ISSN 2320-2602

International Journal of Advances in Computer Science and Technology (IJACST), Vol.3 , No.12, Pages : 01-08
Special Issue of ICCEeT 2014 - Held on 22nd December 2014, Dubai

A Comparative Study On Image Deblurring Techniques

I. M. El-Henawy1 ,A. E. Amin2, Kareem Ahmed3, Hadeer Adel4
1

Computer Science Department, Faculty of Computer Science, ZAGAZIG university,[email protected]

2
Computer Science Department, Faculty of Science, Minia University,[email protected]
3
Multimedia Department, Faculty of Computer Science, BSU University,[email protected]
3
Computer Science Department, Faculty of Computer Science, NUB University, [email protected]

AbstractImage blur is a common problem that occurs when

recording digital images due to camera shake, long exposure time,
or movement of objects. As a result, the recorded image is degraded
and the recorded scene becomes unreadable. Recently, the field of
blur removal has gained increasing interest in a lot of researches.
The problem is known as blind deconvolution if the only available
information is the blurred image and there is no knowledge about
the blurring model or the Point Spread Function (PSF). In this case,
the basic target of the process is to recover both the blur kernel and
the deblurred (latent) image, simultaneously. In this paper, we
introduced a comprehensive study on the image deblurring, type of
blur, noise model and finally a comparative study of different image
deblurring techniques. We performed several experiments to
evaluate these techniques in terms of performance, blur type, Peak
Signal to Noise Ratio and structural similarity (SSIM).

Key words : Image Deblurring, blur, PSF, degradation model,

image enhancement and restoration, PSNR.

INTRODUCTION
Images are widely used in many kinds of applications such
as everyday photography, monitoring, medical imaging,
astronomy, microscopy, and remote sensing. Digital images
are composed of picture elements or pixels that are organized
in a grid. Each pixel contains an intensity value which
determines the tone at a specific point. Unfortunately, all
captured images end up more or less blurry. The motion of
objects or the vibration of the sensor (camera) when pressing
the shutter causes the image to be blurred. There are many
factors that cause the blurring or degradation of the digital
image, such as movement during the capture process, using
long exposure times, using wide angle lens, etc.[2]. However,
there are two main causes for motion blur: (i) the image is
blurred by the camera vibration which causes all pixels in the
image to be affected, and (ii) the image is blurred by object
motion which causes a specific region to be blurred. Image
blur usually devastate the images, and practically it is hard to
avoid it because there is a lot of interference in the
environment. Image deblurring is the process of applying and
solving mathematical models to recover the original (sharp)
image. Recently, image deblurring and restoration become
the main and the most important sub field of digital image
processing. Image deblurring is used to make images sharp
and useful. There are many applications for image deblurring
such as:

Recovering valuable photographs.

Watching distant star fields through ground based telescopes.

Watching space vehicles and satellites.

Radar imaging
Tomography and medical imaging
Microscopy
Iris recognition

Mathematically, the blurring can be defined as a linear

process that leads to image degradation. The blurring model
can be described as a combination of two operations; the first
operation is convolving the original (unblurred or sharp)
image with unknown kernel; the second operation is the
addition of some noise to the resulting image. Equation 1
introduces the notations that are used in the blurring
process[1].
=

(Eq. 1)

Where B is blurred image, I is the unblurred image, K is

unknown linear shift-invariant point spread function (PSF) or
impulse response, and N is the additive noise. For example, if
the original (unblurred) image contains a single bright
(white) pixel and all other pixels are black. When capturing
this image, the single bright pixel is spread over its
neighboring pixels. This single bright pixel is called a point
source. The point spread function (PSF) is a function that
describes the response of an imaging system to a point
source. Point Spread Function (PSF) is the degree to which an
optical system blurs (spreads) a point of light[26]. Optical
Transfer Function (OTF) is the Fourier transfer of the point
Spread Function (PSF) and the PSF is the inverse Fourier
transform of OTF. In the frequency domain, the OTF
describes the response of a linear, position-invariant system
to an impulse[27]. Image deblurring can be viewed as a
process that tries to obtain an approximation of I . One
possible solution to obtain the unblurred image is the blind
deconvolution which requires the estimation of K using
image intensities and gradients to deconvolve B. Equation 1
contains a term for the noise that are consequences of
problems associated with acquisition devices.
In the literature, several image restoration techniques have
been proposed, some of these approaches are non-blind or
they require a priori knowledge to estimate the blur
filter(PSF), and the others are blind which means that they
can perform image Deblurring without a priori knowledge of
the blur filter(PSF). In this paper we study different
approaches and methodologies for image deblurring.
BLURRING
It is well known that an image is clear if we can perceive the
shape of all its objects correctly. For example, a face image is
clear if we can recognize lips, eyes, nose, etc. The shapes of
contained objects are recognized from its edges. By contrast,
blurring means reducing the edge content and making the
transition from one color to another very smooth. One may

ISSN 2320-2602
International Journal of Advances in Computer Science and Technology (IJACST), Vol.3 , No.12, Pages : 01-08
Special Issue of ICCEeT 2014 - Held on 22nd December 2014, Dubai
think that zooming is one type of blurring. Actually, if an
image is zoomed using pixel replication and large zooming
factor, the resulting image is a blurred image. however, the
difference between zooming and blurring is that the number
of pixels in the zoomed image is larger than the number of
pixels in the original image, but the number of pixels in the
original image and blurred one is the same. Blurring can be
thought of as applying some filter to an image. There are
many kind of filters that can be used to perform blurring, such
as average filter, weighted average filter, Gaussian filter and
motion filter. The average filter has three properties: it is odd
ordered, the sum of all its elements equal 1, and all its
elements are the same. The amount of the blurring is
increased when the size of the kernel is increased. That is
because a large number of pixels are included and one
smooth transition is defined. The weighted average filter
gives more weight to the center value. As a result, the
contribution of the center pixel becomes more than the rest of
the pixels. The Gaussian blur uses a Gaussian function to
calculate the transformation that will be applied to each pixel
in the image. It is low-bass pass filter which is always used to
reduce noise and enhance image structure at different scales.
In Gaussian blur, the pixel weights are decreased from kernel
centre to edges according to a bell-shaped curve. The
Gaussian filter blends a specific number of pixels
incrementally, following a bell-shaped curve. The blurring is
dense in the centre and feathers at the edge[3]. It can be used
to give more control over the Blur effect because the
Gaussian blur depends on the Size and Alfa. Motion blur is
the apparent streaking of rapidly moving objects in a still
image[5]. It occurs when the image being recorded is
changed during the recording either due to rapid movement
or long exposure. The basic reason for all motion blur is the
relative motion between the recording device and the scene in
the form of translation, rotation, sudden change of scale, or a
combinations of them [2]. The Motion Blur is a filter that
adds a blur in a specific direction, so the image appears to be
moving. The motion is controlled by an angle (0 to 360
degree) or direction (-90 to 90) and by distance or intensity in
pixels. There are many other types of blur types such as
out-of-focus blur. If a camera is used to convert a 3D real
scene into a 2D image, 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)[6]. The degree of defocus
(diameter of the COC) depends on several factors including:
the focal length, the aperture number of the lens, and the
distance between the camera and the scene.

relation between the blurred and sharp images is expressed

as:
B = A IA
(Eq. 2)
The term A I denotes applying the same vertical blurring
operation to all columns of I. The term I A denotes applying
the same horizontal blurring to all rows of I. The very nave
solution to this linear model is:
I = A B (A )
(Eq. 3)
Unfortunately, the reconstructed image does not resemble the
desired image because we have omitted the noise term. Image
noise is a random variation of intensities (color values) in the
image and is usually generated by the sensor and the circuitry
of the camera. Image noise introduces a set of spurious and
extraneous information, so it should be considered in the
linear blurring model as expressed in equation 4.
B=A IA +E
(Eq. 4)
Where the noise image E is also of size m n. So the sharp
image can be reconstructed using equation 5.
I = A B (A ) A E (A )
(Eq. 5)
The term A E (A ) is informally called the inverted
noise which measures the contribution of the additive noise in
the reconstruction of the sharp image. The more the value of
this term, the more the inverted noise dominates the
reconstruction process. Practically, It is noticed that the
inverted noise always dominates the reconstruction process.
SINGULAR VALUE DECOMPOSITION
In this section, we will consider a blurring model [7] that is
more general than the previous one. Here, the blurring is
performed on the rows and columns of the image
simultaneously. The first step is to reshape the matrices I and
B to be two column vectors of length N = mn by stacking the
columns of these matrices into two long column vectors I
and B . The general linear blurred model can be described
as a relation between the sharp column vector I
and
blurred column vector B as expressed in equation 6.
=

(Eq. 6)

At this point, we will assume that A is a known blurring

matrix of size N N, but in practice, it is constructed from
the imaging system. The term e denotes the column vector of
the noise image E. So, the sharp image can be reconstructed
by solving equation 7.
=

(Eq. 7)

Where the term A e is the inverted noise. If the deblurred

image is very distorted, it means that the inverted noise term
dominates the reconstructed image. The blurring matrix A
can be decomposed into two orthogonal matrices U and V,
and a diagonal matrix using the singular value
decomposition (SVD) as expressed in equation 8.

DEBLURRING TECHNIQUES

GENERAL LINEAR MODEL

(Eq. 8)

Where U U = V V = I and = diag( ) is N N

diagonal matrix which contains a set of nonnegative values
(Singular values) that are organized in the main diagonal in a
nonincreasing order, i.e. 0 . The
number of positive singular values determines the rank of A.
The columns of U are called the left singular vectors. The
columns of V are called the right singular vectors. When i is
increased, the singular values decreased and the
corresponding singular vectors tend to have more sign
changes to represent higher-frequency information. If all
singular values are not zeros, we can directly deduce that the

In this model[7], it is assumed that the blurring process (the

process of converting sharp image into blurred image) is
linear. This assumption is very useful because in many cases
the blur can be well approximated by a linear model. It is also
assumed that the desired sharp image I and the recorded
blurred image B are two grayscale digital images of size
m n. The most simple case of blurring occurs when the
blurring of the rows is performed independently of the
blurring of the columns. In this case, there are two matrices
A and A of sizes m m and n n , respectively. The

ISSN 2320-2602
International Journal of Advances in Computer Science and Technology (IJACST), Vol.3 , No.12, Pages : 01-08
Special Issue of ICCEeT 2014 - Held on 22nd December 2014, Dubai
inverse of the blurring matrix A can be expressed as shown in
equation 9.

When the percentage of the first singular value to the last

singular value, i.e. , is very large, it means that the
solution is highly sensitive to perturbations and rounding
errors.

system is a list of blurred images and the target output is the

corresponding sharp images. After the system is trained on
various types of blurred images, it becomes able to predict an
unknown sharp image from the blurred one. Some of the
techniques that employed neural network, in order to
converge the recorded blurred image to the sharp version of
it, can be found in [32, 33, 34, 35, 37]. Jubien et al. [31]
described two different training algorithms that can be used
by the neural network for blind image restoration, one is
based on the least mean squares (LMS) rule, and the second is
called dubbed algorithm-X. Subashini et al. [29] adopted
back propagation network with gradient decent rule which
consists of three layers. They used highly nonlinear back
propagation neuron for image restoration to get a high quality
restored image and achieves fast neural computation, less
computational complexity due to the less number of neurons
used and quick convergence without lengthy training
algorithm. S. Saadi et al. [30] proposed improving the
training of the neural network using a novel swarm
optimization algorithm called Artificial Bees Colony (ABC),
inspired from the foraging intelligence of honey bees. Z.
Hongying et al. [36] proposed a variational image deblurring
algorithm based on modified Hopfield neural network
(MHNN).

RICHARDSON-LUCY DECONVOLUTION APPROACH

ITERATIVE METHODS

The Lucy-Richardson algorithm, or RichardsonLucy

deconvolution, is an iterative procedure for recovering the
sharp image that has been blurred by a known point spread
function (PSF). It was developed by William Richardson and
Leon Lucy. In this technique, the Pixels in the observed
(blurred) image is represented in terms of the point spread
function and the latent (sharp) image. Some of the researches
that used Lucy-Richardson algorithm can be found in [2, 9,
11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25]. Let
K is the PSF (the portion of light that comes from true
location j and is observed at another location i), x is the pixel
value at location j in the latent (sharp) image, and b is the
observed value at pixel location i. The pixels in the observed
image are represented in terms of the point spread function
and the latent (sharp) image as expressed in equation 14.

J. Biemond [75] discussed using iterative restoration

techniques for the removal of linear blurs from images that
are degraded by pointwise nonlinearities such as film
saturation and additive noise. They show that iterative
algorithms can use various types of prior knowledge about
the class of possible solutions to remove non-stationary blurs.
They presented study on the convergence problem of the
algorithms. They also compared between classical solutions
such as inverse filters, Wiener filters, and constrained
least-squares filters which are shown to be limiting solutions
of variations of the iterations. Finally, they introduced
Regularization as a means for preventing the excessive noise
magnification that is typically associated with ill-conditioned
inverse problems such as the deblurring problem, and they
showed that noise effects can be minimized by terminating
the algorithms after a finite number of iterations.
A. Beck and M. Teboulle [76] studied a class of Iterative
Shrinkage-Thresholding Algorithms (ISTA) for solving
linear inverse problems arising in image processing. They
reported that these methods is attractive due to its simplicity
but they are also known to converge very slowly. Then, they
presented a Fast Iterative Shrinkage-Thresholding Algorithm
(FISTA) which preserves the computational simplicity of
ISTA, but with a global rate of convergence which is proven
to be significantly better, both theoretically and practically.
James L. Herring [77] showed the effectiveness of three
projected iterative methods for image deconvolution:
projectedsuccessive over-relaxation
method (SOR),
projected Landweber method, and an interior point gradient
method. J. G. Nagy et al. [78] introduced a set of Matlabtools
that implement iterative image restoration methods. They
uses Matlabfor the implementation of efficient matrix vector
multiplication, and for solving the linear system for
preconditioners, efficiently. They also combine the powerful
scientific computing and graphics capabilities in Matlabto do
object-oriented programming and operator overloading,
results in a set of usable and extensible classes.

(Eq. 9)

Where is also a diagonal matrix whose elements in the

main diagonal are 1 , for i = 1,2, . N.
Equation 8 and 9 can be reformulated in terms of the left
singular vectors u , right singular vectors v and singular
values as shown in equation 10 and 11.
=

(Eq. 10)
1

(Eq. 11)

So, equation 7 can be rewritten as shown in equation 12.

=
=

(Eq. 12)

(Eq. 13)

(Eq. 14)

The model assumes that x values are Poisson distributed,

which is the most suitable for photon noise in the data. The
model calculates the most likely x given the observed b and
known PSF K . The model finds a recursive formula to
estimate x as expressed in equation 15.
(

=
=

( )

(Eq. 15)
( )

(Eq. 16)

Experimentalresults showed that the iteration converges to

the most likelihood solution of x .
NEURAL NETWORK APPROACH
In this approach, neural network are employed to convert
blurred images into sharp images. The basic idea is that the
system can be trained on various types of blurred images and
their corresponding sharp ones. The input to the training
3

ISSN 2320-2602
International Journal of Advances in Computer Science and Technology (IJACST), Vol.3 , No.12, Pages : 01-08
Special Issue of ICCEeT 2014 - Held on 22nd December 2014, Dubai
WIENER FILTERING

researches that discusses blind deconvolution approach can

be found in [2, 10, 28, 48, 49, 51]. There are two basic
approaches for blind deconvolution. The first is called
projection based blind deconvolution and the second is the
maximum likelihood restoration. The first approach makes
an initial estimation of the point spread function (PSF)
followed by an initial estimation of the true image. Then, the
process is repeated until a predefined convergence criterion is
met. The advantages of this approach is that it is robust
against inaccuracies of support size and it is insensitive to
noise. The disadvantages of this approach is that it is not
unique and can lead to errors if unsuitable local minima has
been initiated. The second approach uses the maximum
likelihood to estimate parameters such as PSF and covariance
matrix. Since the estimation of PSF is not unique, the
approach can consider parameters such as size, symmetry,
and other parameters. The advantages of this approach is low
computational complexity. The approach can obtain blur,
noise and power spectra of the sharp image. The
disadvantage of this approach is that it converges to local
minima of the estimated cost function.

One of the most common technique for image deblurring is

wiener filtering. The wiener filter has a large ability to
remove the blur in images caused by linear motion or
unfocussed optics. The blurred image can be seen as a result
of poor sampling. Each pixel in the image should contain
intensity value for a single stationary point in front of the
capturing device(or camera). Unfortunately, if the camera is
moved or the shutter speed is very slow, a given pixel will be
an amalgram of intensities from points along the line of the
camera's motion[38, 39]. The Wiener filter performs an
optimal trade-off between inverse filtering and noise
smoothing because it removes the additive noise and inverts
the blurring simultaneously. Some of the techniques that used
Wiener filtering can be found on [40, 41, 42, 43, 44, 45, 46,
47]. The model works as described in equations 17-20. Let
the case is the model described in equation 17.
b( , ) = ( , ) ( , ) + ( , )

(Eq. 17)

Where s(x, y) is the unknown sharp image, h(x, y) is the

known impulse response of a linear scale-invariant system,
n(x, y) is some unknown additive noise independent of
s(x, y) , and b (x, y) is the observed blurred image. The
Wiener deconvolution filter finds g(x, y) and use it to
estimate s(x, y), as expressed in equation 18.
( , ) = ( , ) ( , )

SPARSE REPRESENTATION
Most of the blind image deblurring techniques either only
remove simple motion blurring, or need user interactions to
handle more complex cases. Sparse decomposition is a
process that estimates a sparse multi-dimensional vector that
satisfies a set of equations in a linear system given
high-dimensional observed data and a design matrix. It is
used in a wide range of applications such as image deblurring
and restoration. Sparse representation succeeds due to the
development of l1-norm optimization techniques, and also
because natural images are intrinsically sparse in some
domain. Some of the techniques that use sparse
approximation for blind image deblurring can be found on
[55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69]. J.
Yang et al. [54] presented an approach for image
super-resolution using sparse signal representation. From
image statistics, they stated that image patches can be
well-represented as a sparse linear combination of elements
from an appropriately chosen over-complete dictionary. So
they seek a sparse representation for each patch of the
low-resolution input, and then use the coefcients of this
representation to generate the high-resolution output. W.
Dong et al. [52] introduced two adaptive regularization terms
into the sparse representation framework. The first term is a
set of autoregressive (AR) models that are learned from a
specific dataset. The best fitted AR model is adaptively
selected to regularize the image local structures. The second
term is the image non-local self-similarity. Experimental
results on image deblurring and super-resolution shows that
the usage of adaptive sparse domain selection and adaptive
regularization, achieves an improved results in terms of
PSNR and visual perception. J.-F. Cai et al. [53] presented an
approach to remove motion blurring from a single image by
formulating the blind blurring as a new joint optimization
problem, which simultaneously maximizes the sparsity of the
blur kernel and the sparsity of the clear image under certain
suitable redundant tight frame systems (curvelet system for
kernels and framelet system for images). The advantage of
this techniques is that it can recover high-quality images from
given blurred images without any prior knowledge of the blur
kernel.

(Eq. 18)

Where s(x, y) is an estimate of s(x, y) that minimizes the

mean square error. The filter works in the frequency domain.
The first step is to calculate the frequency domain version of
g(x, y), as expressed in equation 19.
( , )=

( , ) ( , )
| ( , )| ( , ) + ( , )

(Eq. 19)

Where G(u, v) is the Fourier transform of g, H(u, v) is the

Fourier transform of h, W(u, v) is the mean power spectral
density of s(x, y), N(u, v) is the mean power spectral density
of the noise n(x, y). Finally, the filtering is performed in the
frequency domain, as expressed in equation 20.
( , )= ( , ) ( , )

(Eq. 20)

Where S(u, v) is the Fourier transform of the estimated sharp

image and B(u, v) is the Fourier transform of the observed
blurred image. If three pixels in a line contain info from the
same point on an image, the digital image will seem to have
been convolved with a three-point boxcar in the time
domain[39]. It is seen that the technique is based on inverse
filtering. Unfortunately, there is a number of drawbacks
associated with the wiener filter: (i) H is unknown. It can be
guessed for a given image but it requires a lot of trials and
efforts to give a good estimation of H; (ii) it is failed in some
cases because the sine function equal 0 at some values of x
and y.
BLIND DECONVOLUTION APPROACH
It is noticeable that inverse filtering and wiener filtering
require accurate estimation of the degradation function and
also require information about the noise model for image
restoration. But it is very hard to obtain from a practical point
of view. So, blind deconvolution approach is not concerned
with an accurate estimation of the degradation function.
Instead, an initial estimation of the degradation function is
calculated. Then the system carries out an iterative process of
blind deconvolution until an accurate estimation is obtained
which minimizes the mean square error. Some of the
4

ISSN 2320-2602

PSNR

International Journal of Advances in Computer Science and Technology (IJACST), Vol.3 , No.12, Pages : 01-08
Special Issue of ICCEeT 2014 - Held on 22nd December 2014, Dubai

40
35
30
25
20
15
10
5
0
Y. Tai et S. Prasad
al[9]
et al[11]
Series1

30

30

O.
Whyte
et al[79]

A.-P.
Yang et
al[32]

30

20

Z.
J.
J. L.
Hongyin L. Yuan
P. Singh S. kumar W. Dong J.-F. Cai
Biemond Herring
g et
et al[73]
et al[38] et al[40] et al[52] et al[53]
et al[75] et al[77]
al[34]
19

30

29

22

20

26

35

33

Figure 1: Mean PSNR (db) for each of the tested techniques

SSIM

0.8
0.6
0.4
0.2
0
Y. Tai et
al[9]
Series1

0.93

S.
O.
Prasad Whyte
et al[11] et al[79]
0.95

0.9

A.-P.
Yang et
al[32]
0.76

Z.
J.
J. L.
Hongyin L. Yuan
P. Singh S. kumar W. Dong J.-F. Cai
Biemond Herring
et al[73]
et al[38] et al[40] et al[52] et al[53]
g et
et al[75] et al[77]
al[34]
0.77

0.93

0.88

0.69

0.69

0.72

0.97

0.98

Figure 1: Mean SSIM for each of the tested techniques

OTHER APPROACHES
As we discussed before, non-blind deconvolution methods
have been widely used for image deblurring. These
techniques can handle only the linear blur model where the
blurred image is generated by a linear convolution of the
latent image and the blur kernel. However, this often fails
because there are several types of outliers that should be
handled. S. Cho et al. [70] proposed a novel blur model that
explicitly takes into account several types of outliers, and
build a robust non-blind deconvolution method that reduces
the visual artifacts caused by outliers. They tested the
proposed system on synthetic and real-world examples. It
was shown that if these outliers are not handled, they cause
severe ringing artifacts even if the kernel is estimated
accurately. So they analyze a few common types of outliers
that cause other deblurring techniques to fail, such as pixel
saturation and non-Gaussian noise. H. Yin and I. Hussain[72]
proposed a blind deconvolution method for image deblurring
based on ICA measure as well as a simple genetic algorithm.
S. El-Regaily, et al. [71] proposed an algorithm for image
deblurring using Genetic Algorithms, by finding proper
parameters and goal function.
O. Whyte et al. [79] propose a new parametrized geometric
model of the blurring process in terms of the rotational
motion of the camera during exposure. The model can
capture non-uniform blur in an image due to camera shake
using a single global descriptor, and can be substituted into
existing deblurring algorithms with only small modifications.
L Yuan et al. [73] proposed an approach that uses the blurred
image and an assistance noisy image to estimate an accurate

blur kernel, which is harder to obtain from a single blurred

image. Then, they used a residual deconvolution to
signicantly reduce ringing artifacts inherent to image
deconvolution. Finally, they used a gain-controlled
deconvolution process to suppress the remaining ringing
artifacts in smooth image regions. A. Gupta et al. [74]
presented an image deblurring technique to estimate spatially
non-uniform blur. They use existing spatially invariant
deconvolution methods to compute initial estimates of the
latent image. They represented the camera motion as a
Motion Density Function (MDF) which records the fraction
of time spent in each discretised portion of the space of all
possible camera poses. They reported that 6D camera motion
is well approximated by 3 degrees of motion, then they
analyse the scope of this approximation.
EXPERIMENTAL RESULTS
To evaluate the effectiveness of different image deblurring
techniques, the quality of the latent image is measured. There
are two types [50] of measures that can be used to evaluate
the quality: subjective measures and objective measures.
Subjective measures measure the users satisfaction or
perception without calculating a numeric quantity through an
explicit formula. This type of metrics is the most accurate
because it is directly trying to accommodate different user
requirements. The second type of metrics is the objective
metrics which use an explicit formula to calculate a numeric
quantity. There are two types of objective metrics that are
widely used in the evaluation of image deblurring techniques:
Peak Signal-To-Noise Ratio (PSNR) and Structural
Similarity (SSIM). PSNR is used to measure the quality of

ISSN 2320-2602
International Journal of Advances in Computer Science and Technology (IJACST), Vol.3 , No.12, Pages : 01-08
Special Issue of ICCEeT 2014 - Held on 22nd December 2014, Dubai
image in dB (decibels). PSNR between two images
(deblurred image and original image) measures how far the
two images are equal.
1

= 10

[13] S. Remmele, and J. Hesser. "Vector extrapolation-based acceleration

of
regularized
Richardson
Lucy
image
deblurring",
InBildverarbeitungfr die Medizin, PP. 400-404. Springer Berlin
Heidelberg, 2009.
[14] Z. Al-Ameen, G. Sulong, M. Gapar, and M. Johar. "Utilizing a
LaplacianRichardson Lucy procedure for deblurring CT medical
images degraded by Gaussian blur." In proceeding of: 5th conference
and exhibition on computer communication. 2012.

(Eq. 21)
)

(Eq. 22)

[15] J. L. Wu, C.-F. Chang, and C.-S. Chen. "An improved

Richardson-Lucy algorithm for single image deblurring using local
extrema
filtering."
In Intelligent
Signal
Processing
and
Communications Systems (ISPACS), International Symposium on, pp.
27-32. IEEE, 2012.

Where I is the original image, X is the restored (deblurred)

image, M is the number of rows, N is the number of columns
and MAX is the maximum possible intensity value of the
image. The Structural Similarity (SSIM) [50] is calculated as
expressed in equation 23.
=

(2
(

+ )(2
+ )
+ )( +
+ )

[16] H. L. Yang, Y.-H. Chiao, P.-H Huang, and S.-H. Lai. "Blind image
deblurring with modified richardson-lucy deconvolution for ringing
artifact suppression." In Advances in Image and Video Technology,
pp. 240-251. Springer Berlin Heidelberg, 2012.
[17] M. K.Khan, S. Morigi, L. Reichel, and F. Sgallari. "Iterative methods
of Richardson-Lucy-type for image deblurring." Numerical
Mathematics: Theory, Methods and Applications 12,2003.

(Eq. 23)

Where and are the means of I and X, respectively;

and are the variance of I and X, respectively; c and c are
used to stabilize the division with weak denominator.
Each of the listed techniques in figure 1 and figure 2 is tested
on 20 different blurred images. We added a Gaussian noise to
each blurred image. The standard deviation of the Gaussian
noise is 0.005 to simulate the realistic conditions during
image acquisition. The obtained PSNR and SSIM are
recorded for each one of the tested technique.

[18] J. L. Wu, C.-F. Chang, and C.-S. Chen. "An adaptive richardson-lucy
algorithm for single image deblurring using local extrema
filtering." Journal of applied science and engineering 16, no. 3, 2013.
[19] Y. W. Tai, P. Tan, L. Gao, and M.S.
Richardson-Lucy deblurring for scenes under projective
motion path, Technical report, KAIST, 2009.

[20] J. J. Ding, W.-D. Chang, Y. Chen, S.-W Fu, C.-W Chang, and C.-C.
Chang. "Image deblurring using a pyramid-based Richardson-Lucy
algorithm." In Digital Signal Processing (DSP), 2014 19th
International Conference on, PP. 204-209. IEEE, 2014.
[21] Y. Chu, S. Li, and M. Wang. "A Modified Richardson-Lucy Method
for Space-Variant Motion Deblurring." In Computer Science &
Service System (CSSS), International Conference on, pp. 1884-1887.
IEEE, 2012.

REFERENCES
[1]

[2]

J.Jiaya. "Single image motion deblurring using transparency." In

Computer Vision and Pattern Recognition, 2007. CVPR'07.
IEEE Conference on, pp. 1-8. IEEE, 2007.

[22] B. Zhao, W. Zhang, and H. Ding. "Novel image deblurring algorithm

based on Richardson-Lucy", Computer Engineering and Applications,
Vol. 47, no. 34, 2011.

D. Singh,R. K. SahuA Survey on Various Image

DeblurringTechniques, International Journal of Advanced Research
in Computer and Communication Engineering, Vol. 2, Issue 12,
December 2013.

[3]

G. Anil, and R. Kumar. "Design and Analysis of an Algorithm for

Image Deblurring using Bilateral Filter",International Journal for
Science and Emerging Technologies with Latest Trends, Vol. 5, No. 1,
PP. 28-34, 2013.

[4]

C. Jian-Feng, H. Ji, C. Liu, and Z. Shen. "Framelet-based blind motion

deblurring from a single image." Image Processing, IEEE Transactions
on 21, no. 2 (2012): 562-572.

[5]

A. K. Soe, and X. Zhang. "A simple PSF parameters estimation method

for the de-blurring of linear motion blurred images using wiener filter
in OpenCV." In Systems and Informatics (ICSAI), 2012 International
Conference on, pp. 1855-1860. IEEE, 2012.

Brown,

[23] H.-L. Yang, P.-H. Huang, and S.-H. Lai. "A novel gradient attenuation
RichardsonLucy algorithm for image motion deblurring." Signal
Processing 103, PP. 399-414, 2014.
[24]

L.P. Torres, "Deblurring AffymetrixGeneChip

Images by Applying the Lucy-Richardson
Algorithm." PhD diss., University of Essex, 2008.

[25] W. D. Chang, J. J. Ding, Y. Chen, C. W. Chang, and C. C. Chang,

Image deblurring with pyramid-based Richardson-Lucy algorithm,
Computer Vision, Graphics, and Image Processing, Sun Moon Lake,
Taiwan, Aug. 2012.
[26] S. S. Al-Amri, and A. S. Ali. "Restoration and Deblured Motion
Blurred Images." International Journal of Computer Science Issues
(IJCSI) 11, no. 1, 2014.

[6]

C. Khare, and K. K. Nagwanshi. "Implementation and analysis of

image restoration techniques." International Journal of Computer
Trends and Technology-May to June, 2011.

[27] D. S. Rao, K. S. Deepthi, and K.M.S. Deep. "Application of Blind

Deconvolution Algorithm for Image Restoration." International
Journal of Engineering Science and Technology (IJEST), 2011.

[7]

P. C. Hansen, , J. G. Nagy, and D. P. O'leary. Deblurring images:

matrices, spectra, and filtering. SIAM, Philadelphia, 2006.

[8]

A. S. Saleh, N. V. Kalyankar and S.D. Khamitar, "Deblured Gaussian

blurred images", Journal of Computing, Vol. 2, No. 4, 2010.

[28] S. H. Umale, and A. M. Sahu. "A Review on Various Techniques for

Image Debluring", International Journal of Computer Science and
Mobile Computing, Vol.3 Issue.4, PP. 263-268, April- 2014.

[9]

Y. Tai, P. Tan, and M. S. Brown, Richardson-Lucy deblurring for

scenes under projective motion path, IEEE Trans. Pattern Anal. Mach.
Intell., vol. 33, no. 8, pp. 16031618, Aug. 2011.

[29] P. Subashini, M. Krishnaveni, and V. Singh. "Image Deblurring Using

Back Propagation Neural Network", World of Computer Science and
Information Technology Journal (WCSIT), ISSN: 2221-0741, Vol. 1,
No. 6, 277-282, 2011

[10] D. Kundur and D. Hatzinakos, A novel blind deconvolution scheme

for image restoration using recursive filtering, IEEE Trans. Signal
Process., Vol. 46, No. 2, PP. 375-390, Feb. 1998.

[30] S. Saadi, A. Guessoum, and M. Bettayeb. "ABC optimized neural

network model for image deblurring with its FPGA implementation",
Microprocessors and Microsystems 37, no. 1 PP. 52-64, 2013.

[11] S. Prasad, "Statistical-information-based performance criteria for

Richardson-Lucy image deblurring." Journals of Optical Society of
America, 19, no. 7, PP. 1286-1296, 2002.

[31] C. M. Jubien, and M. E. Jernigan. "A neural network for deblurring an

image." In Communications, Computers and Signal Processing, IEEE
Pacific Rim Conference on, PP. 457-460. IEEE, 1991.

[12] M. K. Singh, U. S.Tiwary, and Y.-H. Kim. "An adaptively accelerated

Lucy-Richardson method for image deblurring", EURASIP Journal on
Advances in Signal Processing52, 2008.

[32] A.-P. Yang, Z.-X. Hou, and C.-Y. Wang. "Image deblurring based on
wavelet and neural network." In Wavelet Analysis and Pattern
Recognition, 2007. ICWAPR'07. International Conference on, vol. 2,
pp. 647-651. IEEE, 2007.

ISSN 2320-2602
International Journal of Advances in Computer Science and Technology (IJACST), Vol.3 , No.12, Pages : 01-08
Special Issue of ICCEeT 2014 - Held on 22nd December 2014, Dubai
[33] N. Kumar, M. Shrivastava, and N. Rahim. "Image Deblurring Using a
Neural Network Approach", International Journal of Engineering and
Innovative Technology (IJEIT), Vol. 2, No. 3,September 2012

[53] J.-F. Cai, H.Ji, C. Liu, and Z.Shen. "Blind motion deblurring from a
single image using sparse approximation." In Computer Vision and
Pattern Recognition, 2009. CVPR 2009. IEEE Conference on, pp.
104-111. IEEE, 2009.

[34] Z. Hongying, P. Qicong, and W. Yadong. "Variational Image

Deblurring
Using
Modified Hopfield Neural
Network."
In Communications, Circuits and Systems Proceedings, 2006
International Conference on, Vol. 1, PP. 429-432. IEEE, 2006.

[54] J. Yang, J. Wright, T. S. Huang, and Y. Ma. "Image super-resolution

via sparse representation." Image Processing, IEEE Transactions on19,
no. 11, PP. 2861-2873, 2010.

[35] N. Kumar, R. Nallamothu, and A. Sethi. "Neural Network Based Image

Deblurring", 11th symposium on neural network applications in
electrical engineering, NEUREL, Serbia, September 2012.

[55] L. Xu, S. Zheng, and J. Jia. "Unnatural l0 sparse representation for

natural image deblurring." In Computer Vision and Pattern
Recognition (CVPR), 2013 IEEE Conference on, pp. 1107-1114.
IEEE, 2013.

[36] Z. Hongying, P. Qicong ; W. Yadong, "variational image deblurring

using modified hopfield neural network", Communications, Circuits
and Systems Proceedings, 2006 International Conference on , vol.1,
no., pp.429,432, 25-28 June 2006

[56] Y. Zhang, and Z. Yanning "Sparse representation based iterative

incremental image deblurring." In Image Processing (ICIP), 2009 16th
IEEE International Conference on, pp. 1293-1296. IEEE, 2009.

[37] A.-P. Hou, Y.Z.-Xin, and C.-Y. Wang. "image deblurring based on
wavelet and neural network", Wavelet Analysis and Pattern
Recognition, 2007. ICWAPR'07. International Conference on, 2007.

[57] M. Carlavan, and L. B.-Fraud. "Sparse Poisson noisy image

deblurring." Image Processing, IEEE Transactions on 21, no. 4 (2012):
1834-1846.

[38] P. Singh, and L. Jain. "Noise reduction in ultrasound images using

wavelet and spatial filtering techniques." In Information Management
in the Knowledge Economy (IMKE), 2013 2nd International
Conference on, pp. 57-63. IEEE, 2013.

[58] H. Zhang, D. Wipf, and Y. Zhang. "Multi-image blind deblurring using

a coupled adaptive sparse prior." In Computer Vision and Pattern
Recognition (CVPR), 2013 IEEE Conference on, pp. 1051-1058.
IEEE, 2013.

[39] A. Dwivedi, and A. Singh. "An Advanced Filter for Image

Enhancement and Restoration." Journal Open Journal of Advanced
Engineering Techniques OJAET Vol 1, no. 1 PP. 7-10, 2013.

[59] L. Ma, J. Yu, and T.Zeng. "Sparse Representation Prior and Total
Variation--Based Image Deblurring under Impulse Noise." SIAM
Journal on Imaging Sciences 6, no. 4, PP. 2258-2284, 2013.

[40] S. kumar, P. kumar,M. Gupta, A. K. Nagawat, Performance

Comparison of Median and Wiener Filter in Image De-noising,
International Journal of Computer Applications, Vol. 12, Issue
No.-10.5120, PP. 1664-2241, Dec. 2010.

[60] H. Huang, and N. Xiao. "Image deblurring based on sparse model with
dictionary learning [J]." Journal of Information and Computational
Science10, no. 1, PP. 129-137, 2013.
[61] W. Dong, G. Shi, and X. Li. "Image deblurring with low-rank
approximation structured sparse representation." In Signal &
Information Processing Association Annual Summit and Conference
(APSIPA ASC), Asia-Pacific. 2012.

[41] J. Wang, G. Wang, and M. Jiang. "Blind deblurring of spiral CT images

Based on ENR and Wiener filter." Journal of X-Ray Science and
Technology13, no. 1, PP. 49-60, 2005.
[42] P. Bojarczak, and Z. Lukasik. "Image DeblurringWiener Filter Versus
TSVD Approach." Advances in Electrical and Electronic
Engineering 6, no. 2,PP. 86-89, 2011.

[62] H. Li, Y. Zhang, H. Zhang, Y. Zhu, and J. Sun. "Blind image

deblurring based on sparse prior of dictionary pair." In Pattern
Recognition (ICPR), 2012 21st International Conference on, pp.
3054-3057. IEEE, 2012.

[43] A. Kalestynski, "Deblurring Of Images By The Use Of A Computer

Generated Quasi-Wiener Spatial Filter." In Holography'89, pp.
551-556. International Society for Optics and Photonics, 1990.

[63] Q. Liu, D. Liang, Y. Song, J. Luo, Y. Zhu, and W. Li. "Augmented

Lagrangian-based sparse representation method with dictionary
updating for image deblurring." SIAM Journal on Imaging Sciences 6,
no. 3, PP. 1689-1718, 2013.

[44] Y. Zheng, "A Fast Image Deblurring Algorithm Using the Wiener
Filter and the Hartley Transform." In Review of Progress in
Quantitative Nondestructive Evaluation, pp. 735-742. Springer US,
1989.

[64] M. Yin, J. Gao, D. Tien, and S. Cai. "Blind image deblurring via
coupled sparse representation." Journal of Visual Communication and
Image Representation 25, no. 5, PP. 814-821, 2014.

[45] S. Khalil, W. Nourildean, M. Ismaeel. "Image Deblurring and Effect of

Wiener, Regularized, Lucky-Richardson, Blind-Deconvolution,
Average and Median Filters on Blurred Image and Noisy
Image." Tikrit Journal of Engineering Science (TJES) 21, no. 1,PP.
84-95, 2014.

[65] A. M. Deshpande, and S. Patnaik. "Uniform and non-uniform single

image deblurring based on sparse representation and adaptive
dictionary learning." The International Journal of Multimedia & Its
Applications (IJMA), 06 (01), PP. 47-60, 2014.

[46] H. Yang, Z. B. Zhang, D. Y. Wu, and H. Y. Huang. "Image deblurring

using empirical Wiener filter in the curvelet domain and joint non-local
means filter in the spatial domain." The Imaging Science Journal 62,
no. 3, PP. 178-185, 2014.

[66] X. Song, Y. Li, and D. Hu. "Image Deblurring Without Boundary

Condition from Accurate Degradation Model via Sparse
Representation", International Journal of Applied Mathematics and
Statistics 49, no. 19, PP. 220-228, 2013.

[47] P. D.Sankhe, M. Patil, and M. Margaret. "Deblurring of grayscale

images using inverse and Wiener filter." In Proceedings of the
International Conference & Workshop on Emerging Trends in
Technology, pp. 145-148. ACM, 2011.

[67] P. C. Hansen, J. G. Nagy, and K. Tigkos. "Rotational image deblurring

with sparse matrices." BIT Numerical Mathematics, PP. 1-23, 2013.
[68] H. Pan, Z. Jing, M. Lei, R. Liu, B. Jin, and C. Zhang. "A sparse
proximal Newton splitting method for constrained image
deblurring."Neurocomputing 122, PP. 245-257, 2013.

[48] N. B.Patel, K. N. Jariwala"A Survey on Image Enhancement using

Image Super-Resolution and Deblurring Methods", Journal of
Emerging Technologies and Innovative Research (JETIR), Vol. 1, No.
5, 2014

[69] P.Jegatheeswari, J. Amudha, and R. Sudhakar. "Sparse representation

based image deblurring and super resolution." In Advances in
Engineering, Science and Management (ICAESM), 2012 International
Conference on, pp. 468-473. IEEE, 2012.

[49] M. Poulose, "Literature Survey on Image Deblurring

Techniques."International Journal of Computer Applications
Technology and Research, Volume 2 Issue 3, 286 - 288, 2013.

[70] S. Cho, J. Wang, and S. Lee. "Handling outliers in non-blind image

deconvolution." In Computer Vision (ICCV), IEEE International
Conference on, pp. 495-502. IEEE, 2011.

[50] R. Wang, and D. Tao. "Recent Progress in Image Deblurring." arXiv

preprint arXiv:1409.6838, 2014.
[51] A. Levin, Y. Weiss, F. Durand, and W. Freeman, Understanding blind
deconvolution algorithms, IEEE Trans. Pattern Anal. Mach. Intell.,
vol. 33, no. 12, pp. 23542367, Nov. 2011.

[71] S. El-Regaily, M. A. El-Aziz, H. El-Messiry, and M. Roushdy. "Using

GPU-accelerated Genetic Algorithm for non-linear motion deblurring
in a single image." In Informatics and Systems (INFOS), 2012 8th
International Conference on, pp. BIO-174. IEEE, 2012.

[52] W. Dong, L.Zhang, R. Lukac, G. Shi, Image Deblurring and

super-resolution by adaptive sparse domain selection and adaptive
regularization, IEEE Trans. On Image Processing, vol. 20, no. 7, pp.
1838-1857, July 2011.

[72] H. Yin, and I. Hussain. "ICA and genetic algorithms for blind signal
and image deconvolution and deblurring." In Intelligent Data
Engineering and Automated LearningIDEAL 2006, pp. 595-603.
Springer Berlin Heidelberg, 2006.

ISSN 2320-2602
International Journal of Advances in Computer Science and Technology (IJACST), Vol.3 , No.12, Pages : 01-08
Special Issue of ICCEeT 2014 - Held on 22nd December 2014, Dubai
[73] L. Yuan, J. Sun, L. Quan, and H. Shum, Image deblurring with
blurred/noisy image pairs, ACM Trans. Graph., vol. 26, no. 3, pp.
111,2007.
[74] A. Gupta, N. Joshi, C. Lawrence Zitnick, M. Cohen, and B. Curless,
Single image deblurring using motion density functions, in Proc
ECCV, PP. 171184, 2010.
[75] J. Biemond, R. L. Lagendijk, and R. M. Mersereau. "Iterative methods
for image deblurring." Proceedings of the IEEE 78, no. 5, PP. 856-883,
1990.
[76] A. Beck, and Marc Teboulle. "A fast iterative shrinkage-thresholding
algorithm with application to wavelet-based image deblurring."
In Acoustics, Speech and Signal Processing, 2009. ICASSP 2009.
IEEE International Conference on, pp. 693-696. IEEE, 2009.
[77] J. L. Herring,"The Effects of Projection on Iterative Methods in Image
Deblurring." PhD diss., Emory University, 2010.
[78] J. G. Nagy, K. Palmer, and L.Perrone. "Iterative methods for image
deblurring: a Matlab object-oriented approach." Numerical
Algorithms 36, no. 1, PP. 73-93, 2004.
[79] O. Whyte, J.Sivic, A.Zisserman, and Jean Ponce. "Non-uniform
deblurring for shaken images." International journal of computer
vision 98, no. 2, PP. 168-186, 2012.