UNIT-3

IMAGE RESTORATION
IMAGE RESTORATION:
Kestoration improves image in some predefined sense. It is an
Restoration attempts to reconstruct an image that has been degradedobjective process.
by using a priori
knowledge of the degradation phenomenon. These
modeling the degradation and then techniques are oriented toward
applving the inverse process in order to recover the
Onginal image. Restoration techniques are based on mathematical or probabilistic models
OT image processing. Enhancement, on the other hand is based on human subjective
Preferences regarding what constitutes a "good" enhancement result. Image Restoration
refers to a class of methods that aim to remove or reduce the degradations that have
occurred while the digital image was being obtained. All natural images when displayed
have gone through some sort of degradation:
During display mode
Acquisition mode, or
Processing mode
Sensor noise
Blur due to camera mis focus.

Relative object-camera motion

Random atmospheric turbulence
Others

Degradation Model:
an input
Degradation processoperates on a degradation function that operates on
by using the notation f(x,y),
image with an additivenoise term. Input image is represented
terms when combined gives the result
noise term can be represented as n(x,y).These two
g(x,y). If we are given g(x,y), some knowledge about the degradation function H or J
as
some knowledge about the additive noise teem n(x,y), the objective of restoration is
and
estimate f(,y) of the original image. We want the estimate to be as close as
to obtain an
original image. The more we know about h and n,the closer f(x,y)will be
possible to the
invariant process, then degraded image is given in the
tof(x.y). If it is a linear position
spatialdomain by
g(x,y)=f(x,y)*h(x,y+n,y)
 Ay) 1s spatial representation of degradatjon function and symbol * represents
convolution. In frequency domain we may write this equation as
G(u,v)=-F(u,v)H(u,v)+N(u,v)
tems in the capital letters are the Fourier Transform of the corresponding terms in
the spatial domain.

f[x, y) Degradation 8(r. y) Restoration

function filter(s)

Noise

DEGRADATION RESTORATION

Fig: Amodel of the image Degradation/ Restoration

process
Noise Models:
The principal source of noise in digital images arises during image acquisition and
lor transmission. The performance of imaging sensors is affected by avariety of factors,
such as environmental conditions during image acquisition and by the quality of the
sensing elements themselves. Images are corrupted during transmission principally due to
interference in the channels used for transmission. Since main sources of noise presented
in digital images are resulted from atmospheric disturbance and image sensor circuitry,
following assumptions can be made i.e. the noise model is spatial invariant (independent
of spatial location). The noise model is uncorrelated with the object function.
Gaussian Noise:
tractability in both spatial
These noise models are used frequently in practices because of its
random variable is
and frequency domain. The PDF of Gaussian

if as z sb
P(:) -b-a otherwise
deviation.
mean of average value of z, o= standard
Where z represents the gray level, u=

Page51
 P()|

(Gusim

Q.C07

Rayleigh Noise:
no symmetric. It is given by
Unlike Gaussian distribution, the Rayleigh distribution is
the formula.

a)e-*-abzza

The mean and variance of this density is

b(4 -)
4

Ruyligk

(ii) Gamma Noise:

The PDF of Erlang noise is given by

-e,forz20
P(2)={(6-1
0, for z <0
The mean and variance of this density are given by
b
mean: u=- variance: ¡ =

Page 52
 ('amrna

(b-1)1

gamma density
This equation is referredto as
Itsshape is similarto Rayleigh disruption.
gamma function.
it iscorrect only when the denominator is the
(iv) Exponential Noise:
shape. The PDF of exponential noIse 1S given as
Exponentialdistribution has an exponential
ae z0
Páz) o z<0

Where a>0. The mean and variance of this density are given by
1

Exponential

(v) Uniform Noise:

The PDF of uniform noise is given by
1
if as z sh
otherwise

The mean and variance of this noise is

m
a t b (6- a
2 12

Page 53
 P(z)|

Unilorm

(vi) 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
P for z = a
P forzb
otherwise
b>a
Ifb>a, gray levelb willappear as a light dot in image. Level a will appear like a dark dot.

P
Impulse

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)tn(x,y)
or

Gu,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:

Page 54
 i)) Mean Filter:
i) (a)Arithmetic Mean
filter:
*s ue Simplest mean filter. Let Sxv renresents the set of coordinates in the sub
Image of sizem*n centered at noint (y v) The arithmetic mean filter computesthe average
Vanue of the corrupted image oly v) in the areg defined by Sxy. The value of the restored
mage f at any point (x,y) is the arithmetic mean computed using the pixels in the region
defined by Sxy.
1
jr.y)=
(s.te Sy
This operation can be using a convolution mask in
which all coefficients have
value 1/mn A mean filter smoothes local
yariations 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.
(b)Geometric Mean filter:
An image restored using a geometric mean filter is
given by the expression

I|g(3.)
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.
(c) Harmonic Mean filter:
The harmonic mean filtering operation isgiven by the expression

jr,y) =(st)es(s,nEs(s.1P
Ssy (s,i)e Say
The harmonic mean filter works well for salt noise but fails for pepper noise. It does
wellwith Gaussian noise also.
(d) Order statistics filter:
Order statistics filters are spatial filters whose response is based on ordering the pixel
contained inthe image area encompassed by the filter. The response of the filter at any
point is determined by the ranking result.

Page 55
 (e) Median filter:
It is the
best order statistic
levels in the ffiilter; it replaces the value of a pixel by the
Neighborhood of the pixcl. median of gray

f(x.y) = median (g(s.)}

The original of the iis
pixel included in the
POssible because for certain types of randomcomputation of the median of the filter are quite
noise, the provicde excellent noise
capabilities with considerably less blurringthen smoothing reduction
are effective for bipolar and filters of similar size. These
(e) Max and
unipolor impulse noise.
Min filter:
Using the 100th percentile of ranked set of numbers is
called the max filter and is given
by the equation

Îo)=, max (g(s, i)}

It isused for finding the brightest point in an
(s.)e Soy
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 Oth
percentile filter is min filter.

fy) = min (g(s.)}

(s,f)e Sy
This filter is useful for flinging the darkest point in image. Also, it
reduces salt noise
of the min operation.
() Midpoint filter:
The midpoint filter simply computes the midpoint between the maximum and minimum
values in the area encompassed by

max (g(s,t)}+ min

(s,)e Sy (seSy
It comeliness the order statistics and averaging ThÉs filter works best for randomly
distributed noise like Gaussian or uniform noise.
Periodic Noise by Frequency domain filtering:
These types of filters are used for this purpose
Band Reject Filters:
It removes a band of frequencies about the origin of the Fourier transformer.
Ideal Band reject Filter:

Page 56
 An ideal band reject filter is
given by the expression
if D{u, v)<D, -W/2
if D,-WI2s Du,»)s D, +WI2
{1 if D{u, v)>D, +W/2
D(u,v)- the distance from the originof the
W- the
centered frequency rectangle.
width of the band
Do- the radial center of the
frequency rectangle.
Butterworth Band reject Filter:

Gaussian Band reject Filter:

D'u)-)
Hu,v)=1- exp
D(u,) W
These filters are mostly used when the location of noise
component in the frequency
domain is known. Sinusoidal noise can be easily removed by using these kinds
of filters
because it shows two impulses that are mirror images of each other about the origin. Of
the frequency transform.

FIGURE From left to right, perspective plots of ideal, Butterworth (of order 1), and Gaussian bandreject
filters
Band pass Filter:
The function of a band pass filter is opposite to that of a band reject flter It allows aspecfic
frequency band of the image to be passed and blocks the rest of frequencies. The transfer
function of a band pass filter can be obtained from a corresponding band reject filter with
transfer function Hbr(u,v) by using the equation

Hgp(4, v)=l-H(u,v)

Page 57
 Inese filters cannot be abplied directly onan imagebecause it may remove too much details
Ol an mage but these are effective in jsolating the effect of an image of selected frequency
bands.
Notch Filters:
neighborhoods about a
A notch filter rejects (or passes) frequencies in predefined
center frequency.
must appear in symmetric
Due to the symmetry of the Fourier transform notch filters
pairs about the origin.
of radius Do with centers a (uo, vo)
The transfer function of an ideal notch reject filter
and by symmetry at (-u0, vo) 1s

D,(u,v) =Ju-M/2-u,)? +(v-N/2-Y,)'

D,(u, v)=(u-M/2+1,)² +(u-NI2+v,'
Ideal, butterworth, Gaussian notch filters
if D{u,v) SD, or D,(u,v) sD,
|1 otherwise

H(u, )

Hu, r) H(4, v)

FIGURE Perspective plots of (a) ideal, (b) Butterworth (of order 2), and (c) Gaussian
notch (reject) filers.

Page 58
 Inverse Filtering:
esimplest approach to restoration is direct inverse filtering where we complete an
estimate Fu. e)
of the transform of the original image simply by dividing the transtorm
of the degraded image G(u,v) by
degradation function H(u,v)
Glu. v)
Hu, v)
We know that

Gu,v)=H(4,v) F(u,v) +N(4)

Therefore

F(u, v) =F(u, v) + N(u, v)

H(u. v)
From the above equation we observe that we cannot recover the
undegraded image
exactly because N(u,v) is a random function whose Fourier transform is not known.
One approach to get around the zero or small-value problem is to limit the filter
frequencies to values near the origin.
We know that H(0,0) is equal to the average values of h(x,y).
By Limiting the analysis to frequencies near the origin we reduse the probability of
encountering zero values.
Minimum mean Square Error (Wiener) filtering:
The inverse filtering approach has poor performance. The wiener filtering approach
uses the degradation function and statistical characteristics of noise into the restoration
process.

The objective is tofindan estimate J of the uncorrupted imagef such that the mean
square error between them is minimized.
The error measure is given by

e'= E{[f()-•n)
Where E{.} is the expected value ofthe argument.
zero
We assume that the noise and the image are uncorrelated one or the other has
mean.
levels in the degraded
The gray levels in the estimate are a linear function of the
image.
Page 59
 H'(u, v)S, (u, v)
Ls,(u, v)lH(u, v)|? +S,(u. v) J
H'u, v)

Where H(u,v)= degradation function

H*(u,v)-complex conjugate of H(u,v)
|H(u,v)-H* (u,v) H(u,v)
Sa(4,v)-N(u,v)=power spectrum ofthe noise
S{u,v-F(u,v)|=power spectrum of the underrated image
The power spectrum of the undegraded image is rarely known. An approach used
frequently when these quantities are not known or cannot be estimated then the
expression used is

Hu, v) |Hu, v)l + K

Where K is a specified constant.

Constrained least squares filtering:

The wiener filter has a disadvantage that we need to know the power spectra of the
undegraded image and noise. The constrained least square filtering requires only the
knowledge of only the mean and variance of the noise. These parameters usually can
be calculated from a given degraded image this is the advantage with this method. This
method produces a optimal result. This method require the optimal criteria which is
important we express the
g(r. y) =, y) *fn, y) + n*. y) in vector-matrix form

The optimality criteria for restoration is based on a measure of smoothness, such as the
second derivative of an image (Laplacian).
The minimum of a criterion function C defined as

Page 60
 Subject to the constraint

I- HtF-tf
Where w?
is a euclidean vector norm f is estimate of the undegraded
image. is laplacian operatot.
The frequency domain solution to this optimization problem is given by

H'(u,e)

Where y is a parameter that must be adjusted so that the constraint is satisfied.

P(u,v) is the Fourier transform of the laplacian operator

-1 0

Page 61