
2
0 0
1, ,
M Nj ux M vy N
x y
F u v f x y eMN
1 1
2
0 0
, ,M N
j ux M vy N
u v
f x y F u v e
http://ivm.sjtu.edu.cn
http://ivm.sjtu.edu.cn
• Fourier spectrum
• Phase angle
• Power spectrum
where R(u, v) and I(u, v) are the real and imaginary parts of F(u, v), respectively
2 2, , ,F u v R u v I u v
-1
,, tan
,
I u vu v
R u v
2
2 2
, ,
, ,
P u v F u v
R u v I u v
http://ivm.sjtu.edu.cn
• For display, it is common practice to shift the original point to the center (M/2, N/2). According to the shift theorem, we need to multiply the input image function by (-1)x+y
prior to computing the Fourier transform.:
, 1 ( / 2, / 2)x y
f x y F u M v N
http://ivm.sjtu.edu.cn
• What’s the value of the transform at (u, v)=(0, 0)?
• The average of f(x, y).
1 1
0 0
10,0 ,
M N
x y
F f x yMN
http://ivm.sjtu.edu.cn
A simple example
http://ivm.sjtu.edu.cn
Translation and Rotation
http://ivm.sjtu.edu.cn
The 2-D Convolution Theorem
• 2-D circular convolution
• The 2-D convolution theorem is given by
and, conversely,
1 1
0 0
, , ,,M N
m n
h x y f m n h x m y nf x y
, ,, ,h x y F uf y vx v H u
, ,, ,h x y F u vf y vx H u
http://ivm.sjtu.edu.cn
2D convolution theorem example
*
f(x,y)
h(x,y)
g(x,y)
|F(sx,sy)|
|H(sx,sy)|
|G(sx,sy)|
http://ivm.sjtu.edu.cn
Curious Things about FT on Images
Example:cheetah pic
http://ivm.sjtu.edu.cn
Curious Things about FT on Images
This is the magnitude transform of the cheetah pic
http://ivm.sjtu.edu.cn
Curious Things about FT on Images
This is the phase transform of the cheetah pic
http://ivm.sjtu.edu.cn
Curious Things about FT on Images
Example:zebra pic
http://ivm.sjtu.edu.cn
Curious Things about FT on Images
This is the magnitude transform of the zebra pic
http://ivm.sjtu.edu.cn
Curious Things about FT on Images
This is the phase transform of the zebra pic
http://ivm.sjtu.edu.cn
Curious Things about FT on Images
Reconstruction with zebra phase, cheetah magnitude
http://ivm.sjtu.edu.cn
Curious Things about FT on Images
Reconstruction with cheetah phase, zebra magnitude
http://ivm.sjtu.edu.cn
Filtering in frequency domain
http://ivm.sjtu.edu.cn
http://ivm.sjtu.edu.cn
Steps for filtering in frequency domain
1) Multiply the input image by (-1)x+y to center the transform
2) Zero padding in case of aliasing
3) Compute F(u, v), the DFT of the image from (2)
4) Multiply F(u, v) by a filter function H(u, v)
5) Compute the inverse DFT of the result in (4)
6) Obtain the real part of the result in (5)
7) Multiply the result in (6) by (-1)x+y
8) Crop the image
http://ivm.sjtu.edu.cn
Correspondence Between Filtering in the
Spatial and Frequency Domains
• h(x, y) is a spatial filter, and can be obtained from the response of a frequency domain filter to an impulse.
• h(x, y) sometimes is referred to as the impulse response of H(u, v)
http://ivm.sjtu.edu.cn
Gaussian low-pass filter and high-pass filter in
the frequency and spatial domain
http://ivm.sjtu.edu.cn
A Gaussian kernel gives less weight to pixels further from the center of the window
This kernel is an approximation of a Gaussian function:
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 90 90 90 90 90 0 0
0 0 0 90 90 90 90 90 0 0
0 0 0 90 90 90 90 90 0 0
0 0 0 90 0 90 90 90 0 0
0 0 0 90 90 90 90 90 0 0
0 0 0 0 0 0 0 0 0 0
0 0 90 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
1 2 1
2 4 2
1 2 1
Correspondence Between Filtering in the
Spatial and Frequency Domains
http://ivm.sjtu.edu.cn
Correspondence Between Filtering in the
Spatial and Frequency Domains
http://ivm.sjtu.edu.cn
Correspondence Between Filtering in the
Spatial and Frequency Domains
http://ivm.sjtu.edu.cn
Image Smoothing
• Ideal low-pass filters
0
0
1 if ,,
0 if ,
D u v DH u v
D u v D
http://ivm.sjtu.edu.cn
• Ideal low-pass filters
0
0
1 if ,,
0 if ,
D u v DH u v
D u v D
Image Smoothing
http://ivm.sjtu.edu.cn
• Ideal low-pass filters
Image Smoothing
http://ivm.sjtu.edu.cn
• Butterworth low-pass filters
2
0
1,
1 ,n
H u vD u v D
Image Smoothing
http://ivm.sjtu.edu.cn
• Butterworth low-pass filters
2
0
1,
1 ,n
H u vD u v D
Image Smoothing
http://ivm.sjtu.edu.cn
http://ivm.sjtu.edu.cn
• Gaussian low-pass filters
2 2, 2,
D u vH u v e
Image Smoothing
http://ivm.sjtu.edu.cn
http://ivm.sjtu.edu.cn
• Gaussian low-pass filters
Image Smoothing
http://ivm.sjtu.edu.cn
Image Sharpening
• A high-pass filter is obtained from a given low-pass filter
, 1 ,HP LPH u v H u v
http://ivm.sjtu.edu.cn
Image Sharpening
http://ivm.sjtu.edu.cn
Image Sharpening
http://ivm.sjtu.edu.cn
• Ideal high-pass filters
0
0
0 if ,,
1 if ,
D u v DH u v
D u v D
Image Sharpening
http://ivm.sjtu.edu.cn
• Butterworth high-pass filters
2
0
1,
1 ,n
H u vD D u v
Image Sharpening
http://ivm.sjtu.edu.cn
• Gaussian high-pass filters
,,
2 2D u v 2H u v 1 e
Image Sharpening
http://ivm.sjtu.edu.cn
• The Laplacian in the frequency domain
, 2 2 2H u v 4 u v
, , ,12 H uf v F vx y u
Image Sharpening
http://ivm.sjtu.edu.cn
Sharpening Spatial Filters
• The Laplacian
▫ Laplacian operator
▫ X-direction
▫ Y-direction
2 22
2 2
f ff
x y
2
2
ff x 1, y f x 1, y 2 f x, y
x
2
2
ff x, y 1 f x, y 1 2 f x, y
y
http://ivm.sjtu.edu.cn
Sharpening Spatial Filters
• The discrete Laplacian of two variables
2 f x, y f x 1, y f x 1, y f x, y 1 f x, y 1 4 f x, y
http://ivm.sjtu.edu.cn
• Enhancement is achieved using the equation
• In frequency domain
, , ,
, ,
,
,
,
1
1
1 2 2
F u v H u v F u v
1 H u v F u v
1 4 D u v F
g
u
x y
v
2g x, y f x, y c f x, y
Image Sharpening
http://ivm.sjtu.edu.cn
• High-boost filtering
▫ Spatial domain:
▫ Frequency domain
, , ,1
LPLP H u v F uf x y v
mask LPg x, y f x, y f x, y
Image Sharpening
maskg x, y f x, y k g x, y
* ,, ,1
LP1 k 1 H u v ug vx Fy
http://ivm.sjtu.edu.cn
• High-frequency-emphasis filtering:
* ,, ,1
HPg 1 k H u vx u vy F
Image Sharpening
http://ivm.sjtu.edu.cn
• A slightly more general formulation of high-frequency-emphasis filtering
* ,, ,1
1 2 HPk k H ug x y v F u v
Image Sharpening
http://ivm.sjtu.edu.cn
Take another look
Important things are to be repeated
for 3 times !
http://ivm.sjtu.edu.cn
Homework One
Page327: problem 4.8(b), 4.22, 4.28, 4.33, 4.35
http://ivm.sjtu.edu.cn
Thank You!