IMAGE RESTORATION
Restoring an image from the noise.
What way it is different from the image
enhancement technique , which are used for the
noise removal.
More complex and mathematical than
enhancement.
A model for Image
Degradation/Restoration process
g(x,y)
f(x,y) Degradation Degradation fˆ ( x, y)
function + function
H H
η(x,y)
Noise
g ( x, y) h( x, y) * f ( x, y) ( x, y)
G(u, v) H (u, v) F (u, v) N (u, v)
Noise Models
Gaussian Noise
Mathematically tractable in
both spatial and frequency
domain. PDF of a
Gaussian random variable
z is given by z Intensity
( z z )2 z M ean (average)
1
p( z ) e 2 2
Sandard deviation
2
Variance of z
2
Rayleigh Noise
PDF is given by
2 ( z a ) 2
( z a)e b for z a
p( z ) b
0 for z a
mean z a b / 4
b( 4 )
Displacement from origin
variance
2
Skewed to the right
4
Erlang (gamma) noise
Exponential noise
This is a special case of Erlang Noise with
b=1
Uniform Noise
Impulse (Salt and Pepper Noise)
PDF of impulse noise is
Pa for z a
p( z ) Pb for z b
0
otherwise
If b>a , intensity b will appear as alight dot
in the image a will appear as dark dot
Periodic Noise
Restoration in the
presence of noise
only-Spatial filtering
g ( x, y ) f ( x, y ) ( x, y )
and
G (u, v) F (u, v) N (u, v)
Mean Filters
Arithmetic mean filter
1
f ( x, y ) g ( s, t )
mn ( s ,t )S xy
Geometric mean filter
1
mn
fˆ ( x, y ) g ( s, t )
( s ,t )S x , y
This filter achieves smoothing comparable to
arithmetic mean filter , but it tends to lose less
details in the process.
Harmonic mean filter
mn
fˆ ( x, y )
1
( s ,t )S xy g ( s, t )
This filter works well for salt noise, but fails the
pepper noise. Works well with other type of
noise like Gaussian noise.
Contraharmonic mean filter
g ( s, t )
( s ,t )S xy
Q 1
Where Q is order
fˆ ( x, y ) of the filter
g ( s, t )
( s ,t )S xy
Q
Well suited for reducing or virtually eliminating
the effects of salt and pepper noise.
+ve Q removes eliminates pepper noise.
-ve Q eliminates salt noise.
It cannot do both simultaneously.
If Q=0, it reduces to arithmetic mean filter
If Q=-1 , it reduces to harmonic mean filter.
Order statistics filter
Median Filter
fˆ ( x, y) median{g ( s, t )}
( s ,t )S xy
Max and Min filter
fˆ ( x, y) max {g ( s, t )}
( s ,t )S xy
Useful in finding the brightest points in an
image, due to this pepper noise will be reduced
fˆ ( x, y) min {g ( s, t )}
( s ,t )S xy
Useful in finding the darkest points in an image,
due to this salt noise will be reduced
Mid point filter
ˆf ( x, y) 1 max {g ( s, t )} min {g ( s, t )}
2 ( s ,t )S xy ( s ,t )S xy
Computes the midpoint between the maximum
and minimum values in the sub image.
It works best for randomly distributed noise ,
like random and Gaussian noise.
Alpha trimmed mean filter
Suppose , we delete d/2 lowest and the d/2
highest intensity values of g(s,t) in the
neighborhood Sxy . Let gr(s,t) represents the
minimum mn-d pixels. This is filter formed by
averaging these remaining pixels.
1
fˆ ( x, y )
mn d
g ( s, t )
( s ,t )S xy
r
d ranges from mn-1.
When d=0 , it reduces to arithmetic mean filter
If d=mn-1, this filter reduces median filter.
Useful in noise, which is a combination of salt &
pepper and Gaussian noise
X-ray
image
Result of
Result of arithmetic
Image
geometric mean mean filter
corrupted by
[3x3] [3x3]
additive
Gaussian noise
Estimation of the
degradation function
There are three ways it can be estimated
Observation
Experimentation
Mathematical Modeling
Estimation by Image Observation
Let the image is degraded by an unknown
degradation function H.
Assume that H is linear and position invariant.
Gather information from the image.
Look at the small rectangular section of a
blurred image, which contain sample structure.
To reduce the noise effect , look at the strong
signal content area (area of high contrast).
Now process the sub-image to arrive at a result
that it a unblurred as possible.
(Using sharpening filter or by processing small areas
in hand)
Let the observed sub-image be denoted by
g s ( x, y )
and the processed sub-image fˆs ( x, y)
Assuming negligible noise effect , we have
Gs (u, v)
H s (u, v)
Fs (u, v)
Based on the information of Position invariance,
we can deduce complete degradation function
H(u,v) from the characteristics of above function
Suppose the radial plot of Hs(u,v) has the
approximate shape of the Gaussian curve , same
information is used for H(u,v) in the larger scale.
Estimation by Experimentation
If the equipment , with which we obtained the
degraded image is available, accurate estimation
of the degradation function can be obtained.
With various system settings of the equipments
, obtain a degraded as closely as possible to the
given degraded image.
Using the system settings (Response) get an
impulse response by imaging an impulse (small
dot of light.
An impulse is simulated by bright dot of light as
bright as possible to reduce the effect of noise to
negligible value.
Then Fourier transform of an impulse is a
constant
G(u, v)
H (u, v)
A
Where A is constant describing strength of the
impulse.
Estimation by modeling
The mathematical degradation model proposed
by Hufnagel and Stanley [1964] is based on the
physical characteristics of the atmospheric
turbulence. 5
k (u 2 v 2 )
H (u, v) e
6
Where k is the constant depends on the nature of
the turbulence.
Image with
Image with severe
negligible turbulence
turbulence k=0.0025
Image with Image with
mild low
turbulence turbulence
k=0.001 k=0.00025
Another way of mathematical modeling is starting
from the basic principle . Let us consider a case
where image is blurred due to the uniform linear
motion between the image and the sensor during
the image acquisition.
Let an image f(x,y) undergoes a planar motion
and that x0(t) and y0(t) are time varying
components of motion in x and y direction
respectively.
The total exposure at any point of recording
medium is obtained by integrating the
instantaneous exposure over time interval during
which the imaging system shutter is open.
T
g ( x, y ) f [ x x0 (t ), y y0 (t )]
0
Where g ( x, y ) is the blurred image