2. A Model of the Image
Degradation/Restoration Process
3. Degradation
Degradation function H
Additive noise
Spatial domain
Frequency domain
)
,
( y
x
)
,
(
)
,
(
*
)
,
(
)
,
( y
x
y
x
f
y
x
h
y
x
g
)
,
(
)
,
(
)
,
(
)
,
( v
u
N
v
u
F
v
u
H
v
u
G
5. Noise Models
Sources of noise
Image acquisition, digitization,
transmission
White noise
The Fourier spectrum of noise is
constant
Assuming
Noise is independent of spatial
coordinates
Noise is uncorrelated with respect to
the image itself
6. Gaussian noise
The PDF of a Gaussian random variable,
z,
Mean:
Standard deviation:
Variance:
2
2
2
/
)
(
2
1
)
(
z
e
z
p
2
7. 70% of its values will be in the range
95% of its values will be in the range
)
(
),
(
)
2
(
),
2
(
8. Rayleigh noise
The PDF of Rayleigh noise,
Mean:
Variance:
a
z
a
z
e
a
z
b
z
p
b
a
z
for
0
for
)
(
2
)
(
/
)
( 2
4
/
b
a
4
)
4
(
2
b
9.
10. Erlang (Gamma) noise
The PDF of Erlang noise, ,
is a positive integer,
Mean:
Variance:
0
for
0
0
for
)!
1
(
)
(
1
z
z
e
b
z
a
z
p
z
a
b
b
a
b
2
2
a
b
0
a b
11. Exponential noise
The PDF of exponential noise, ,
Mean:
Variance:
0
for
0
0
for
)
(
z
z
ae
z
p
z
a
a
1
2
2 1
a
0
a
12. Uniform noise
The PDF of uniform noise,
Mean:
Variance:
otherwise
0
if
1
)
(
b
z
a
a
b
z
p
2
b
a
12
)
( 2
2 a
b
13. Impulse (salt-and-pepper) noise
The PDF of (bipolar) impulse noise,
: gray-level will appear as a
light dot, while level will appear like
a dark dot
Unipolar: either or is zero
otherwise
0
for
for
)
( b
z
P
a
z
P
z
p b
a
a
b b
a
a
P b
P
14. Usually, for an 8-bit image, =0
(black) and =0 (white)
b
a
15. Modeling
Gaussian
Electronic circuit noise, sensor noise due
to poor illumination and/or high
temperature
Rayleigh
Range imaging
Exponential and gamma
Laser imaging
16. Impulse
Quick transients, such as faulty switching
Uniform
Least descriptive
Basis for numerous random number
generators
17.
18.
19.
20. Periodic noise
Arises typically from electrical or
electromechanical interference
Reduced significantly via frequency
domain filtering
21.
22. Estimation of noise parameters
Inspection of the Fourier spectrum
Small patches of reasonably constant
gray level
For example, 150*20 vertical strips
Calculate , , , from
a b
S
z
i
i
i
z
p
z )
(
S
z
i
i
i
z
p
z )
(
)
( 2
2
23.
24. Restoration in the Presence of Noise
Only-Spatial Filtering
Degradation
Spatial domain
Frequency domain
)
,
(
)
,
(
)
,
( y
x
y
x
f
y
x
g
)
,
(
)
,
(
)
,
( v
u
N
v
u
F
v
u
G
25. Mean filters
Arithmetic mean filter
Geometric mean filter
xy
S
t
s
t
s
g
mn
y
x
f
)
,
(
)
,
(
1
)
,
(
ˆ
mn
S
t
s xy
t
s
g
y
x
f
1
)
,
(
)
,
(
)
,
(
ˆ
26. Harmonic mean filter
Works well for salt noise, but fails fpr
pepper noise
xy
S
t
s t
s
g
mn
y
x
f
)
,
( )
,
(
1
)
,
(
ˆ
27. Contraharmonic mean filter
: eliminates pepper noise
: eliminates salt noise
xy
xy
S
t
s
Q
S
t
s
Q
t
s
g
t
s
g
y
x
f
)
,
(
)
,
(
1
)
,
(
)
,
(
)
,
(
ˆ
0
Q
0
Q
28.
29.
30. Usage
Arithmetic and geometric mean filters:
suited for Gaussian or uniform noise
Contraharmonic filters: suited for
impulse noise
31.
32. Order-statistics filters
Median filter
Effective in the presence of both bipolar
and unipolar impulse noise
)}
,
(
{
median
)
,
(
ˆ
)
,
(
t
s
g
y
x
f
xy
S
t
s
33. Max and min filters
max filters reduce pepper noise
min filters salt noise
)}
,
(
{
max
)
,
(
ˆ
)
,
(
t
s
g
y
x
f
xy
S
t
s
)}
,
(
{
min
)
,
(
ˆ
)
,
(
t
s
g
y
x
f
xy
S
t
s
34. Midpoint filter
Works best for randomly distributed noise,
like Gaussian or uniform noise
)}
,
(
{
min
)}
,
(
{
max
2
1
)
,
(
ˆ
)
,
(
)
,
(
t
s
g
t
s
g
y
x
f
xy
xy S
t
s
S
t
s
35. Alpha-trimmed mean filter
Delete the d/2 lowest and the d/2 highest
gray-level values
Useful in situations involving multiple
types of noise, such as a combination of
salt-and-pepper and Gaussian noise
xy
S
t
s
r t
s
g
d
mn
y
x
f
)
,
(
)
,
(
1
)
,
(
ˆ
36.
37.
38.
39. Adaptive, local noise reduction filter
If is zero, return simply the value
of
If , return a value close to
If , return the arithmetic
mean value
2
)
,
( y
x
g
2
2
L
)
,
( y
x
g
2
2
L
L
m
L
L
m
y
x
g
y
x
g
y
x
f
)
,
(
)
,
(
)
,
(
ˆ
2
2
40.
41. Adaptive median filter
= minimum gray level value in
= maximum gray level value in
= median of gray levels in
= gray level at coordinates
= maximum allowed size of
min
z
max
z
med
z
xy
z
max
S
xy
S
xy
S
xy
S
xy
S
)
,
( y
x
42. Algorithm:
Level A: A1=
A2=
If A1>0 AND A2<0, Go to
level B
Else increase the window size
If window size
repeat level A
Else output
min
z
zmed
max
z
zmed
max
S
med
z
43. Level B: B1=
B2=
If B1>0 AND B2<0, output
Else output
min
z
zxy
max
z
zxy
xy
z
med
z
44. Purposes of the algorithm
Remove salt-and-pepper (impulse) noise
Provide smoothing
Reduce distortion, such as excessive
thinning or thickening of object
boundaries
45.
46. Periodic Noise Reduction by Frequency
Domain Filtering
Bandreject filters
Ideal bandreject filter
2
D
v)
D(u,
if
1
2
D
v)
D(u,
2
D
if
0
2
D
v)
D(u,
if
1
)
,
(
0
0
0
0
W
W
W
W
v
u
H
2
/
1
2
2
)
2
/
(
)
2
/
(
)
,
( N
v
M
u
v
u
D
47. Butterworth bandreject filter of order n
Gaussian bandreject filter
n
D
v
u
D
W
v
u
D
v
u
H 2
2
0
2
)
,
(
)
,
(
1
1
)
,
(
2
2
0
2
)
,
(
)
,
(
2
1
1
)
,
(
W
v
u
D
D
v
u
D
e
v
u
H
52. Notch filters
Ideal notch reject filter
otherwise
1
D
v)
(u,
D
or
D
v)
(u,
D
if
0
)
,
( 0
2
0
1
v
u
H
2
/
1
2
0
2
0
1 )
2
/
(
)
2
/
(
)
,
( v
N
v
u
M
u
v
u
D
2
/
1
2
0
2
0
2 )
2
/
(
)
2
/
(
)
,
( v
N
v
u
M
u
v
u
D
53. Butterworth notch reject filter of
order n
n
v
u
D
v
u
D
D
v
u
H
)
,
(
)
,
(
1
1
)
,
(
2
1
2
0
54. Gaussian notch reject filter
2
0
2
1 )
,
(
)
,
(
2
1
1
)
,
(
D
v
u
D
v
u
D
e
v
u
H
55.
56. Notch pass filter
)
,
(
1
)
,
( v
u
H
v
u
H nr
np
59. Interference noise pattern
Interference noise pattern in the spatial
domain
Subtract from a weighted
portion of to obtain an
estimate of
)
,
(
)
,
(
)
,
( v
u
G
v
u
H
v
u
N
)}
,
(
)
,
(
{
)
,
( 1
v
u
G
v
u
H
y
x
)
,
(
)
,
(
)
,
(
)
,
(
ˆ y
x
y
x
w
y
x
g
y
x
f
)
,
( y
x
g
)
,
( y
x
)
,
( y
x
f
60. Minimize the local variance of
The detailed steps are listed in Page
251
Result
)
,
(
ˆ y
x
f
)
,
(
)
,
(
)
,
(
)
,
(
)
,
(
)
,
(
)
,
(
2
2
y
x
y
x
y
x
y
x
g
y
x
y
x
g
y
x
w
65. H is linear if
Additivity
)]
,
(
[
)]
,
(
[
)]
,
(
)
,
(
[
2
1
2
1
y
x
f
bH
y
x
f
aH
y
x
bf
y
x
af
H
)]
,
(
[
)]
,
(
[
)]
,
(
)
,
(
[
2
1
2
1
y
x
f
H
y
x
f
H
y
x
f
y
x
f
H
66. Homogeneity
Position (or space) invariant
)]
,
(
[
)]
,
(
[ 1
1 y
x
f
aH
y
x
af
H
)]
,
(
)]
,
(
[
y
x
g
y
x
f
H
67. In terms of a continuous impulse
function
d
d
y
x
f
y
x
f )
,
(
)
,
(
)
,
(
d
d
y
x
f
H
y
x
f
H
y
x
g
)
,
(
)
,
(
)]
,
(
[
)
,
(
68.
d
d
y
x
h
f
d
d
y
x
H
f
d
d
y
x
f
H
y
x
g
)
,
,
,
(
)
,
(
)
,
(
)
,
(
)
,
(
)
,
(
)
,
(
69. Impulse response of H
In optics, the impulse becomes a point
of light
Point spread function (PSF)
All physical optical systems blur
(spread) a point of light to some
degree
)]
,
(
[
)
,
,
,
(
y
x
H
y
x
h
)
,
,
,
(
y
x
h
70. Superposition (or Fredholm) integral of
the first kind
d
d
y
x
h
f
y
x
g
)
,
,
,
(
)
,
(
)
,
(
71. If H is position invariant
Convolution integral
)
,
(
)]
,
(
[
y
x
h
y
x
H
d
d
y
x
h
f
y
x
g
)
,
(
)
,
(
)
,
(
72. In the presence of additive noise
If H is position invariant
)
,
(
)
,
,
,
(
)
,
(
)
,
(
y
x
d
d
y
x
h
f
y
x
g
)
,
(
)
,
(
)
,
(
)
,
(
y
x
d
d
y
x
h
f
y
x
g
73. If H is position invariant
Restoration approach
Image deconvolution
Deconvolution filter
)
,
(
)
,
(
)
,
(
)
,
( y
x
y
x
f
y
x
h
y
x
g
)
,
(
)
,
(
)
,
(
)
,
( v
u
N
v
u
F
v
u
H
v
u
G
74. Estimating the Degradation Function
Estimation by image observation
In order to reduce the effect of noise in
our observation, we would look for
areas of strong signal content
)
,
(
ˆ
)
,
(
)
,
(
v
u
F
v
u
G
v
u
H
s
s
s
75. Estimation by experimentation
Obtain the impulse response of the
degradation by imaging an impulse
(small dot of light) using the same
system settings
Observed image
The strength of the impulse
A
v
u
G
v
u
H
)
,
(
)
,
(
)
,
( v
u
G
A
76.
77. Estimation by modeling
Hufnagel and Stanley
Physical characteristic of atmospheric
turbulence
6
5
2
2
)
(
)
,
( v
u
k
e
v
u
H
82. If and
T
at
t
x /
)
(
0 0
)
(
0
t
y
ua
j
T
T
at
u
j
T t
x
u
j
e
ua
ua
T
dt
e
dt
e
v
u
H
0
]
/
[
2
0
)]
(
[
2
)
sin(
)
,
( 0
83. If and
T
at
t
x /
)
(
0 T
bt
t
y /
)
(
0
)
(
)]
(
sin[
)
(
)
,
(
vb
ua
j
e
vb
ua
vb
ua
T
v
u
H
84.
85. Inverse Filtering
Direct inverse filtering
Limiting the analysis to frequencies
near the origin
)
,
(
)
,
(
)
,
(
)
,
(
)
,
(
)
,
(
ˆ
v
u
H
v
u
N
v
u
F
v
u
H
v
u
G
v
u
F
86.
87. Minimum Mean Square Error (Wiener)
Filtering
Minimize
Terms
= degradation function
= complex conjugate of
=
}
)
ˆ
{( 2
2
f
f
E
e
)
,
( v
u
H
)
,
( v
u
H )
,
( v
u
H
2
)
,
( v
u
H )
,
(
)
,
( v
u
H
v
u
H
88. = power spectrum
of the noise
= power spectrum
of the undegraded image
2
)
,
(
)
,
( v
u
N
v
u
S
2
)
,
(
)
,
( v
u
F
v
u
S f
93. Constrained Least Squares Filtering
Vector-matrix form
, , :
:
g
)
,
(
)
,
(
*
)
,
(
)
,
( y
x
y
x
f
y
x
h
y
x
g
1
MN
MN
MN
H
η
f
η
Hf
g
94. Minimize
Subject to
1
0
1
0
2
2
)
,
(
M
x
N
y
y
x
f
C
2
2
ˆ η
f
H
g
95. The solution
Where is the Fourier transform
of the function
)
,
(
)
,
(
)
,
(
)
,
(
*
)
,
(
ˆ
2
2
v
u
G
v
u
P
v
u
H
v
u
H
v
u
F
)
,
( v
u
P
0
1
0
1
4
1
0
1
0
)
,
( y
x
P
96.
97. Computing by iteration
Adjust so that
f
H
g
r ˆ
a
2
2
η
r
99. Algorithm
1: Specify an initial value of
2: Compute
3: Stop if is satisfied;
otherwise return to Step 2 after
increasing if or
decreasing if .
a
2
2
η
r
a
2
2
η
r
a
2
2
η
r