MODULE-4 VII SEM DIGITAL IMAGE PROCESSING [17EC72]

MODULE -4
Image Restoration

Restoration: Noise models, Restoration in the Presence of Noise Only using Spatial
Filtering and Frequency Domain Filtering, Linear, Position-Invariant Degradations,
Estimating the Degradation Function, Inverse Filtering, Minimum Mean Square Error
(Wiener) Filtering, Constrained Least Squares Filtering.
[Text: Chapter 5: Sections 5.2, to 5.9]

RBT: L1, L2, L3

TEXT, REFERENCE & ADDITIONAL REFERENCE BOOKS:

BOOK TITLE/AUTHORS/PUBLICATION /Web links/Channel

Digital Image Processing- Rafel C Gonzalez and Richard E. Woods,

T-1.
PHI 3rd Edition 2010.

Digital Image Processing- S.Jayaraman, S. Esakkirajan,

AR-1
T.Veerakumar, Tata McGraw Hill 2014.

Fundamentals of Digital Image Processing-A. K. Jain, Pearson

AR-2
2004.

WR-1 https://nptel.ac.in/courses/117/105/117105135/

WR-2 https://www.youtube.com/watch?v=DSGHkvQBMbs

Channel https://www.youtube.com/channel/UCD0mXLuJqtvPxBqRI1r66cg/videos

Introduction to Image Restoration

Dr Ganesh V Bhat (Dept. of ECE Canara Engineering College India) Page 1

MODULE-4 VII SEM DIGITAL IMAGE PROCESSING [17EC72]

The degradation process is generally modelled as a degradation function H and

the additive noise term 𝜂(𝑥, 𝑦) together they yield 𝑔(𝑥, 𝑦). Given 𝑔(𝑥, 𝑦), some
knowledge about H and 𝜂(𝑥, 𝑦) it is the objective of restoration to estimate 𝑓̂(𝑥, 𝑦). Of
course this estimate should be as close as possible to 𝑓(𝑥, 𝑦) .

where
𝑔(𝑥, 𝑦) the degraded image
𝑓(𝑥, 𝑦) the original image
H an operator that represents the degradation process
𝜂(𝑥, 𝑦) the external noise which is assumed to be image-independent

If the degradation function H is a linear, shift-invariant process, then the degraded

image is given in the spatial domain by 𝑔(𝑥, 𝑦) = ℎ(𝑥, 𝑦) ∗ 𝑓(𝑥, 𝑦) + 𝜂(𝑥. 𝑦) where ℎ(𝑥, 𝑦)
is the spatial representation of the degradation function, and * indicates convolution.
As we know from the Convolution Theorem this can be rewritten in the frequency
domain. 𝐺(𝑢, v) = 𝐻(𝑢, v)𝐹(𝑢, v) + 𝑁(𝑢, 𝑣) where the capital letters are the Fourier
transforms of the respective function

Possible classification of restoration methods

Restoration methods could be classified as follows:

From the viewpoint of the Signal:
• deterministic: we work with sample by sample processing of the observed
(degraded) image
• stochastic: we work with the statistics of the images involved in the process
From the viewpoint of the Degradation Process:
• non-blind: the degradation process H is known
• blind: the degradation process H is unknown
• semi-blind: the degradation process H could be considered partly known
From the viewpoint of implementation:
• direct
• iterative
• recursive

Dr Ganesh V Bhat (Dept. of ECE Canara Engineering College India) Page 2

MODULE-4 VII SEM DIGITAL IMAGE PROCESSING [17EC72]

Noise Models:
The principle sources of noise in digital images arise during image acquisition
and/or transmission. The performance of imaging sensors are affected by a variety
of factors during acquisition, such as
• Environmental conditions during the acquisition
• Light levels (low light conditions require high gain amplification)
• Sensor temperature (higher temp implies more amplification noise)
Images can also be corrupted during transmission due to interference in the channel
for example
• Lightning
• Atmospheric disturbances
Depending on the specific noise source, a different model must be selected that
accurately reproduces the spatial characteristics of the noise.

Gaussian Noise: Because of its mathematical

tractability in both the spatial and frequency
domain, Gaussian noise models (aka normal
distribution) are used frequently in practice.
The PDF (Probability density function) of a
Gaussian random variable 𝑍 is given by
1 2 2
𝑝𝐺 = 𝑒 −(𝑧−𝑢) /(2𝜎 )
√2𝜋𝜎
where 𝑍 represents the grey level, µ the mean
value and 𝜎 the standard deviation. The
application of the Gaussian model is so convenient that it is often used in
situations in which they are marginally applicable at best.

Rayleigh Noise: Unlike Gaussian

distribution, the Rayleigh distribution is no

symmetric. The PDF of Rayleigh noise is

defined as

the mean and variance are given by

Dr Ganesh V Bhat (Dept. of ECE Canara Engineering College India) Page 3

MODULE-4 VII SEM DIGITAL IMAGE PROCESSING [17EC72]

Erlang Noise/ Gamma Noise: The PDF

of Erlang noise is given by

Its shape
is similar
to Rayleigh disruption. This equation is referred
to as gamma density it is correct only when the
denominator is the gamma function.
Where 𝑎 > 0 𝑎𝑛𝑑 𝑏 is a positive integer. The mean
and variance are then given by

Exponential Noise: Exponential distribution

has an exponential shape. The PDF of exponential
noise is given as

Where a>0. The mean and

variance of this density are
given as
The exponential noise model is a special case of the
Erlang noise model with .

Uniform Noise: The PDF of uniform noise is given

by
The mean and variance of this noise is

Impulse (salt & pepper) Noise: In this

case, the noise is signal dependent, and is
multiplied to the image. The PDF of bipolar
(impulse) noise is given by

Dr Ganesh V Bhat (Dept. of ECE Canara Engineering College India) Page 4

MODULE-4 VII SEM DIGITAL IMAGE PROCESSING [17EC72]

If b>a, gray level b will appear as a light dot in image. Level a will appear like a dark
dot. Because impulse corruption is generally large compared to the signal strength, the
assumption is usually that a and b are digitised as saturated values thus black (pepper)
and white (salt).
Periodic Noise: Periodic noise typically arises from electrical or electromechanical
interference during image acquisition and is spatial dependent.

Estimation of Noise parameters

In rare cases the parameters of noise PDFs may be known from sensor specifications,
but it is often necessary to estimate them for a particular imaging arrangement. If the
imaging system is available, one simple way to study noise characteristics would be to
capture multiple images of homogeneous environments, such as solid grey cards.
In many cases the noise parameters have to be estimated from already captured
images. The simplest way is to estimate the mean and variance of the grey levels from
small patches of reasonably constant grey level.
The parameters of periodic noise are typically estimated by inspection of the Fourier
spectrum. The frequency spikes can often be detected by visual analysis.
Automatic analysis is possible if the noise spikes are either exceptionally pronounced or
a priori knowledge about their location is known.
The presented noise models provide a useful tool for approximating a broad range of
noise corruption situations found in practice.
For example:
Gaussian noise: arises in images due to sensor noise caused by poor illumination and/or
high temperature, and electronic circuit noise
The Rayleigh noise model is used to characterise noise phenomena in range images
Exponential and Erlang noise models find their application in Laser imaging
Impulse noise takes place in situations where high transients (faulty switching) occurs

Restoration in the presence of Noise only- Spatial filtering:

When the only degradation present in an image is noise, i.e.
g(x,y)=f(x,y)+η(x,y) or
G(u,v)= F(u,v)+ N(u,v)
The noise terms are unknown so subtracting them from g(x,y) or G(u,v) is not a realistic
approach. In the case of periodic noise it is possible to estimate N(u,v) from the
spectrum G(u,v). So N(u,v) can be subtracted from G(u,v) to obtain an estimate of
original image. Spatial filtering can be done when only additive noise is present. The
following techniques can be used to reduce the noise effect:

Dr Ganesh V Bhat (Dept. of ECE Canara Engineering College India) Page 5

MODULE-4 VII SEM DIGITAL IMAGE PROCESSING [17EC72]

A Mean Filter:
Arithmetic Mean filter: It is the simplest mean filter. A mean filter smoothes local
variations in image Noise is reduced as a result of blurring. For
every pixel in the image, the pixel value is replaced by the
mean value of its neighboring pixels with a weight .
This will resulted in a smoothing effect in the image. This
operation can be using a convolution mask in which all coefficients have value 1/mn
Geometric Mean filter: An image restored using a geometric
mean filter is given by the expression
Here, each restored pixel is given by the product of the
pixel in the sub image window, raised to the power 1/mn. A
geometric mean filters but it to loose image details in the
process.
Harmonic Mean filter: The harmonic mean filtering
operation is given by the expression.
The harmonic mean filter works well for salt noise but
fails for pepper noise. It does well with Gaussian noise also.
Contraharmonic mean filter Contraharnomic mean filter
reduces salt and pepper noise, Positive Q is suitable for
eliminating pepper noise. Negative Q is suitable for
eliminating salt noise. Cannot do both simultaneously.
For Q = 0, the filter reduces to an arithmetic mean filter.
For Q = -1, the filter reduces to a harmonic mean filter.
Where, Q = the filter order.

B Order statistics filter: Order statistics filters are spatial filters whose response
is based on ordering the pixel contained in the image area encompassed by the filter.
The response of the filter at any point is determined by the ranking result.
Median filter: It is the best order statistic filter; it replaces the
value of a pixel by the median of gray levels in the
Neighborhood of the pixel. The original of the pixel is
included in the computation of the median of the filter are
quite possible because for certain types of random noise, the provide excellent noise
reduction capabilities with considerably less blurring then smoothing filters of similar
size. These are effective for bipolar and unipolor impulse noise.
Max and Min filter: Using the l00th percentile of ranked set of
numbers is called the max filter and is given by the equation
It is used for finding the brightest point in an image. Pepper
noise in the image has very low values, it is reduced by max
filter using the max selection process in the sublimated area
sky. The 0th percentile filter is min filter. This filter is useful
for flinging the darkest point in image. Also, it reduces salt
noise of the min operation.

Dr Ganesh V Bhat (Dept. of ECE Canara Engineering College India) Page 6

MODULE-4 VII SEM DIGITAL IMAGE PROCESSING [17EC72]

Midpoint filter: The midpoint filter

simply computes the midpoint
between the maximum and minimum
values in the area encompassed by It comeliness the order statistics and averaging.
This filter works best for randomly distributed noise like Gaussian or uniform noise.

Alpha-Trimmed Mean Filter: The alpha-timmed mean filter is based on order statistics
and varies between a median and a mean filter.
It is used when an image contains both short
and long tailed types of noise (e.g. both
Gaussian and salt and pepper noise).
Suppose that we delete the d/2 lowest and the
d/2 highest intensity values of g(r, c) in the neighborhood 𝑆𝑥𝑦 . Let 𝑔𝑅 (𝑟. 𝑐) represent the
remaining 𝑚𝑛 − 𝑑 pixels in 𝑆𝑥𝑦 A filter formed by averaging these remaining pixels is
called an alpha-trimmed mean filter. The form of this filter is

Periodic Noise Reduction by Frequency domain filtering/ Selective

Filtering: Bandreject and Bandpass Filters:
There are applications in which it is of interest to process specific bands of frequencies
or small regions of the frequency rectangle. Filters in the first category are called band
filters. If frequencies in the band are filtered out, the band filter is called a bandreject
filter.
Bandpass and bandreject filter transfer functions in the
frequency domain can be constructed by combining
lowpass and highpass filter transfer functions, with the
latter also being derivable from lowpass functions.
Furthermore, a bandpass filter transfer function is
obtained from a bandreject function in the same manner
that we obtained a highpass from a lowpass transfer
function:

Below table shows Bandreject filter transfer functions with C0 as the center of the band
with W as the width of the band, and D(u,v) is the distance from the center of
the transfer function to a point (u,v) in the frequency rectangle.

Dr Ganesh V Bhat (Dept. of ECE Canara Engineering College India) Page 7

MODULE-4 VII SEM DIGITAL IMAGE PROCESSING [17EC72]

Adding lowpass and highpass transfer functions to form Gaussian and Butterworth
bandreject functions presents some difficulties: we have no direct control over W, and
the value of H(u,v) is not 0 at C0.
Fortunately, instead of adding lowpass and highpass transfer function, an alternative is
to modify the expressions for the Gaussian and Butterworth highpass transfer
functions so that they will satisfy the three requirements
stated earlier
To illustrate the procedure for a Gaussian function, we
begin by changing the
point at which

For Gaussian highpass filter (GHPF) transfer function

To get

A plot of this function shows that, below C0 the function

behaves as a lowpass Gaussian function, at C0 the
function will always be 0, and for values higher than the function behaves as a
highpass Gaussian function. Parameter W is proportional to the standard deviation and
thus controls the “width” of the band.
The only problem remaining is that the function is not always 1 at the origin. A simple
modification of as given below removes this
shortcoming:

Notch Filters
Notch filters are the most useful of the selective filters. A notch filter rejects (or passes)
frequencies in a predefined neighborhood of the frequency rectangle. Zero-phase-shift
filters must be symmetric about the origin (center of the frequency rectangle), so a
notch filter transfer function with center at (u0,v0) must have a corresponding notch at
location (-u0,-v0)
Notch reject filter transfer functions are constructed as products of highpass filter
transfer functions whose centers have been translated to the centers of the notches.
The general form is:

Dr Ganesh V Bhat (Dept. of ECE Canara Engineering College India) Page 8

MODULE-4 VII SEM DIGITAL IMAGE PROCESSING [17EC72]

Where Hk(u,v) and H-k(u,v) are highpass filter transfer functions whose centers are at
(u0,v0) and (-u0,-v0) respectively. These centers are specified with respect to the center
of the frequency rectangle, (M/2,N/2), where, as usual, M and N are the number of
rows and columns in the input image. Thus, the distance computations for each filter
transfer function are given by
and

For example, the following is a Butterworth notch reject filter transfer function of order
n, containing three notch pairs is give as :

Where and Dk(u,v) and D-k(u,v) are given by above two Eqs. The constant D0k is the same
for each pair of notches, but it can be different for different pairs. The general equations
for Ideal, Butterworth, Gaussian notch filters are as given below

Linear position invariant degradation models

Definition:
We again consider the general degradation model
g ( x, y ) = H  f ( x , y )  + n ( x , y )
If we ignore the presence of the external noise n( x, y) we get
g( x, y ) = H  f ( x, y ) 
H is linear if
H  k1 f1 ( x, y ) + k2 f 2 ( x, y )  = k1H  f1 ( x, y )  + k2 H  f 2 ( x, y ) 
H is position (or space) invariant if
H  f ( x − a, y − b)  = y ( x − a, y − b )

Dr Ganesh V Bhat (Dept. of ECE Canara Engineering College India) Page 9

MODULE-4 VII SEM DIGITAL IMAGE PROCESSING [17EC72]

In this course we will be dealing with linear, space invariant type of degradations, as in
a real life problem many types of degradations can be approximated by linear, position
invariant processes.

Advantage: Extensive tools of linear system theory become available.

Disadvantage: In some real life problems nonlinear and space variant models would be
more appropriate for the description of the degradation phenomenon.

Typical linear position invariant degradation models:

Motion blur. It occurs when there is relative motion between the object and the camera
during exposure.
1 L L
 , if −  x, y 
h ( x, y ) =  L 2 2
0, otherwise
Atmospheric turbulence. It is due to random variations in the reflective index of the
medium between the object and the imaging system and it occurs in the imaging of
astronomical objects.
 x2 + y 2 
h( x, y ) = K exp  − 
 2 2 
Uniform out of focus blur
 1
 , if x 2 + y 2  R
h( x, y ) =   R
0, otherwise
Uniform 2-D blur
 1 L L
 2 , if −  x, y 
h( x, y ) =  ( L) 2 2
0,
 otherwise

Characteristic metrics for degradation models

Blurred Signal-to-Noise Ratio (BSNR): a metric that describes the degradation model.
 1 2
 MN   g ( x, y ) − g ( x, y )  
BSNR = 10log10 
i j

  n2 
 
g ( x, y ) = y ( x, y ) − n( x, y ) , g ( x, y) = E{g ( x, y)} ,  n2 : variance of additive noise
Improvement in SNR (ISNR): validates the performance of the image restoration
algorithm.
 f ( x, y ) − y ( x, y )  
 
2
 x y 
ISNR = 10log10  2 
   f ( x, y ) − fˆ ( x, y )  
 x y   


Dr Ganesh V Bhat (Dept. of ECE Canara Engineering College India) Page 10

MODULE-4 VII SEM DIGITAL IMAGE PROCESSING [17EC72]

where fˆ ( x, y ) is the restored image. Both BSNR and ISNR can only be used for
simulation with artificial data.

Derive a mathematical model of Motion Blur

Due to Stabilization / Motion of Object, An image blurred by uniform linear motion
between Image and the sensor during image acquisition; Image f(x,y) is in planar
motion ; x0(t) and y0(t) time varying components of motion. Assuming shutter opening
and closing takes place instantaneously, and optical imaging process is perfect, not
affected by image motion. If T is the duration of the exposure, and g(x,y) is blurred
image.
T
g ( x, y) =  f [ x − x0 (t ), y − y0 (t )]dt
0
Taking the Fourier transformation on both sides we have
+ +
G(u, v) =   g ( x, y)e− j 2 (ux+vy )dt
− −
+ + T
= − [0 f [ x − x0 (t ), y − y0 (t )]dt ]e− j 2 (ux+vy )dxdy
−
T + +
=  [  f [ x − x0 (t ), y − y0 (t )]e− j 2 (ux+vy )dxdy]dt ; f ( x − x0 , y − y0 )  F (u, v)e − j 2 (ux +vy )
0 0

0 − −

=  F (u, v)e− j 2 (ux0 (t )+vy0 (t ))dt

T
0

G(u, v) = F (u, v) e− j 2 (ux0 (t )+vy0 (t ))dt = H (u, v) F (u, v)

T
0

H (u, v) =  e− j 2 [ux0 (t )+vy0 (t )]dt

T
Where
0
T
if x0 (t ) = at / T , y0 (t ) = 0 then H (u, v) = sin( ua)e− j ua
 ua
T
if x0 (t ) = at / T , y0 (t ) = bt / T then H (u, v) = sin[ (ua + vb)]e− j (ua+vb)
 (ua + vb)
T
For constant motion H (u, v) = sin( (ua + vb))e− j (ua +vb)
 (ua + vb)

Dr Ganesh V Bhat (Dept. of ECE Canara Engineering College India) Page 11

MODULE-4 VII SEM DIGITAL IMAGE PROCESSING [17EC72]

Inverse Filtering
The objective is to minimize
J (f ) = n(f ) = g − Hf
2 2

We set the first derivative of the cost function equal to zero

J (f )
= 0  −2H T (g − Hf ) = 0
f
f = (HT H)-1 HTg
If M = N and H−1 exists then
f = H -1g
According to the previous analysis if H (and therefore H-1 ) is block circulant the above
problem can be solved as a set of M  N scalar problems as follows
H  (u, v)G (u, v)  
−1 H (u , v)G (u , v)

F (u, v) = 2
 f (i , j ) =   2

H (u, v)  H (u, v) 

Computational issues concerning inverse filtering and pseudoinverse filtering:

(I) Suppose first that the additive noise n( x, y) is negligible. A problem arises if
H (u, v) becomes very small or zero for some point (u, v) or for a whole region in
the (u, v) plane. In that region inverse filtering cannot be applied. Note that in
most real applications H (u, v) drops off rapidly as a function of distance from the
origin. The solution is that if these points are known they can be neglected in the
computation of F (u, v) .

(II) In the presence of external noise we have that

H  (u , v) ( G (u , v) + N (u , v) ) H  (u, v)G (u, v) H  (u, v) N (u, v)
Fˆ (u , v) = 2
= 2
+ 2

H (u , v) H (u, v) H (u, v)
N (u, v)
Fˆ (u, v) = F (u, v) +
H (u , v)
If H (u, v) becomes very small, the term N (u, v) dominates the result. The solution
is again to carry out the restoration process in a limited neighborhood about the
origin where H (u, v) is not very small. This procedure is called pseudoinverse
filtering. In that case we set
 H  (u, v)G (u, v)
 2
H (u , v)  T
 H (u , v)

Fˆ (u, v) = 
0 H (u, v)  T



The threshold T is defined by the user. In general, the noise may very well possess
large components at high frequencies (u, v) , while H (u, v) and G(u, v) normally will be
dominated by low frequency components.

Dr Ganesh V Bhat (Dept. of ECE Canara Engineering College India) Page 12

MODULE-4 VII SEM DIGITAL IMAGE PROCESSING [17EC72]

Constrained least squares (CLS) restoration

It refers to a very large number of restoration algorithms. The problem can be
formulated as follows.
minimize
J (f ) = n(f ) = g − Hf
2 2

subject to

2
Cf
where Cf is a high pass filtered version of the image. The idea behind the above
constraint is that the high pass version of the image contains a considerably large
amount of noise! Algorithms of the above type can be handled using optimization
techniques. Constrained least squares (CLS) restoration can be formulated by choosing
an f to minimize the Lagrangian
(
min g − Hf +  Cf
2 2
)
Typical choice for C is the 2-D Laplacian operator given by
 0.00 − 0.25 0.00 
C = − 0.25 1.00 − 0.25
 
 0.00 − 0.25 0.00 
 represents either a Lagrange multiplier or a fixed parameter known as regularisation
parameter and it controls the relative contribution between the term g − Hf
2
and the
2
term Cf . The minimization of the above leads to the following estimate for the original
image
( )
−1
f = HT H +  CTC H Tg

Issues concerning the CLS method:

Choice of 
The problem of the choice of  has been attempted in a large number of studies and
different techniques have been proposed.
With larger values of  , and thus more regularisation, the restored image tends to have
more ringing. With smaller values of  , the restored image tends to have more

Dr Ganesh V Bhat (Dept. of ECE Canara Engineering College India) Page 13

MODULE-4 VII SEM DIGITAL IMAGE PROCESSING [17EC72]

amplified noise effects. The variance and bias of the error image in frequency domain
are
2
M N
H (u , v)
Var ( ) =  2

( H (u, v) )
n
2 2
+  C (u , v )
2
u =0 v =0

F ( u, v )  2 C (u, v )
2 4
M −1 N −1
Bias( ) =    2
n
u =0 v =0
( H ( u, v ) 2
+  C ( u, v )
2 2
)
The minimum MSE is encountered close to the intersection of the above functions. A
good choice of  is one that gives the best compromise between the variance and bias
of the error image.

Wiener estimator (stochastic regularisation)

The image restoration

problem can be viewed as a
system identification
problem as follows:

The objective is to minimize the following function

 

 ( f ( x. y ) − fˆ ( x. y ) dxdy )
2
=
− −

from Parseval ' s thereom

  2

=   f ( x. y ) − fˆ ( x. y )
− −
dxdy

  2

=   F ( x. y ) − Fˆ ( x. y )
− −
dudv

To do so the following conditions should hold ie (output can be estimated as weighted

sum of input and error must be orthogonal to the observation about the mean):
(i) F̂ = WG = WHF + WN
(ii) F − Fˆ = (1 − WH ) F − WN

From (i) and (ii) we have that

  2

= 
− −
(1 − WH ) F − WN dudv

   (1 − WH ) F dudv ;
 
= − WN f ( x, y ) and  ( x, y ) are uncorrelated
2 2
sin ce
− −

Dr Ganesh V Bhat (Dept. of ECE Canara Engineering College India) Page 14

MODULE-4 VII SEM DIGITAL IMAGE PROCESSING [17EC72]

Minimize the above equation by differentiating it and evaluate it to zero w.r.t the filter
coefficients we get (Note that the above equation is integrated sum of two squares i.e

z
( zz ) = 2z )
* *


z
⎯⎯ ((
→ 2 − 1 − W *H * H F + W * N ) 2 2
)=0
2
H F H*
W = *
=
F +N H +N
2 2 2 2 2 2
H F
H * ( u, v ) S ff ( u , v ) H * ( u, v ) H * ( u, v )
W* = = =
H ( u, v ) S ff ( u, v ) + S ( u , v ) H ( u , v ) + S ( u , v ) S ff ( u , v ) H (u , v ) + K
2 2 2

F(u,v)| and |N(u,v)| are known

approximately, or K is a constant
(w.r.t. u and v) chosen empirically
to our knowledge of the noise level.
Wiener filter so obtained is more
robust to noise, and preserves
high-frequency details.
Note: The filter obtained in
frequency domain can also be
represented in spatial domain
using the relation between thee
correlation and power spectrum Winer filtes can also be represented as
W = Rfg Rgg −1 = Rff HT (HRff HT + Rnn )−1
Computational issues : The noise variance has to be known, otherwise it is estimated
from a flat region of the observed image. In practical cases where a single copy of the
degraded image is available, it is quite common to use S yy (u, v) as an estimate of
S ff (u, v) . This is very often a poor estimate.
Wiener smoothing filter: In the absence of any blur, H (u, v) = 1 and
S ff (u, v) (SNR)
W (u, v) = =
S ff (u, v) + Snn (u, v) (SNR) + 1
(i) ( SNR)  1  W (u, v)  1
(ii) ( SNR)  1  W (u, v)  ( SNR)
( SNR) is high in low spatial frequencies and low in high spatial frequencies so W (u, v)
can be implemented with a lowpass (smoothing) filter.

Dr Ganesh V Bhat (Dept. of ECE Canara Engineering College India) Page 15

MODULE-4 VII SEM DIGITAL IMAGE PROCESSING [17EC72]

1
Relation with inverse filtering: If Snn (u, v) = 0  W (u, v) = which is the inverse filter
H (u, v)
 1
 H (u, v) H (u, v)  0

If S nn (u, v) → 0  lim W (u, v) = 
Snn →0
 0 H (u, v) = 0


which is the pseudoinverse filter.

University Theory Questions

1. With a block diagram, briefly explain the model of image degradation- restoration
process.

2. Discuss various mean filters and order statistics filters in image restoration system
3. Explain the Weiner filtering method of restoring images
4. Justify the statements “median filter is an effective tool to minimize salt and pepper
noise” with an example.

5. Discuss the concept of inverse filtering and its drawbacks

6. Explain the process for removal of periodic noise in an image
7.
8.

9.

10.

FURTHER READING

Further reading: http://www.ent.mrt.ac.lk/~ranga/courses/en5204_2013/L01.pdf

NPTEL course Link: https://nptel.ac.in/courses/117/104/117104069/

Dr Ganesh V Bhat (Dept. of ECE Canara Engineering College India) Page 16