International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169

Volume: 6 Issue: 6 184 - 188

______________________________________________________________________________________
Image Restoration Techniques Using Fusion to Remove Motion Blur

Neha Jat Jyoti Jain

Sagar Institute of Research and Technology, Sagar Institute of Research and Technology,
Bhopal, Madhya Pradesh, India Bhopal, Madhya Pradesh, India
Email : [email protected] Email : [email protected]

Abstract—Restoration techniques are oriented towards modelling the degradation and applying inverse process to recover the original image.
The image gets blurred due to relative motion between object and detector (Motion Blur), and/or improperly focused image capturing device.
This work presents a comparison of different image restoration process where, different filtering method are used with RL-Deconvolutionfor
different applications. The proposed approach combines two different restoration method by using DWT.

Keywords: Lucy Richardson method, wiener filter, image fusion,DWT (Discrete wavelet Transform)

__________________________________________________*****_________________________________________________
I. INTRODUCTION II. IMAGE RESTORATION
Image restoration is the process of recovering an A. Wiener filter[5]
image that has been degraded by using a priori knowledge of Wiener filter is used to yield an estimated image of desired
the degradation phenomenon. Restoration techniques or targeted arbitrary process by using linear Time-
involves modelling of the degradation function and applying Invariant(LTI) processing of recognized noise, basically
the inverse process to recover the original image. [1] wiener filter minimizes the mean square error of estimated
pixel value and the desired pixel value.
Digital image Restoration techniques falls into
The main objective of wiener filter is to figure out
digital image processing. In digital Image restoration
statistical estimate of an anonymous signal employing an
degraded, noisy or/and Blurred images consider as an Input
associated signal as an input and filtering the acknowledge
and put effort to estimate the original and de-noised image.
signal to yield estimated output
image may have got corrupted by any unknown phenomenon
or external interference like, when the image of source is Weiner Filtering is also a non-blind technique for
captured/ taken on a detector or couple charged device reconstructing the degraded image in the occurrence of
(CCD) and there will be relative motion between them,then known PSF. It removes the additive noise and
during the exposure the point sources of real image got shift/ inverts the blurring at the same time. It not only performs
drift in captured image, which depends upon the relative the deconvolution by inverse filtering (high pass filtering) but
motion between them (This phenomenon is known as motion in addition removes the noise by means of a firmness
Blur). operation (low pass filtering). It evaluates through an
Motion blur is an undesired effect on an image due to estimation of the desired noiseless image. The input to a
camera shake during exposure, a long aperture time, wiener filter is a degraded image corrupted through additive
atmospheric turbulence, wrong focusing of lens, relative noise. The output image is computed by means of a filter
motion between Photographic device and original scene. In through means of the subsequent expression
image processing system, it is Important to regain the
original image from degraded image which is blurred by a 𝑓” = 𝑔 ∗ (𝑓 + 𝑛)
degradation Function[2][3][4].The degradation producing ill-
effect of blur is termed as the point spread function (PSF). 𝑓 is the new image, 𝑛 is the noise, 𝑓" is the estimated image
Any type of blur is characterized by the PSF. The and 𝑔 is the wiener filters response.
electromagnetic radiation or other imaging wave’s
propagated from a point source or point object is known as B. Lucy-Richardson Deconvolution Method[6]
the PSF. The quality of any imaging system depends on the When an image is captured on a detector such as
degree of spreading (blurring) of the point object. The photographic film or a charge coupled device (CCD), it will
PSFdefines the impulse response of a point source[3]. get slightly blurred, with an ideal point source not appearing as
a point but being spread out, that is known as the point spread
function. Non-point sources are generally the sum of many
individual point sources, and pixels in a recorded image can be

184
IJRITCC | June 2018, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 6 Issue: 6 184 - 188
______________________________________________________________________________________
represented in terms of the point spread function and the latent i.e. the image is divided into very small blocks called pixels.
image as An image can be defined as a two dimensional function I [2]
𝐼 = 𝑓(𝑥, 𝑦)
𝑑𝑖 = 𝑝𝑖𝑗 𝑢𝑗
𝑗

Where, f(x,y) g(x,y) f(x,y)

Degradation Restoration
𝑝𝑖𝑗 is PSF (Point Spread Function), Function + Filter
𝑗 and 𝑖 are; true image and recorded image,
𝑢𝑗 is value at location 𝑗,
𝑑𝑖 is observed value at location 𝑖, ƞ(x,y)

The calculations are performed using Poisson

Distribution
Where 𝑥 and 𝑦 are spatial coordinates. 𝑥, 𝑦 represents a
(𝑡−1) ⅆ𝑖 pixel. 𝐼 is the intensity or grey level value which is the
𝑢𝑗 = 𝑢𝑗 𝑡 𝑝
𝑐𝑖 𝑖𝑗 amplitude of 𝑓 at any point 𝑥, 𝑦 . If the values of the
𝑖 coordinates (spatial coordinates) and the amplitude are finite
Where, and discrete, then it is called digital image .the degraded
images 𝑔 𝑥, 𝑦 can Be represented as
𝑐𝑖 = 𝑝𝑖𝑗 𝑢𝑗 𝑡
𝑗 𝑔 𝑥, 𝑦 = 𝑕 𝑥, 𝑦 ∗ 𝑓 𝑥, 𝑦 + ƞ 𝑥, 𝑦
C. Regularized Filter[7]
Regularized filtering is used effectively when observed image are represented using the PSF an image
constraints like smoothness are applied on the recovered image undeniably is to avoid the need for restoring the image.
and limited information is known about the additive noise. The Furthermore, proposed work presents comparative analysis
blurred and noisy image is restored by a constrained least of different non-blind image restoration techniques based on
square restoration algorithm that uses a regularized filter. Wiener and Lucy-Richardson algorithm for different types of
Regularized restoration provides similar results as the wiener image layouts. [8]
filtering but it has a very different viewpoint. In regularized
 Lucy Richardson fused with Wiener restoration
filtering less prior information is required to apply restoration.
The regularization filter is often chosento be a discrete
Laplacian. ThisFilter can be understood as an approximation III. Lucy Richardson fused with regularized filter
of a Wiener filter.
𝑦 𝑚, 𝑛 = 𝑕 𝑚, 𝑛 ∗ 𝑥 𝑚, 𝑛 + 𝑢(𝑚, 𝑛) IV. LITERATURE SURVEY
Where,
Li Yang et.al in 2016[3] proposed a Method to the relative
∗ is 2D-Convolution, motion between the camera and the object scene produce
defect in recorded image which is termed as Motion Blur. The
𝑕 𝑚, 𝑛 is Point Spread Function (PSF),
degraded image, which is degraded by the relative motion and,
𝑥 𝑚, 𝑛 is original image, the image Must have Motion Blur with blur direction are
crucial for image restoration. After image recognition and
𝑢 𝑚, 𝑛 𝑖𝑠 𝑎𝑑𝑑𝑖𝑡𝑖𝑣𝑒 𝑛𝑜𝑖𝑠𝑒.
image analysis, the author developed an approach which is
Regularization deconvolution concentrates mainly on two based on the fractional directional derivative operator with a
desirable goals. non-casual solution to restore an image from Motion blurred
image a partial Derivative operator is summarized on
1. The closeness of model fit.
fractional derivative
2. The closeness of the model behavior
Real and synthesized motion Blurred images are
The model which will closely fulfills the model requirement
evaluated quantitatively and qualitatively performance
are,
evaluated. Noise is the crucial aspect of authored method. The
[𝑦(𝑚, 𝑛) − 𝑕 𝑚, 𝑛 ∗ 𝑥(𝑚, 𝑛)]2 experiments are established on a non-casual fractional partial
𝑚 ,𝑛
derivative mask, which ensures That the method will provide
better immunity to noise in comparison and accuracy of
D. Model of degradation analyzing motion Blur than the method based on integer order
A Model of image degradation and restoration process differentiation.
the basic unit of an image is called a pixel or image element
185
IJRITCC | June 2018, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 6 Issue: 6 184 - 188
______________________________________________________________________________________
Haiying Liu et.al in 2016[4] proposed a Total Variation Experimental Conventional Authored
Regulization Algorithm.Image taken / captured from Image Size
Images method Method
unfavorable conditions are more tend to got blurred by Article 482 X 482 17.0 7.5
unknown kernels and may corrupted by measurement noise. Parrots 753 X 502 25.4 8.5
The authored paper such as a new total variation (TV) Street 1024 X 683 52.8 10.4
minimization based method for blindly deburredthose images. Big Ben 1920 X 1080 141.9 16.8
The proposed method is dissimilar to the alternative
optimization based algorithm, by adopting a joint estimation Table 2: Processing Time [sec]
strategy that removes the unknownblurringkernel and
unknown image by iterative manner. In each iteration two K. Panfilova et.al [8]proposedmethod to solved the problem
separate image de-noisingsub problems / sub process are of linear blur (Linear Blur Compensation). To restored image,
performed that provides fast implementation. In authored authors usedLucy-Richardson iterative deconvolution and its
paper experiment perform on gray-scale test images as well as sub variants. The main disadvantage of conventional Lucy-
colored images. The proposed new method deburredimage Richardson iterative deconvolution are edge artifacts that
with considerably fast than conventional algorithm technique. forms during deconvolution process and less information of
optimal numbers of algorithm iteration. To lessen the issue of
[13] + [9] Proposed Method those problems authors proposed to extend the test subjected
SSIM 0.7595→0.9125 0.7595→0.9739 image beyond to its original border reducing its brightness to
zero at newly created borders and used an empirical
PSNR (dB) 21.4131→25.6472 21.4131→31.4932
benchmark to define the point where the iterative Lucy-
Identified kernel L2 Richardson procedure should end. The limit of iterations is
0.0090 0.0033
error
based on some analysis of the changes in the image at each
Eclipsed Time (sec) 77.9242 48.9063 iteration. The proposed changes (method) resulted in decrease
Table 1 : Performance comparison for grey image of image distortion by more than 50% (in RMS). This study
makes possible to easily estimate the required numbers of
Tomio goto et.al[1]is proposed a blind method that rapidly iterations for better performance of algorithms.
restores blurred images implying local patches. Author’s
proposed method a little section of test image (blurred image) Jian-Jiun Ding et.al in 2014[2]proposed a Pyramid-Based
is used for PSF (Point spread function) calculation Richardson-Lucy Algorithm.In image deburring, it is
(estimation). Author’s experiment results presents that the important to reconstruct images with small error, high
proposed method generates accurate de-blurred images, and perception quality, and less computational time. In this paper,
processing time is significantly lower than that of the a blurred image reconstruction algorithm, which is a
conventional de-blurring method. combination of the Richardson-Lucy (RL) deconvolution
approach and a pyramid structure, is proposed. The RL
approach has good performance in image reconstruction.
However, it requires an iterative process, which costs a lot of
computation time, and the reconstructed image may suffer
from a ringing effect. In the proposed algorithm, we
decompose a blurred image from a coarse scale to a fine scale
and progressively utilize the RL approach with different
number of iterations for each scale. Since the number of
iterations is smaller for the large scale part, the computation
time can be reduced and the ringing effect caused from details
can be avoided. Simulation results show that our proposed
algorithm requires less computation time and has good
Figure 1 : Block Diagram of Restoration through Local Patched
performance in blurred image reconstruction.

186
IJRITCC | June 2018, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 6 Issue: 6 184 - 188
______________________________________________________________________________________
Year Proposed Title and Author Methodology Conclusion
Result based on the Wavelet
O. Rockinger, " Image sequence fusion image Fusion with Wiener
Wiener filter and Blind
Oct. 1997 using a shift-invariant wavelet transform”, filter provided the better
Deconvolution
IEEE, 26-29 Oct. 1997[5] results that iterative Blind
Deconvolution
Winer filter with Image
I.W. Selesnick, “The design of Wiener Filter, Blind
Fusion with better acceptable
May 2002 approximate Hilbert transform pairs of Deconvolution algorithm and
restoration results Image
wavelet bases”, IEEE, 2016[9] Wavelet Transform
Restoration
K.Jaya Priya, Dr. R.S. Rajesh, “Dual Tree Better Restoration based on
Complex Wavelet Transform based Face Wiener Filter, Lucy-Richardson neural network with
July 2016 Recognition with Single View”, deconvolution algorithm & PSNR=30.1135 as compare to
Ubiquitous Computing communication Neural Network approach Lucy-Richardson, Inverse
Journal, July 2016[10] filter and Wiener filter
Wiener filter, Regularized
filter is the best techniques
Paul Hill, Nishan Canagarajah and Dave Wiener Filter, Lucy-Richardson when no noise in image. If
Feb. 2015 Bull, “Image Fusion Using Complex algorithm, Regularized filter & noise is presented with blur
Wavelets”, BMVS, pp. 487-496, 2015[6] Blind deconvolution algorithm the Lucy-Richardson
Algorithm is the best
techniques
It integrates complementary
R.J. Sapkal, S.M. Kulkarni, “Image
information to give a better
Oct. 2012 Fusion based on Wavelet Transform for Wavelet Transform
visual picture of a scenario
Medical Application”[11]
suitable for processing

Input
V. PROPOSED CONCEPT Image

Restoration techniques are located against base of degradation

Initial PSF
and applying the inverse process in order to recover the
original image. There will be present several kind of
degradation that need to be consider to restore degraded Blur the
image. Such kind of degrading are blurring, noise and color Image

imperfections. This study focus on approach to take measure

PSF (Point Spread Function) parameters from blurred and
Deco nvolution Deco nvolution
noisy images and presents relative analysis of different using Lucy using Different
restoration techniques. Methodology used will maintain the Richardson Filter
requiredinformation from both images

Image Fusion
using Wavelet

Inverse DWT
Analysis

MSE and PSNR

Calculation

Output image

Figure 2: Process flow

187
IJRITCC | June 2018, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 6 Issue: 6 184 - 188
______________________________________________________________________________________
Figure 2shows sequence of procedure to be taken. First a [8]. K. Panfilova and S. Umnyashkin, "Linear blur
test subjected image is distorted and degraded in such ways compensation in digital images using Lucy-Richardson
that, I’ll look similar to actual distorted image by assuming method," IEEE, 2016.
initial PSF and, then applying blur to it using assumed PSF as [9]. I. W. SELESNICK, "The design of approximate Hilbert
criteria. Then the blur simulated image will get passed into transform pairs of wavelet bases," IEEE Transactions on
two different image restoration method, in our case RL- Signal Processing, vol. 50, pp. 1144-1152, 2002.
deconvolution and different filter (Wiener Filter, Blind [10]. P. Jaya and R. RENGAN SIVAGURUNATHAN, "Dual
Deconvolution and Regularized Filter). After the Tree Complex Wavelet Transform based Face Recognition
deconvolution process, resulted images then fused together by with Single View," Ubiquitous Computing communication
using Wavelet Fusion technique. For final end result, Inverse Journal, 2016.
DWT Analysis carried out to calculate SNR and MSE. [11]. R. Sapkal and S. Kulkarni, "Image Fusion based on
Wavelet Transform for Medical," International Journal of
Engineering Research and Applications, vol. 2, no. 5, pp.
VI. CONCLUSION
624-627, 2012.
Image restoration is an active research area and various
researchers work to get better the efficiency by applying
different method. Primarily image restoration is done by using
numerus restoration techniquesuch as Weiner filter,
Richardson-Lucy Blind Deconvolution algorithm, Inverse
filter. The proposed procedure focus to consider PSF as
constrain since image are distorted and degraded. Moreover,
it will improve the objectivity of the image and removes the
noise and blurry content in the image. This work is present a
relative study of different non-blind image restoration
procedure will support on Wiener filter,Regularized Filterand
Lucy-Richardson method. The performances of these
techniques will evaluated and compared.

REFERENCES

[1]. T. Goto, H. Senshiki, S. Hirano and M. Sakurai, "Fast blind

restoration of blurred images based on local patches," 2016
International Conference on Electronics, Information, and
Communications (ICEIC), 2016.
[2]. D. Jian-Jiun, C. Wei-De, C. Yu, F. Szu-Wei, C. Chir-Weei
and C. Chuan-Chung, "Image deblurring using a pyramid-
based Richardson-Lucy algorithm," 19th International
Conference on Digital Signal Processing, 2014.
[3]. Y. Li , "Image Restoration from a Single Blurred
Photograph," 3rd International Conference on Information
Science and Control Engineering (ICISCE), 2016.
[4]. H. Liu, J. Gu, M. Q. -H. Meng and W.-S. Lu, "Fast
Weighted Total Variation Regularization Algorithm for
Blur Identification and Image Restoration," IEEE Access,
2016.
[5]. O. Rockinger, "Image sequence fusion using a shift-
invariant wavelet transform," Proceedings of International
Conference on Image Processing, Santa Barbara, vol. 3, pp.
288-291, 1997.
[6]. P. hill, N. Canagarajah and d. Bull, "Image Fusion Using
Complex Wavelets," BMVS, pp. 487-496, 2015.
[7]. S. Eddins, "Image deblurring using regularization," 21 july
2008. [Online]. Available:
http://blogs.mathworks.com/steve/2008/07/21/image-
deblurring-using-regularization/.

188
IJRITCC | June 2018, Available @ http://www.ijritcc.org
_______________________________________________________________________________________