Digital Image Processing

Chapter 5: Image Restoration

A Model of the Image
Degradation/Restoration Process
 Degradation
Degradation function H
Additive noise ( x, y )
Spatial domain
g ( x, y ) h ( x, y ) * f ( x , y ) ( x, y )
Frequency domain

G (u, v) H (u , v) F (u, v) N (u , v)
 Restoration

g ( x, y ) Restoration Filter f ( x, y )
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
 Gaussian noise
The PDF of a Gaussian random
variable, z,
1 ( z ) 2 / 2 2
p( z ) e
2
Mean:
Standard deviation:
Variance:
2
 70% of its values will be in the range

( ), ( )
95% of its values will be in the range

( 2 ), ( 2 )
 Rayleigh noise
The PDF of Rayleigh noise,
2 ( z a ) 2 / b
( z a )e for z a
p( z ) b
0 for z a
Mean: a b/4
b( 4 )
Variance: 2

4
 Erlang (Gamma) noise
The PDF of Erlang noise, a0 , b
is a positive integer,
a b z b 1 a z
e for z 0
p( z ) (b 1)!
0 for z 0
b
Mean:
a
b
Variance: 2
2

a
 Exponential noise
The PDF of exponential noise, a0 ,

ae a z for z 0
p( z )
0 for z 0
1
Mean:
a
1
Variance: 2
2

a
 Uniform noise
The PDF of uniform noise,

1
if a z b
p( z ) b a
0 otherwise
ab
Mean:
2
Variance: 2
(b a ) 2

12
 Impulse (salt-and-pepper) noise
The PDF of (bipolar) impulse noise,

Pa for z a

p ( z ) Pb for z b
0 otherwise

b a : gray-level b will appear as a
light dot, while level a will appear like
a dark dot
Unipolar: either Pa or Pb is zero
 Usually, for an 8-bit image, a =0
(black) and b =0 (white)
 Modeling
Gaussian
Electronic circuit noise, sensor noise due
to poor illumination and/or high
temperature
Rayleigh
Range imaging
Exponential and gamma
Laser imaging
 Impulse
Quick transients, such as faulty switching
Uniform
Least descriptive
Basis for numerous random number

generators
 Periodic noise
Arises typically from electrical or
electromechanical interference
Reduced significantly via frequency
domain filtering
 Estimation of noise parameters
Inspection of the Fourier spectrum
Small patches of reasonably constant
gray level
For example, 150*20 vertical strips
Calculate , , a , b from
zi p ( zi )
zi S

( zi ) p ( zi )
2 2

zi S
Restoration in the Presence of Noise
Only-Spatial Filtering

Degradation
Spatial domain
g ( x, y ) f ( x, y ) ( x, y )
Frequency domain

G (u, v) F (u, v) N (u, v)

 Mean filters
Arithmetic mean filter

f ( x, y ) 1
g ( s, t )
mn ( s ,t )S xy
Geometric mean filter
1
mn
f ( x, y ) g ( s, t )
( s ,t )S xy
 Harmonic mean filter
Works well for salt noise, but fails fpr
pepper noise

mn
f ( x, y )
1

( s ,t )S xy g ( s, t )
 Contraharmonic mean filter
Q0 : eliminates pepper noise

Q0 : eliminates salt noise

g ( s ,
( s ,t )S xy
t ) Q 1

f ( x, y )
g ( s ,
( s ,t )S xy
t ) Q
 Usage
Arithmetic and geometric mean filters:
suited for Gaussian or uniform noise
Contraharmonic filters: suited for
impulse noise
 Order-statistics filters
Median filter
Effective in the presence of both bipolar
and unipolar impulse noise

f ( x, y ) median{g ( s, t )}
( s ,t )S xy
 Max and min filters
max filters reduce pepper noise

f ( x, y ) max {g ( s, t )}
( s ,t )S xy

min filters salt noise

f ( x, y ) min {g ( s, t )}
( s ,t )S xy
 Midpoint filter
Works best for randomly distributed
noise, like Gaussian or uniform noise

f ( x, y ) 1 max {g ( s, t )} min {g ( s, t )}
2 ( s ,t )S xy ( s ,t )S xy
 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

1
f ( x, y )
mn d
g ( s, t )
( s ,t )S xy
r
 Adaptive, local noise reduction filter
If 2
is zero, return simply the value
of g ( x, y )
If 2 2 , return a value close to
L
g ( x, y )
If
L , return the arithmetic
2 2

mean value mL

f ( x, y ) g ( x, y )
2

g ( x, y ) mL
2
L
 Adaptive median filter
z min = minimum gray level value in
S xy
z max = maximum gray level value in
S xy
z med = median of gray levels in S xy
z xy = gray level at coordinates ( x, y )
S max = maximum allowed size of S xy
 Algorithm:
Level A: A1= z med z min
A2= z med z max
If A1>0 AND A2<0, Go to
level B
Else increase the window size
If window size S max
repeat level A
Else output z med
 Level B: B1= z xy z min
B2= z xy z max
If B1>0 AND B2<0, output z xy
Else output z med
 Purposes of the algorithm
Remove salt-and-pepper (impulse) noise
Provide smoothing

Reduce distortion, such as excessive

thinning or thickening of object

boundaries
 Periodic Noise Reduction by
Frequency Domain Filtering

Bandreject filters
Ideal bandreject filter
W
1 if D(u, v) D 0
2

W W
H (u, v) 0 if D 0 D(u, v) D 0
2 2
1 W
if D(u, v) D 0

2

D(u, v) (u M / 2) (v N / 2)
2

2 1/ 2
 Butterworth bandreject filter of order n
1
H (u , v) 2n
D(u , v)W
1 2 2
D (u , v) D0
Gaussian bandreject filter
2
1 D 2 ( u ,v ) D02

2 D ( u ,v )W
H (u, v) 1 e
 Bandpass filters

H bp (u , v) 1 H br (u , v)
 Notch filters
Ideal notch reject filter

0 if D1 (u, v) D 0 or D 2 (u, v) D 0
H (u, v)
1 otherwise

D1 (u , v) (u M / 2 u0 ) (v N / 2 v0 )
2
2 1/ 2

D (u , v) (u M / 2 u )
2 0
2
(v N / 2 v )
0
2 1/ 2
 Butterworth notch reject filter of
order n
1
H (u , v) n
D 2

1 0

D1 (u , v) D2 (u , v)
 Gaussian notch reject filter

1 D1 ( u ,v ) D2 ( u ,v )
2
2 D0
H (u, v) 1 e
 Notch pass filter

H np (u, v) 1 H nr (u, v)
 Optimum notch filtering
 Interference noise pattern

N (u, v) H (u , v)G (u, v)

Interference noise pattern in the spatial
domain
( x, y ) {H (u , v)G (u, v)}
1

Subtract from g ( x, y ) a weighted

portion of ( x, y ) to obtain an
estimate of f ( x, y )
f ( x, y ) g ( x, y ) w( x, y ) ( x, y )
 Minimize the local variance of f ( x, y )
The detailed steps are listed in Page
251
Result

g ( x, y ) ( x, y ) g ( x, y ) ( x, y )
w( x, y )
2 ( x, y ) 2 ( x, y )
Linear, Position-Invariant Degradations

Input-output relationship

g ( x, y ) H [ f ( x, y )] ( x, y )

( x, y ) 0

g ( x, y ) H [ f ( x, y )]
 H is linear if
H [af1 ( x, y ) bf 2 ( x, y )]
aH [ f1 ( x, y )] bH [ f 2 ( x, y )]
Additivity

H [ f1 ( x, y ) f 2 ( x, y )]
H [ f1 ( x, y )] H [ f 2 ( x, y )]
 Homogeneity

H [af1 ( x, y )] aH [ f1 ( x, y )]
Position (or space) invariant

H [ f ( x , y )] g ( x , y )]
 In terms of a continuous impulse
function

f ( x, y ) f ( , ) ( x , y )d d

g ( x, y ) H [ f ( x, y )]

H f ( , ) ( x , y )d d

g ( x, y )
H f ( , ) ( x , y ) d d

f ( , ) H ( x , y ) d d




f ( , )h( x, , y, )d d

 Impulse response of H

h( x, , y, ) H [ ( x , y )]
In optics, the impulse becomes a point
of light
Point spread function (PSF)

h ( x, , y , )
All physical optical systems blur
(spread) a point of light to some
degree
 Superposition (or Fredholm) integral of
the first kind

g ( x, y )



f ( , )h( x, , y, )d d
 If H is position invariant

H [ ( x , y )] h( x , y )

Convolution integral

g ( x, y )


f ( , )h( x , y )d d
 In the presence of additive noise

g ( x, y )



f ( , )h( x, , y, )d d ( x, y )
If H is position invariant
g ( x, y )



f ( , )h( x , y )d d ( x, y )
 If H is position invariant

g ( x , y ) h ( x , y ) f ( x, y ) ( x, y )
G (u, v) H (u, v) F (u, v) N (u, v)
Restoration approach
Image deconvolution
Deconvolution filter
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

Gs (u , v)
H s (u , v)
Fs (u , v)
 Estimation by experimentation
Obtain the impulse response of the
degradation by imaging an impulse
(small dot of light) using the same
system settings
G (u, v)
H (u , v)
A
Observed image G (u , v )
The strength of the impulse A
 Estimation by modeling
Hufnagel and Stanley
Physical characteristic of atmospheric
turbulence

5
k (u 2 v 2 ) 6
H (u , v) e
 Image motion
T
g ( x, y ) f [ x x0 (t ), y y0 (t )]dt
0

G (u , v) g ( x, y )e j 2 ( u x v y )
dx dy

T
f [ x x (t ), y y (t )]dt

0 0
0

j 2 ( u x v y )
e dx dy

f [ x x0 (t ), y y0 (t )]
T

0
e j 2 ( u x v y )

dx dy dt
 T
G (u , v) F (u , v)e j 2 [ u x0 ( t ) v y0 ( t )]
dt
0
T
F (u , v) e j 2 [ u x0 ( t ) v y0 ( t )]
dt
0

F (u , v) H (u , v)
Where
T
H (u , v) e j 2 [ u x0 ( t ) v y0 ( t )]
dt
0
 If x0 (t ) at / T and y0 (t ) 0
T
H (u , v) e j 2 [ u x0 ( t )]
dt
0
T
e j 2 [ u at / T ]
dt
0

T j ua
sin( ua)e
ua
 If x0 (t ) at / T and y0 (t ) bt / T

H (u , v)
T j ( ua vb )
sin[ (ua vb)]e
(ua vb)
Inverse Filtering
Direct inverse filtering
Limiting the analysis to frequencies
near the origin

G (u , v)
F (u, v)
H (u , v)
N (u , v)
F (u, v)
H (u , v)
Minimum Mean Square Error (Wiener)
Filtering

Minimize
2
e E{( f f ) }
2

Terms
H (u , v ) = degradation function

H (u , v) = complex conjugate of H (u , v)

2
H (u , v) = H (u, v) H (u, v)

 2
S (u, v) N (u, v) = power spectrum
of the noise
2
S f (u , v) F (u, v) = power spectrum
of the undegraded image
 Wiener filter
H * (u , v ) S (u , v )
F (u , v)
f
2
G (u , v)
S f (u , v) H (u , v) S (u , v)
H * (u , v)
2
G (u, v)
H (u, v) S (u , v) / S f (u , v)
1 H (u , v)
2

2
G (u , v)
H (u , v) H (u , v) S (u , v) / S f (u , v)
 White noise

1 H ( u , v )
2

F (u, v) 2
G (u, v)
H (u , v) H (u , v) K
Constrained Least Squares Filtering

Vector-matrix form
g ( x, y ) h ( x, y ) * f ( x, y ) ( x, y )

g Hf

g , f , : MN 1
H : MN MN
 Minimize

M 1 N 1
C f ( x, y ) 2 2

x 0 y 0

Subject to
2

g Hf
2
 The solution
H * (u , v )
F (u , v) 2 2
G (u , v)
H (u , v) P(u , v)
Where P (u , v ) is the Fourier transform
of the function
0 1 0
P ( x, y ) 1 4 1
0 1 0
 Computing by iteration

r g Hf
Adjust so that
2 2
r a
 Computation
M 1 N 1
r r 2 ( x, y )
2

x 0 y 0
M 1 N 1
1
m
MN
( x, y)
x 0 y 0
2

( x, y) m
M 1 N 1
1

2

MN x 0 y 0

2
MN [ 2 m2 ]
 Algorithm
1: Specify an initial value of
2: Compute
2 2
3: Stop if r a is satisfied;
otherwise return to Step 2 after
increasing if 2 2
r a or

decreasing
2 2
if r . a
Geometric Mean FIlter

F (u, v)
1


H * (u, v) H * (u , v)
G (u, v)
2
H (u , v) S (u , v )

2
H (u, v)

S ( u , v )

f
Geometric Transformations
Spatial transformations
x ' r ( x, y )
y ' s ( x, y )

Tiepoints
 Bilinear equations

x' r ( x, y ) c1 x c2 y c3 xy c4
y ' s ( x, y ) c5 x c6 y c7 xy c8
 Gray-level interpolation
v( x' , y ' ) ax'by ' cx' y ' d