Content uploaded by Alaa Hamid Mohammed
Author content
All content in this area was uploaded by Alaa Hamid Mohammed on Nov 22, 2020
Content may be subject to copyright.
IMAGE SUPER-RESOLUTION USING SPARSE
REPRESENTATION AND NOVELTY NOISE REMOVAL
SUPER-RESOLUTION
Abstract— The reconstruction of a composite image with a
super-resolution will produce a high-resolution image from a
low-resolution picture. Since the super-resolution problem is
not well answered, a narrow method of optimizing numeric
algorithm is implemented to remove image noise. The total
variation of the image is minimized subject to noise statistical
constraints. The restrictions are imposed on the use of
Lagrange multipliers. The solution is derived from the gradient
projection method. It is the solution of a partial differential
equation based on time based on several constraints described.
As t—-~0o, the solution converges into a continuous image. It is
a fast, relatively easy numerical algorithm. The measurements
are typically state of the art with very noisy videos. The process
is non-invasive, which gives straight boundaries to the picture.
A first step to the transfer can be translated. level range of the
normal image at a pace equivalent to the level curvature
separate from the magnitude of the feature gradient, and a
second stage for the return to the boundary point of the
picture.
Keywords— Super-Resolution, REAL, Novelty Noise
Removal, Super-Resolution.
I.INTRODUCTION
The noise possibly occurs in photos. The method for
production of pictures, picture editing, picture transfer, etc.
may be introduced. Such random distortions render it
difficult to process any picture needed. First of all, the
improvement-oriented feature in refs is carried out. Is very
effective in restoring blurred pictures, but most of the
oscillational noise may be "frozen"[1]. Only a small noise
level becomes dangerous if high precision, e.g. in the
processing of subcells (subpixels), is necessary. The most
widely employed techniques in use are focused on the
smallest square criterion to approximate a true noise signal.
The rationale derives from the statistical argument that the
least quadratic formula is the most efficient for an ensemble
of possible photographs[2]. This is based on the L 2
standard. However, the correct value for photographs was
conjectured in ref. that the usual total variance (TV) is not
the limit of L 2. Basically, L1 derivative standards are the
TV requirements, so that L1 estimate procedures are more
suitable for image estimates (restoration). If reliable
estimates of discontinuities in approaches are required
Typically, techniques for calculating L~ go back. The area of
small total variance functions plays an significant part. The L
1 equation is non-linear and computationally complicated as
contrasted with the least-square methods for efficiently
understanding and calculating closed-form linear
solutions.[3]. Reciently, there has been revived interest in the
topic of mathematical data estimation L 1, see ref, for
example. Based on our prior experience of shock-related
change, we suggest that photos be identified by reducing the
overall variance standard of the expected solution. As a
nonlinear time based PDE, we extract a selective
minimization algorithm where noise statistics agree on the
constraints. Classic methods try to decrease / delete the
portion of noise before further processing. That is the way
that report takes place. However, hybrid algorithms that
combine denoising with other analytical picture noise tasks
that be built utilizing the same TV / L1 principle. It is
understood that the recorded Footage is decaying due to a
combination of noise[4], including the shooting environment,
with high resolution camera video processing[5]. Because
noise is comprised of HF components, noise-mixed pictures
are super-resolutionary in terms of photo quality and
incongruous images. Therefore, noise from initial recordings
can be minimized in order to prevent degradations related to
vibration when surveillance camera imagery is super-
resolved. In comparison to the standard method of
regularization of the total variance, we have attempted to
increase the image quality of videos by enabling more
precise separation of the video elements, taking into
consideration the average variability of the time axis.. The
bulk of this article has the following form[6]. At college. At
work. 2, we study full variance of the -order and the space
for the restricted variance functions of the fractional-order.
We then add other property to this space. Section 3 provides
detailed algorithms of noise reduction based on a mixed
paradigm and absolute variance Nonlinear. At college. At
work. 4, Comprehensive experiments and associated reviews
are described[7]. The conclusion is outlined in Sect. 5.
A. The proposed single-color picture algorithm for
super resolution.
The single-frame superresolution (SR) reconstruction
replaces a deformed and low-resolution (LR) observing file
with the lost high-frequency information of the initial
image[8]. The goal of the SR restoration is to fix the inherent
defects in LR artifacts, so that the restored high-resolution
(HR) image shows better visual effects.[9].
II. METHODOLOGY
AND
DESIGN
Software production framework deployment. Figure 2
displays the whole stream row of the scheme and method
phases. If configuration and parameter control are required in
this document, programming[10]. The closure of this region
shall be closed with the solution description and the date of
incorporation of research proposal into the report..
Saab Hazim Ahmed
Department of Information Technology,
Altinbas University
Saeb.h.ahmed@gmail.com
Alaa Hamid Mohammed
Department of Computer Engineering,
Karabuk University, Turkey
Aallaaha12@gmail.com
Sefer KURNAZ
Department of Electrical And Computer
Engineering, Altinbas University
sefer.kurnaz@altinbas.edu.tr
978-1-7281-9090-7/20/$31.00 ©2020 IEEE
Authorized licensed use limited to: ULAKBIM UASL - Altinbas Universitesi. Downloaded on November 22,2020 at 15:29:32 UTC from IEEE Xplore. Restrictions apply.
III.PROPOSED
METHODOLOGY
Problem: consider the anonymity point k, which is a
sequence of movement elements (a1, a1, a2, .... am, the
sensing nodes collection (n 1, n2 ...) which does nothing to
detect (d1, d2,. ..), dp respectively[11].
A. The proposed single-color picture algorithm for
super resolution
A color image of HR with color image from an LR input is
to be reconstructed. A sparse representation vector for a D 1
dictionary acquired from LR patching is defined for each
LR patch patch feedback on the basis of sparse
representation model.[12]. You should think about searching
for the sparsest variable of y as:
ԡߙԡ
ݏǤݐǤԡܦ
ן
െݕԡ߳ (1)
where ԡǤԡ
is the ܮ
-norm. is the sparse representation
vector
The HR patch x performance is then generated according
to the vector and Dh from HR patches [13].
Optimization is an NP-hard problem (1). Usually, it is
intractable to solve the issue. The L1 regularization
technique is also well known, mostly used to research
fragmented SR dictionaries and recreate SR.[14]. The sparse
optimization L1 problem is established as follow-up:
ሼܦǡןሽൌ
ሼǡሽ
ԡݕെܦܽԡ
ଶ
ଶ
ܦܽԡߣ
ଶ
ଶ
ԡԡܽԡ
ଵ
(2)
The L1 regularization problem can be translated into a
quadratic optimization problem and can be overcome
effectively[15]. Nevertheless, more variations in the sparse
repre-send coefficient calculation may be induced by the ܮͳ
regulationari zation. [16]. This aims to yield solutions that
are typically not sparse enough and possible to obtain strong
sparse approximation results for image data influenced by
noise.
The L2/3 regularization [17] The sparsity of the vectors
has been shown to be more efficient. The regularisation
L2/3 is less vibrational than the regularization process L1
and more stable than the.
Our algorithm therefore constructs[18], as follows, a joint
sparse SR dictionary model based on L2/3 regularization:
ሼܦǡןሽൌ
ሼǡሽ
ԡݕെܦܽԡ
ଶ
ଶ
ܦܽԡߣ
ଶ
ଶ
ԡԡܽԡ
ଶଷ
ൗ
(3)
where ԡǤԡ
ଶଷ
ൗ
is the L2/3-norm. is a regularization
parameter that can balance the sparsity of the solution and
the fidelity of the approximation to y. ܦൌܦ
ܰ
ൗ
ܦ
ܯ
ൗ denote
ܪܴܮܴ
ൗ dictionary pairs. ݕൌܻ
ܰ
ൗ
ܻ
ܯ
ൗ represents HR/LR The
reference patches of the exercise and {N, M} are the
proportions of the image LR and HR respectively. A method
of closed-form requirement for HR / LR dictionary pair
planning [16] is adopted in the sparse coding phase, and the
K-singular value decomposition method [19]is used in the
dictionary atom updating phase
B. Nonlinear partial differential equations based
denoising algorithms.
Let the observed intensity function ݑ
ሺݔǡݕሻ denote the pixel
values of a noisy image forݔǡݕא. Let ݑሺݔǡݕሻ denote the
desired clean image, so
ݑ
ሺݔǡݕሻൌݑሺݔǡݕሻ݊ሺݔǡݕሻǡ (3)
when ݊ is the additive noise.
We, of course, wish to reconstruct ݑ from ݑ
. Most
conventional variational methods involve a least squares ܮ
ଶ
fit because this leads to linear equations. The first attempt
along these lines was made by Phillips [19]and later refined
by Twomey [18] in the one-dimensional case. In our two
Figure 2 : REAL's system architecture.
Random Forest
+
+
R
ESUL
Normalization
Nonlinear based
i
Sparse representation
START
Authorized licensed use limited to: ULAKBIM UASL - Altinbas Universitesi. Downloaded on November 22,2020 at 15:29:32 UTC from IEEE Xplore. Restrictions apply.
dimensional continuous framework their constrained
minimization problem is
minimize ൫ݑ
௫௫
ݑ
௬௬
൯
ଶ
ஐ
(4)
subject to constraints involving the mean
ݑൌ
ஐ
ݑ
ஐ
(5)
and standard deviation
ሺݑെݑ
ሻ
ଶ
ஐ
ൌߪ
ଶ
(6)
The resulting linear equation is now easy to solve for
advanced computer linear algebra. Nonetheless, the findings
of the same constraints (but stronger than the MEM) are
shocking again-see, for example, ref.
The ܮ
ଵ
norm is usually prevented because language
anomalies including ȁݑȁ
ஐ
dx It generates special
distributions as coefficients (such as 6 functions) which can
not be handled in a strictly algebraic way. Still, I do not
know., if ܮ
ଶ
and ܮ
ଵ
Approximations are placed side by side
in a computer panel, it's clear the ܮ
ଵ
Close-up seems
different than the "same"ܮ
ଶ
approximation[21]. The "same"
means subject to similar restrictions. It could be at least
partially psychological because in shock measurements it is
well understood that the ܮ
ଵ
The standard gradient is an
adequate area. It is literally the realm of total constant
variance: BV. We get rid of false oscillations easily, when
there are simple signals here.
In ref. [6] A new technique for enhancing photos called
Shock filters was created by the first reviewer. This had
similar shock wave calculations in computational fluid
mechanics. The nature of oscillation free discontinuities and
the importance of the TV norm is investigated here.
N the first two authors' paper [15] The theory of total
variance preservation was also adopted. There were finite
differentiation systems developed, which were used to boost
slightly blurred images dramatically while retaining the
original picture variation.
Additionally, in [7] A fascinating stability restoration
algorithm focused on mean curvature movement was
developed by Alvarez, Lions, and Morel.. [22]. The mean
variance curvature is literally the derivative from Euler-
Lagrange. Therefore, we notice that the space for BV
functions is the right class for many basic tasks.
And our restricted question of minimization is:
Minimize ඥݑ
௫
ଶ
ݑ
௬
ଶ
݀ݔ݀ݕ
ஐ
(7)
Subject to restrictionsݑ
.
We have put on the same two limitations in our research so
far as:
ݑ݀ݔ݀ݕ
ஐ
ൌݑ
ஐ
݀ݔ݀ݕ (8)
This constraint signifies the fact that the white noise
݊ሺݔǡݕሻ in (2.1) is of zero mean and.
IV.SIMULATION
RESULTS
We must first discuss two significant issues for the proposed
algorithm in this section, including the scale of the training
dictionary and the regularisation parameter.ߣ. Instead, the
proposed approach was equated with other SR approaches,
such as bicubic interpolation (BI), sparse SR coding(ScSR),
NCSR channels and MCcSR. BI is the most popular
reconstruction process. ScSR is a minimal restoration
approach that uses widely studied dictionaries to restore
photos of HR. NCSR is a sparse type of SR which focuses
on a sparse book and is not organized locally. MCcSR is an
interpretation to color SR by researching colored
dictionaries that help edge comparisons.
Joint SR dictionaries with the same instruction sets are
eligible. See Section 1 Images from regularly seen sets like
Set5, Set14 etc. See pictures. The recovered artifacts with
different approaches are tested visually and qualitatively by
means of the peak signal tonoise ratio ( PSNR) and
structural similitude (SSIM). They use PSNR and SSIM
only on the light axis of the restoved HR images and the
initialHR images to perform the quantitative tests.
A.
Super-resolution image with sparse depiction.
In Figure 4 our findings are contrasted with five neighbors
and one female embedders. In both cases, our approach
provides finer points and more accurately reconstructs the
particulars of the scene. On the appearance of the baby there
are major improvements.
Authorized licensed use limited to: ULAKBIM UASL - Altinbas Universitesi. Downloaded on November 22,2020 at 15:29:32 UTC from IEEE Xplore. Restrictions apply.
Figure 1 Comparing our system 's output with the
neighborhood embedding (NE) outputs. The NE strategy is
similar to ours in the sense that both approaches generate the
picture patch with the low detail patch utilizing the linear
combination weighs. The NE approach uses fixed k-
neighbours, compared to our process, to find the
rehabilitation supports directly from sampled training parts
and does not have a dictionary training level. To obtain a
reasonable estimate we use the same 100 000 patch pairs of
NE method. Instead we look for the most visually
appreciable results. Our process is a lot faster but also
lightweight shaper results.
B. Noise Removal Super-resolution
The results of noise reduction by BM3D and the proposed
method are compared to confirm the efficacy of the
proposed method. Table I present the TV Regularization
Experimental conditions. Figure 4 displays performance
images for BM3D and our proposed approach after the noise
reduction processes. Unlike BM3D, the components of the
texture are affected, while the approach suggested clearly
shows the details of the face in the picture and removes
noise on the skin. It can be verified that virtually all the
noise originally present was removed.
The super-resolution approach is used for the outputs of
noise reduction. The proposed method and BM3D are
doubled each time with the super-resolution technique. And
even when comparing with BM3D results, the disparity can
be clearly checked. On the other side, there is no need for
the accent. However, there is no no evidence of noise
improvement on the picture, and the accuracy of noise
removal using our proposed method can be confirmed.
V.
CONCLUSIONS
This paper suggests a new approach to a one-picture
super resolution based on sparse representations of high-
and low-resolution image pairs, jointly trained
dictionaries. Both locally and globally, there are
compatibilities between the adjacent patches.
Experimental tests show the sparsity effectiveness as a
precedent for the super-resolution of patch-based images
for the general and the nose. We presented in this paper
the entire regularization cycle, extended towards the
time axis in order to eliminate noise from the image. As
a consequence, noise can be exactly isolated, effective
noise reduction can be achieved for each part and a
high-definition super-resolution image can be produced
with a reduced noise. Potential issues include an
improvement in the efficacy of time axis noise
insulation. Nonetheless, the specification of the
Figure 1: a: Sparse Recovery Figure 1: b: bicubic Interpolation Figure 3: a: Original
Figure 2 :a : Input Image
Figure 3: b: Noisy Image
Figure 2 :b : bicubic Interpolation Figure 2 :c : Our Mathod
Authorized licensed use limited to: ULAKBIM UASL - Altinbas Universitesi. Downloaded on November 22,2020 at 15:29:32 UTC from IEEE Xplore. Restrictions apply.
appropriate dictionary size for natural image patches in
terms of SR assignments remains one of the most
relevant issues for potential investigations. Tighter
connections to the compressed sensing theory can result
in conditions for the appropriate patch size, features to
be used and also approaches to training the coupled
dictionaries.
R
EFERENCES
[1] X. Liu, Y. Chen, Z. Peng, and J. Wu, “Infrared
image super-resolution reconstruction based on
quaternion and high-order overlapping group sparse
total variation,” Sensors, vol. 19, no. 23, p. 5139,
2019.
[2] W. Witwit, Y. Zhao, K. Jenkins, and S. Addepalli,
“Global motion based video super-resolution
reconstruction using discrete wavelet transform,”
Multimed. Tools Appl., vol. 77, no. 20, pp. 27641–
27660, 2018.
[3] A. H. Mohammed, “A REVIEW SOFTWARE
DEFINED NETWORKING FOR INTERNET OF
THINGS,” 2020 Int. Congr. Human-Computer
Interact. Optim. Robot. Appl., no. August, 2020.
[4] M. M. Hamdi, L. Audah, S. A. Rashid, A. H.
Mohammed, S. Alani, and A. S. Mustafa, “A
Review of Applications, Characteristics and
Challenges in Vehicular Ad Hoc Networks
(VANETs),” in 2020 International Congress on
Human-Computer Interaction, Optimization and
Robotic Applications (HORA), 2020, pp. 1–7.
[5] R. Nayak, B. K. Balabantaray, and D. Patra, “A
New Single-Image Super-Resolution Using
Efficient Feature Fusion and Patch Similarity in
Non-Euclidean Space,” Arab. J. Sci. Eng., pp. 1–25,
2020.
[6] A. H. Ahmed, A. T. Al-Heety, B. Al-Khateeb, and
A. H. Mohammed, “Energy Efficiency in 5G
Massive MIMO for Mobile Wireless Network,” in
2020 International Congress on Human-Computer
Interaction, Optimization and Robotic Applications
(HORA), 2020, pp. 1–6.
[7] Y. Wang, J. Wang, F. Li, Y. Zhou, M. Xiong, and
M. Li, “Face Image Super-resolution Reconstruction
via Mapping Matrix and Multilayer Model.,”
IAENG Int. J. Comput. Sci., vol. 47, no. 1, 2020.
[8] S. KURNAZ and A. H. Mohammed, “Secure Pin
Authentication in Java Smart Card Using Honey
Encryption,” in 2020 International Congress on
Human-Computer Interaction, Optimization and
Robotic Applications (HORA), 2020, pp. 1–4.
[9] X. U. Zhigang, M. A. Qiang, and Y. Feixiang,
“Single color image super-resolution using sparse
representation and color constraint,” vol. 31, no. 2,
pp. 266–271, 2020.
[10] Z. A. Sahib, O. N. Uçan, M. A. Talab, M. T.
Alnaseeri, A. H. Mohammed, and H. A. Sahib,
“Hybrid Method Using EDMS & Gabor for Shape
and Texture,” in 2020 International Congress on
Human-Computer Interaction, Optimization and
Robotic Applications (HORA), 2020, pp. 1–6.
[11] A. G. Mushref, A. H. Mohammed, and O. Bayat,
“Rayleigh Leistungs Relation and Rician Fading
Channels in Qam Using Simulink Environment,” in
2020 International Congress on Human-Computer
Interaction, Optimization and Robotic Applications
(HORA), 2020, pp. 1–5.
[12] R. A. Dehkordi, H. Khosravi, and A. Ahmadyfard,
“Single image super-resolution based on sparse
representation using dictionaries trained with input
image patches,” IET Image Process., vol. 14, no. 8,
pp. 1587–1593, 2020.
[13] A. H. Mohammed, M. M. Hamdi, S. A. Rashid, and
A. M. Shantaf, “An Optimum Design of Square
Microstrip Patch Antenna Based on Fuzzy Logic
Rules,” in 2020 International Congress on Human-
Computer Interaction, Optimization and Robotic
Applications (HORA), 2020, pp. 1–7.
[14] S. Ayas and M. Ekinci, “Single image super
resolution using dictionary learning and sparse
coding with multi-scale and multi-directional Gabor
feature representation,” Inf. Sci. (Ny)., vol. 512, pp.
1264–1278, 2020.
[15] A. H. Mohammed, A. M. Shantaf, and M. Khalaf,
“The Probe into Reflection Mobile Radio
Propagation,” in 2020 International Congress on
Human-Computer Interaction, Optimization and
Robotic Applications (HORA), 2020, pp. 1–4.
[16] J. Zhang, M. Shao, L. Yu, and Y. Li, “Image super-
resolution reconstruction based on sparse
representation and deep learning,” Signal Process.
Image Commun., p. 115925, 2020.
[17] A. M. Shantaf, S. Kurnaz, and A. H. Mohammed,
“Performance Evaluation of Three Mobile Ad-hoc
Network Routing Protocols in Different
Environments,” in 2020 International Congress on
Human-Computer Interaction, Optimization and
Robotic Applications (HORA), 2020, pp. 1–6.
[18] A. H. Mohammed, R. M. KHALEEFAH, and A. H.
AlMarzoogee, “The Method Of Calibration
Compensation For Fiber Nonlinearity-A Review,”
in 2020 International Congress on Human-
Computer Interaction, Optimization and Robotic
Applications (HORA), 2020, pp. 1–8.
[19] B. Deka, S. Datta, H. U. Mullah, and S. Hazarika,
“Diffusion-weighted and spectroscopic MRI super-
resolution using sparse representations,” Biomed.
Signal Process. Control, vol. 60, p. 101941, 2020.
[20] A. H. Mohammed, “An Optimum Design of Square
Microstrip Patch Antenna Based on Fuzzy Logic
Rules,” 2nd Int. Congr. Human-Computer Interact.
Optim. Robot. Appl., 2020.
[21] S. Vilov, B. Arnal, E. Hojman, Y. C. Eldar, O. Katz,
and E. Bossy, “Super-resolution photoacoustic and
ultrasound imaging with sparse arrays,” Sci. Rep.,
vol. 10, no. 1, pp. 1–8, 2020.
[22] Z. Xu, Q. Ma, and F. Yuan, “Single color image
super-resolution using sparse representation and
color constraint,” J. Syst. Eng. Electron., vol. 31, no.
2, pp. 266–271, 2020.
Authorized licensed use limited to: ULAKBIM UASL - Altinbas Universitesi. Downloaded on November 22,2020 at 15:29:32 UTC from IEEE Xplore. Restrictions apply.
