International Journal of Research in Advent Technology, Vol.2, No.

3, March 2014
E-ISSN: 2321-9637

A Comprehensive Review on Image Restoration Techniques

Biswa Ranjan Mohapatra, Ansuman Mishra, Sarat Kumar Rout
Department of CSE, BIET, Bhadrak
Odisha

Abstract- Image restoration is an art to improve the quality of image via estimating the amount of noises and blur involved in
the image. With the passage of time, image gets degraded due to different atmospheric and environmental conditions, so it is
required to restore the original image using different image processing algorithms. There is a wide spread application of image
restoration in today’s world. Application area varies from restoration of old images in museum and radar based image
acquisition and restoration. This paper gives a review of different image restoration techniques used.

Keywords- Blur, image restoration, image acquisition

1. INTRODUCTION n ( i, j ) the external noise which is assumed to be image-
Image restoration is based on the attempt to improve the independent.
quality of an image through knowledge of the physical
process which led to its formation. The purpose of image
restoration is to "compensate for" or "undo" defects which
degrade an image. Degradation comes in many forms such as
motion blur, noise, and camera mis-focus. In cases like
motion blur, it is possible to come up with a very good
estimate of the actual blurring function and "undo" the blur to
restore the original image. In cases where the image is Figure 1: Image degradation and restoration techniques
corrupted by noise, the best we may hope to do is to
compensate for the degradation it caused. Image restoration Noise Models
differs from image enhancement in that the latter is concerned
more with accentuation or extraction of image features rather In image processing there are different noise models
than restoration of degradations. Image restoration problems available.
can be quantified precisely, whereas enhancement criteria are Gaussian Noise can be represented as
difficult to represent mathematically. ( z − µ )2
1 −
Image restoration started in 1950’s. There are several p( z) = e 2σ 2
(2)
application domain of image restoration like scientific 2πσ
exploration, legal investigations, film making and archivals, Rayeligh’s noise can be represented as
image and video decoding and consumer photography. The ( z − µ )2
2 −
main area of application is image reconstruction in radio p ( z) = ( z − a)e b
(3)
astronomy, radar imaging and tomography. b
This paper discusses the importance of image restoration Salt and Pepper noise can be represented as
p ( z ) = Paδ ( z − a ) + Pbδ ( z − b )
techniques and reviews different image restoration techniques
(4)
available.

2. IMAGE RESTORATION There are different kinds of blurs involved in images some of
Image restoration uses a priori knowledge of the the blurs are discussed below.
degradation. It models the degradation and applies inverse
process. It formulates and evaluates the objective criteria of
goodness. The distortion can be modelled as noise or a Blur Models
degradation function. To restore an image from a noise model, Motion blur occurs when there is relative motion between the
different filters like median filter, homomorphic filters are object and the camera during exposure. This can be in the
used. To get rid of periodic noises, butterworth lowpass filter, form of a translation, a rotation, a sudden change of scale, or
butterworth band reject filters and notch filters are used. To some combinations of these.
restore an image from linear degradation, inverse and pseudo 1 L L
 − ≤
h (i ) =  L
inverse filtering, wiener filtering and blind de-convolution are
used. 2 2 (5)
A simplified version for the image restoration process  0 else
model isy ( i, j ) = H  f ( i, j )  + n ( i, j ) (1) Atmospheric turbulence occurs due to random variations in
the reflective index of the medium between the object
Where y ( i, j ) is the degraded image, f ( i, j ) is the original and the imaging system and it occurs in the imaging of
image, H an operator that represents the degradation process, astronomical objects.

101
 International Journal of Research in Advent Technology, Vol.2, No.3, March 2014
E-ISSN: 2321-9637

 i2 + j 2  original Image median filtered Image

h ( i, j ) = K exp  − 2 
(6)
 2σ 
When a camera images a 3-D scene onto a 2-D imaging plane,
some parts of the scene are in focus while other parts are not.
If the aperture of the camera is circular, the image of any
point source is a small disk, known as the circle of confusion
(COC). The degree of defocus (diameter of the COC) depends
on the focal length and the aperture number of the lens, and
the distance between camera and object. An accurate model
not only describes the diameter of the COC, but also the Figure 2: Noisy Image and Median filtered image
intensity distribution within the COC. However, if the degree
of defocusing is large relative to the wavelengths considered, original Image median filtered Image
a geometrical approach can be followed resulting in a uniform
intensity distribution within the COC. Uniform out of focus
 1
 i2 + j 2 ≤ R
blur is defined by h ( i, j ) =  π R (7)
 0 else
Uniform 2-D blur is defined by
1 L L median filter restoration
 − ≤ i, j ≤
h ( i ) =  L2 2 2 (8)
 0 else
Performance Indices
Blurred Signal-to-Noise Ratio (BSNR) is a metric that
describes the degradation model.
 1 
∑∑ h ( i, j ) − h% ( i, j )
2

 MN 
BSNR = 10 log10  
i j

 σ n2 
 
 
(9) Figure 3: Image degradation and restoration techniques
Here h ( i, j ) = y ( i, j ) − n ( i, j ) , h% ( i, j ) = E  h ( i, j )  Figure 3 show the same noisy image heavily corrupted by salt
and σ 2
n is variance of additive noise. Improvement in SNR and pepper noise and median filter of window size (9x9) is
used. The higher the window size of median filter there is a
(ISNR) validates the performance of the image restoration higher chance of image degradation.
algorithm.
  f ( i, j ) − y ( i, j )  
∑∑
2
Table 1: PSNR value (dB) for % of salt and pepper noise
 i j  (Image: Cameraman, Filter: Median)
ISNR = 10 log10  2 
(10)
 ∑∑  f ( i, j ) − f% ( i, j )   10% 20% 50% 90%
 i j  1 MF (3X3) 33.7 27.28 14.12 6.54
where f% ( i, j ) is the restored image. 2 MF (5X5) 31.44 30.67 12.88 7.29
3 MF (9X9) 29.47 28.99 16.54 8.09
Median Filter To restore an image from linear degradation inverse
Figure 2 shows an image, heavily corrupted by salt and filter, pseudo inverse filter, Weiner filter and blind de-
pepper noise and 3x3 median filtered is used to remove the convolution is used. These techniques are discussed below.
noise.
Inverse Filtering

102
 International Journal of Research in Advent Technology, Vol.2, No.3, March 2014
E-ISSN: 2321-9637

1
H% ( u, v ) =
show noisy with degraded image
Inverse filter can be expressed as and
H ( u, v )
the recovered image can be expressed as
F% ( u, v ) = G ( u, v ) H% ( u, v )
Degraded Image Restored Image

Figure 4: Image degradation and restoration using inverse

filter
Pseudo Inverse Filter
pseudo-inverse restoration
Pseudo inverse filter can be expressed as

 1
 H ( u, v ) ≥ ε
H% ( u , v ) =  H ( u , v ) (11)
 0
 H ( u, v ) < ε

Degraded Image Restored Image

Figure 6: Image degradation and restoration using pseudo

Figure 5: Image degradation and restoration using pseudo inverse filter
inverse filter
Weiner Filter
Blurred Image
The main disadvantage of Weiner filter is that it
can’t handle noises. So minimum mean square error filtering
(Weiner filter) is used which incorporates both the
degradation function and statistical characteristics of noise in
to image restoration process. In this method it is assumed that
the noise and degradation function are uncorrelated. One of
them has zero mean. The objective function of Weiner filter is
as follows,
min E f ( x, y ) − f% ( x, y ) ≡ min Cee (ω x , ω y ) (12)
2

hr ( x , y ) hr ( x , y )

H% (ω x , ω y )
H r (ω x , ω y ) =
Cnn (ω x , ω y )
(13)

H (ω x , ω y ) +
2

C ff (ω x , ω y )
The main disadvantage of Weiner filter is that the power
spectra of undergraded image and power spectra of noise
must be known.

103
 International Journal of Research in Advent Technology, Vol.2, No.3, March 2014
E-ISSN: 2321-9637

1−α
Restoration of Blurred, Noisy Image Using Estimated NSR  
 H ( u, v ) 
α   (14)
 H ( u, v ) 
F ( u, v ) =  
 G ( u, v )
 H ( u, v ) 2   Sn ( u, v )  
   H ( u, v ) + β 
2


 
 S f ( u , v )  
Geometric Mean filter restoration

Restoration of Blurred, Quantized Image Using Computed NSR

Figure 8: Image degradation and restoration using geometric

mean filter restoration

Blind Deconvolution Algorithm

The Blind Deconvolution Algorithm can be used

effectively when no information about the distortion (blurring
and noise) is known. The algorithm restores the image and the
point-spread function (PSF) simultaneously. The accelerated,
damped Richardson-Lucy algorithm is used in each iteration.
Additional optical system (e.g. camera) characteristics can be
used as input parameters that could help to improve the
wiener filter restoration quality of the image restoration. Blind deconvolution is the
problem of recovering a sharp version of an input blurry
image when the blur kernel is unknown. Mathematically
y=k⊗x
Where x is a visually plausible sharp image, and k is a non
negative blur kernel, whose support is small compared to the
image size.
Blurred Image

Figure 7: Image degradation and restoration using Weiner

filter

Geometric Mean Filter

Geometric mean filter is the generalization form of Weiner

filter. The mathematical expression of geometric mean filter
is defined as

104
 International Journal of Research in Advent Technology, Vol.2, No.3, March 2014
E-ISSN: 2321-9637

di
u tj+1 = u tj ∑
Deblurring with Undersized PSF
pij (16)
i ci

 d 
u tj+1 = u tj  t ⊗ pˆ 
u ⊗ p 
(17)

A = Blurred and Noisy deconvlucy(A,PSF)

Deblurring with Oversized PSF

deconvlucy(A,PSF,NI,DP) deconvlucy(A,PSF,NI,DP,WT)

Figure 10: Image degradation and restoration using

Richardson-Lucy blind deconvolution algorithm

3. CONCLUSIONS
This paper gives a review of different image restoration
Deblurring with INITPSF
algorithms. Image restoration is an active research area and
various researchers work to improve the efficiency of the
different algorithms by developing more efficient algorithms.
But primarily image restoration is done mostly using Weiner
filter, Richardson-Lucy Blind Deconvolution algorithm,
Inverse and Pseudo-inverse filter.

REFERENCES

1. D A Fish, A M Brinicombe, E R Pike, “Blind deconvolution by means

of the Richardson-Lucy algorithm,” J. Opt. Soc. Am. A., vol. 12, no. 1,
Jan 1995, pp 58 – 65
2. D. Kundur and D. Hatzinakos, “Blind image deconvolution,” IEEE
Signal Processing Magazine, 1996.
3. C Khare, K K Nagawanshi, “Implementation and analysis of image
restoration techniques,” International Journal of Computer Trends and
Technology, May 2011
Figure 9: Image degradation and restoration using blind 4. H C Andrew, B R Hunt, “Digital image restoration,” Prentice Hall
deconvolution algorithm 5. K. R. Castleman, “Digital Image Processing”, International Edition,
Prentice-Hall, Inc., 1996.
6. Neelamani R., Choi H., and Baraniuk R. G., "Forward: Fourier-wavelet
Richardson-Lucy Deconvolution Algorithm regularized deconvolution for ill-conditioned systems", IEEE Trans. on
The Richardson–Lucy deconvolution algorithm has Signal Processing, Vol. 52, No 2 (2003) 418-433.
become popular in the fields of astronomy and medical 7. C. Helstrom, “Image Restoration by the Method of Least Squares”, J.
Opt. Soc.Amer., 57(3): 297-303, March 1967.
imaging. Initially it was derived from Bayes’s theorem in the 8. Amandeep Kaur, Vinay Chopra, “A comparative study and analysis of
early 1970’s by Richardson and Lucy. image restoration techniques using different image formats,”
International Journal of Science and Emerging Technologies with latest
Pixels in the observed image can be represented by trends, 2, 1, pp. 7-14, 2012
9. P. Campisi and K. Egiazarian, “Blind image deconvolution theory and
d i = ∑ pij u j (15) applications , CRC Press,2006.
j 10. K. H. Yap and L. Guan, “A computational reinforced learning scheme
to blind image deconvolution , ” IEEE Transactions on Evolutionary
Computation, vol. 6, no. 1, pp. 2-15, 2002.

105