Professional Documents
Culture Documents
PDF Intelligent Technologies and Applications First International Conference Intap 2018 Bahawalpur Pakistan October 23 25 2018 Revised Selected Papers Imran Sarwar Bajwa Ebook Full Chapter
PDF Intelligent Technologies and Applications First International Conference Intap 2018 Bahawalpur Pakistan October 23 25 2018 Revised Selected Papers Imran Sarwar Bajwa Ebook Full Chapter
https://textbookfull.com/product/software-technologies-
applications-and-foundations-staf-2018-collocated-workshops-
toulouse-france-june-25-29-2018-revised-selected-papers-manuel-
mazzara/
https://textbookfull.com/product/computational-intelligence-and-
intelligent-systems-10th-international-symposium-
isica-2018-jiujiang-china-october-13-14-2018-revised-selected-
papers-hu-peng/
https://textbookfull.com/product/web-and-big-data-apweb-
waim-2018-international-workshops-mwda-bah-kgma-dmmooc-ds-macau-
china-july-23-25-2018-revised-selected-papers-leong-hou-u/
https://textbookfull.com/product/geo-informatics-in-sustainable-
ecosystem-and-society-6th-international-conference-
gses-2018-handan-china-september-25-26-2018-revised-selected-
Imran Sarwar Bajwa
Fairouz Kamareddine
Anna Costa (Eds.)
Intelligent Technologies
and Applications
First International Conference, INTAP 2018
Bahawalpur, Pakistan, October 23–25, 2018
Revised Selected Papers
123
Communications
in Computer and Information Science 932
Commenced Publication in 2007
Founding and Former Series Editors:
Phoebe Chen, Alfredo Cuzzocrea, Xiaoyong Du, Orhun Kara, Ting Liu,
Krishna M. Sivalingam, Dominik Ślęzak, and Xiaokang Yang
Editorial Board
Simone Diniz Junqueira Barbosa
Pontifical Catholic University of Rio de Janeiro (PUC-Rio),
Rio de Janeiro, Brazil
Joaquim Filipe
Polytechnic Institute of Setúbal, Setúbal, Portugal
Ashish Ghosh
Indian Statistical Institute, Kolkata, India
Igor Kotenko
St. Petersburg Institute for Informatics and Automation of the Russian
Academy of Sciences, St. Petersburg, Russia
Takashi Washio
Osaka University, Osaka, Japan
Junsong Yuan
University at Buffalo, The State University of New York, Buffalo, USA
Lizhu Zhou
Tsinghua University, Beijing, China
More information about this series at http://www.springer.com/series/7899
Imran Sarwar Bajwa Fairouz Kamareddine
•
Intelligent Technologies
and Applications
First International Conference, INTAP 2018
Bahawalpur, Pakistan, October 23–25, 2018
Revised Selected Papers
123
Editors
Imran Sarwar Bajwa Anna Costa
Department of Computer Science and IT Department of Computer Engineering
Islamia University of Bahawalpur and Digital Systems
Baghdad, Pakistan University of Sao Paulo
São Paulo, Brazil
Fairouz Kamareddine
Mathematical and Computer Sciences
Heriot-Watt University
Edinburgh, UK
This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Preface
The present book includes the papers accepted for the First International Conference on
Intelligent Technologies and Applications (INTAP 2018), held in Bahawalpur,
Pakistan, during October 23–25, 2018, organized by the Artificial Intelligence
Research Group with the collaboration of the Sir Sadiq Association of Computing and
hosted by the Islamia University of Bahawalpur. The conference was sponsored by the
Higher Education Commission, Pakistan.
The conference was organized in 13 simultaneous tracks: AI and Health (5), Sen-
timent Analysis (5), Intelligent Applications (7), Social Media Analytics (6), Business
Intelligence (6), Natural Language Processing (5), Information Extraction (4), Machine
Learning (6), Smart Systems (5), Semantic Web (6), Decision Support Systems (6),
Image Analysis (7), and Automated Software Engineering (6).
We received 251 submissions, from 28 countries and districts representing all
continents. After a blind review process, only 67 were accepted as full papers and
seven were selected as short papers based on the classifications provided by the Pro-
gram Committee, resulting in an acceptance rate of 29%. The selected papers reflect
state-of-the-art research work in the different domains and applications of artificial
intelligence and highlight the benefits of intelligent and smart systems in various fields
of life. These high-quality standards will be maintained and reinforced at INTAP 2019,
to be held at Harriot-Watt University, Scotland, and in future editions of this
conference.
Furthermore, INTAP 2018 included four plenary keynote lectures given by Letizia
Jaccheri (Norwegian University of Science and Technology, Norway), Julia Sidorova
(Blekinge Institute of Technology, Sweden), M. Abbas Choudhary (DIHE, Karachi),
Irfan Hyder (IoBM, Karachi), Dr. Riaz ul Amin (BUITEMS, Quetta), and Dr. Khurram
Khurshid (IST, Islamabad). We would like to express our appreciation to all of them
and in particular to those who took the time to contribute with a paper to this book.
On behalf of the conference Organizing Committee, we would like to thank all
participants. First of all, the authors, whose quality work is the essence of the con-
ference., and the members of the Program Committee, who helped us with their
expertise in reviewing and selecting the quality papers for this book. It is well known
that organizing an international conference requires the effort of many individuals. We
wish to thank also all the members of our Organizing Committee, whose work and
commitment were invaluable.
General Co-chairs
Imran Sarwar Bajwa The Islamia University of Bahawalpur, Pakistan
Mark G. Lee University of Birmingham, UK
Anna Helena Reali Costa University of São Paulo, Brazil
Program Co-chairs
Fairouz Kamareddine Heriot-Watt University, UK
Imran Ghani Indiana University of Pennsylvania, USA
Jamal Bentahar Concordia University, Canada
Dayang Norhayati A. Universiti Teknologi Malaysia, Malaysia
Jawawi
Organizing Committee
Amir Hussain University of Stirling, UK
Irfan Hyder Institute of Business Management, Pakistan
Omair Shafiq Carleton University, Canada
M. Abbas Choudhary Dadabhoy Institute of Higher Education, Karachi,
Pakistan
Noreen Jamil Unitec Institute of Technology, New Zealand
Ghulam Alli Mallah SALU, Khairpur, Pakistan
Riaz ul Amin BUITEMS University, Quetta, Pakistan
Aman Ullah Yasin CASE, Islamabad, Pakistan
Imran Memon Zhejiang University, China
Rafaqut Hussain Kazmi The Islamia University of Bahawalpur, Pakistan
Program Committee
Adel Al-Jumaily University of Technology Sydney, Australia
Adina Florea University Politehnica of Bucharest, Romania
Adriano V. Werhli Universidade Federal do Rio Grande, Brazil
Agostino Poggim Università degli Studi di Parma, Italy
Ales Zamuda University of Maribor, Slovenia
Alexander Gelbukh National Polytechnic Institute, Mexico
Amin Beheshti Macquarie University, Australia
Anand Nayyar Duy Tan University, Vietnam
António Luís Lopes Instituto Universitário de Lisboa, Portugal
Anna Helena Reali Costa University of São Paulo, Brazil
Alvaro Rubio-Largo Universidade NOVA de Lisboa, Portugal
VIII Organization
Invited Speakers
Letizia Jaccheri Norwegian University of Science and Technology,
Norway
Julia Sidorova Blekinge Institute of Technology, Sweden
M. Abbas Choudhary Dadabhoy Institute of Higher Education, Karachi,
Pakistan
Syed Irfan Hyder Institute of Business Management, Karachi, Pakistan
Contents
AI and Health
Sentiment Analysis
Intelligent Applications
Parsing RDFs to Extract Object Oriented Model Using Apache Jena . . . . . . . 119
Umar Farooq Shafi, Hina Sattar, Imran Sarwar Bajwa, and Amna Ikram
Business Intelligence
Information Extraction
A Graph Theory Based Method to Extract Social Structure in the Society . . . 437
Wajid Rafique, Maqbool Khan, Nadeem Sarwar, Muhammad Sohail,
and Asma Irshad
Machine Learning
Smart Systems
Semantic Web
Enrich Existing Ontologies with New Knowledge from Existing Artifacts . . . 630
Amna Mehmood, Ayesha Mehmood, and Bakhtawer Akhtar
Image Analysis
1 Introduction
is similar in all the Machine learning algorithms i.e., Mean Square Error (MSE), PSNR
(Peak Signal to Noise Ratio) and SSIM (Structural Similarity Index Metric).
Dabov [2] et al., proposed an image denoising algorithm using sparse transfor-
mation of 2D image fragments into 3D data arrays resulting a group of jointly filtered
image blocks. The collaborative filtering technique consists of three successive steps
such as, 3D Transformation of a group, shrinkage of a spectrum and Inverse-3D
transformation which results a successive image denoising technique. The metric used
in his 3D fusion technique is Peak Signal to Noise Ratio (PSNR) and Perspective
visualization. He has used different types of wavelet transformation techniques such as
DCT, DST and Walsh-Hadmard Transforms with different wavelets (Haar, Bior, Sym
etc.) into consideration with the BM3D Denoising technique.
An another algorithm improving the performance of denoising is done by imple-
menting Non-Local Means (NLM) and Block Matching 3D (BM3D) Fusion technique
has been proposed by Talebi [3] et al. by designing a global filter which can be applied
to the images to enhance the patch based methods by estimating a pixel from all the
pixels of the image. He stated that the denoising approaches such as bilateral filter,
LARK, BM3D, NLMS and (Patch Based Locally optimum filter) PLOW are data
dependent filtering schemes which uses each pixel individually from the neighbouring
pixels. In his work, he used Nystroms extension for image segmentation and decom-
posed them into corresponding Eigen vectors and applied them to an iterative and
trucking filter (e.g., BM3D and NLMS) and calculated PSNR for those images which
were later compared with his global filtered techniques.
Many Shallow Deep learning and Machine learning algorithms has come across
after that, but the real question is that, did the deep learning algorithms can keep up an
accuracy with the existing models? This question was later on explained by Burger [4]
et al., in his proposed paper. In his work, he applied a simple Multi-Layer Perception
(MLP) algorithm to the image patches and observed the resultant metrics (PSNR) for
different types of Machine Learning techniques such as GSM, KSVD and BM3D
techniques at different Noise levels. But the main disadvantage is that he compiled that
model using GPU instead of CPU which would be very difficult in the real time
scenarios as only few men can afford them.
Image Denoising of Ultra Low dose CT scan images were done by Nishio [5] et al.
In his work, he investigated the performance of Patch based Neural Network trained for
Ultra Low DCT images using Convolution Auto Encoders (CAE) and compared its
performance with NLMS & BM3D techniques using PSNR.
SSIM. As DAE ignores 2D image structure and CAE preserves the 2D image
structural data he used CAE instead of DAE for denoising medical Image datasets.
Denoising Auto Encoders using CNN architecture was designed by Gondara [6]
et al. In his work, he explained that the performance of image denoising can be further
enhanced by exploiting strong spatial correlations by constructing DAEs with Con-
volution layers. He tested his architecture on mini-MIAS database and Dental Radio
Graph database and achieved a strong SSIM of 0.89 and 0.90 when comparing to
NLMS and Median Filters.
So In this paper, we proposed a new Tensorflow and Theano architectures for DAE
which is having only three Dense layers and achieved a maximum accuracy of 99.98%
and 99.97% with Visual Proximities and compared with the novel architecture of our
Enhanced Medical Image De-noising Using Auto Encoders and MLP 5
Multi Linear Perception (MLP) Neural Network which achieved 99.98% accuracy for
the lower samples of Dental Radio Graphy database and mini-MIAS database. These
architectures are trained for 250 epochs except for MLP which is trained for just 85
epochs to achieve a saturation stage of 99.99% accuracy.
The architectures in the paper will be explained later in the Sect. 2 & Keras
Backgrounds in Sect. 3 and the architectures of our models in Sect. 4, results along
with tensor graphs will be explained in the Sect. 5 and the conclusion, future work (i.e.,
for real time scenarios) will be explained in Sect. 6 following with References. The
main difference between auto encoders and MLPs is that, the generated MSE is
truncated back to the hidden layers in MLP. The weight sharing mechanism between
encoders and decoders in the network enables fast self-learning process but at the same
time it restricts the degree of freedom of the network.
Auto Encoders are self-supervised, feed forward and non-recurrent neural networks
which has a belief in deep learning to find applications in Image Dimensionality
reduction, data denoising, clustering, machine translation and anomaly detection etc.,
y ¼ s ðWx þ bÞ ð1Þ
Where b is the biasing function and “s” could be any non-linear function of h(x).
Now the reconstructed signal function (r) can be rewrite in the form of
r ¼ sðW0 y þ b0 Þ ð2Þ
As said earlier, the weight sharing concept of the auto encoder, that is the weights
of the encoder and decoder are simply a transpose of each other, which we can see that
in the Eqs. 1 and 2. Hence it gives a deterministic approach.
The architecture of the stacked denoising auto encoder (SDAE) is best explained by
Gondara [6] in his work on Medical Image Denoising using Conventional Auto
Encoders.
Since each neuron of the hidden layer is connected to all the neurons of the
previous hidden layers, it is also called as Fully Connected Layers.
Unlike DAE’s or CAE’s, MLP performs Back Propagations, i.e., feed backing the
error signal to the hidden layers to change the weights of the neurons in the hidden
layers. The activity of the neurons can be expressed by an activation function,
y ¼ RðWx þ bÞ ð3Þ
In our presenting work, we used “Sigmoid” Activation Function for each neuron in
the hidden layers to limit its probabilities from {0, 1} (Fig. 3).
3 Keras Environments
While designing a neural network using Keras, we have used two types of back-
grounds, they are, Tensorflow and Theano, so as to ensure which Backend gives the
maximum accuracy at which epoch. In our work, we have tested SDAE on mini-MIAS
database and Dental Radio Graphy database on both Keras environments and found
that Theano has achieved a maximum accuracy of 99.97% at 141th epoch and Ten-
sorflow has achieved a slight more accuracy of 99.98% at 152nd epoch. Whereas, MLP
has achieved 99.99% in retrieving the original image at 87th epoch.
In this section, we are going to discuss about the architectures of the proposed Stacked
Denoising Auto Encoders and Multi-Layer Perception Algorithms.
As we can see in the Fig. 4 the stacked Denoising architecture consists of one input
layer of size (64, 64, 1) and two fully connected layers having input dimension 64 with
an activity function of Sigmoid to each neurons in the fully connected layers (Fig. 5).
Figure 6 shows the similar architecture of SDAE but the difference is that the
background environment has been changed to Theano instead of Tensorflow. So as the
input dimension has been changed to “1” instead of 64 and the shape of the input image
has been changed to (1, 64, 64). And the rest of the architecture is similar to that of
SDAE using Tensorflow. It’s like a mirror to the original architecture of Tensorflow
(Fig. 7).
In the MLP architecture, we have added some dropout layers when comparing to
SDAEs, as it supports Back propagation of the resulting errors to the hidden layers
8 S. S. Kunapuli et al.
Fig. 4. Stacked denoising auto encoders (SDAE) architecture using Tensorflow background
Fig. 5. Depicts the SDAE tensorflow architecture which was visualized using tensor board.
Fig. 6. Stacked denoising auto encoders (SDAE) architecture using Theano background
making variation weights of the neurons and also restricting the degree of freedom of
the network architecture. Hence, we used Dropouts to regularize the network archi-
tecture (Figs. 8 and 9).
Enhanced Medical Image De-noising Using Auto Encoders and MLP 9
Fig. 7. Depicts the SDAE Theano architecture which was visualized using Tensor board.
5 Results
In this section, the performance of our algorithm is compared with the results of the
previous models with the help of a tensor board and images.
10 S. S. Kunapuli et al.
Fig. 10. Graph visualization for decreasing loss during recovering till the last pixel of the
medical images
Enhanced Medical Image De-noising Using Auto Encoders and MLP 11
Fig. 11. Depicts Added Gaussian Noise to the Medical images (As we added same amount of
noise as input to the every architecture, we are representing this Fig. 11 as noise image to all the
architectures)
Fig. 12. Depicts the original and retrieved images using SDAE tensorflow architecture. You can
see there is a slight difference between the images which also negotiable in bio-medical
Instrumentation system but yet it yields a fruit-full results.
Even it got 99.99% of accuracy, there is some slight variation of the Denoised from
the original images. So in a case of diagnosing cancers, it could be somewhat crucial
problem (as we are processing the black box). Hence we are undergoing some pre-
processing steps, which will be extended later to our work so as to achieve 100%
accuracy which will be our future scope (Figs. 13, 14, 15 and 16).
Fig. 13. Depicts the SDAE output using Theano architecture (Here you can see that as model is
same but the architecture is different the fall of error will be different but yields very similar
results to the SDAE tensorflow architecture.)
12 S. S. Kunapuli et al.
Fig. 14. Depicts the original and retrieved images using SDAE Theano architecture.
Fig. 15. Depicts the MLP output during denoising (As we can see the difference between these
graphs as how it falls from the peak value while denoising the noisy images.)
Fig. 16. Depicts the original and denoised Images using MLP architecture.
The Aswan Dam is a huge granite barrier a mile and a quarter long which now
controls the waters of the Nile after centuries of alternate flood and drought, saves
Egypt from famine, and adds millions of acres to her irrigable lands.
It brings one close to the days of the Scriptures when he can put
his hand on the very same things that were touched by old Pharaoh;
and can visit the temples in which he worshipped, or sit on the
monuments erected in his honour, and look at the tomb in which his
royal bones were laid away. One feels closer still when he can look
at the royal mummy itself and actually see the hardhearted old
heathen almost as he was when alive, as I did at the museum the
other day.
This Pharaoh, Rameses II, was one of the greatest kings of
ancient Egypt. His temples are scattered throughout the Nile valley
and his statues are the largest ever discovered. One was found in
the Nile delta which measures forty-two feet in height, and there are
others sixty-six feet high at Abu Simbel in Nubia, about as far up the
Nile as Chicago is distant from the mouth of the Hudson. They are
seated on thrones and are hewn from the solid rocks. These figures
stand in front of the temple, also cut out of rock. This building is said
to have been erected by him in honour of his favourite wife,
Nefertari, and there are statues of his children about it. These show
that he was very much of a family man, for inscriptions on the
various monuments mention one hundred and sixty-two of his
children by name.
CHAPTER XVI
THE NILE IN HARNESS
For the last two days I have been steaming through one of the
oldest lands of the globe. I have been travelling up the Nile through
the country which belonged to Noah’s grandson, Cush, who was
Ham’s eldest son, and which was known to the Greeks and Romans
in later days as Ethiopia. The Egyptians called it Nubia, from their
word noub, which means gold, and it is known that a large part of the
gold of ancient time came from it.
Ancient Nubia had a considerable population, and was noted for
its riches and power. It was already a flourishing country about the
time of the Pyramid builders, while in the most prosperous days of
Old Egypt it had large towns and magnificent temples dedicated to
the worship of the Egyptian gods. On my way here I passed Abu
Simbel, a great temple on the bank of the Nile, which was cut out of
the rocks by Rameses II, the Pharaoh of the Bible. Farther down the
river lies the Temple of the Lions, where that same old king was
himself worshipped as a god.
Until 1100 b.c. this country was a dependency of the Pharaohs. It
then became independent, and later its armies overran and
conquered Egypt. As other nations came into this part of the Nile
valley they sent their armies against the Nubians, but were driven
back, and at the time the Romans came the country was ruled by a
succession of queens named Candace, one of whom made war
upon the Romans. The Nubian people very early adopted
Christianity, but later, when the Mohammedans took possession of
Egypt and the Upper Nile valley, they were converted to Islam. They
are still followers of the Prophet, and were among the boldest
soldiers of the fanatical Mahdi in his fights against the troops of
Egypt and Great Britain.
A land with such a history ought to be a rich one. The Nubia of to-
day is about as barren as any country on earth. With the exception of
a narrow band along the Nile, it is altogether desert. Beginning in the
sands of Libya, it extends several hundred miles eastward to the
Red Sea, but only in a few places has the soil enough moisture to
furnish even a scanty pasturage for camels and sheep. The bulk of
the desert population is made up of Bisharin Bedouins, living in tents
made of matting and moving about from place to place with their
flocks. Each tribe has a certain number of wells, and water is the
principal part of its visible wealth. The British officials of the Sudan
have surveyed these wells and investigated their depth and the
quality of the flow of the water. The government has also sunk some
new wells and found water at a depth of about one hundred feet.
Nubia is now a part of the Upper Nile valley, a cultivated strip, in
places only a quarter of a mile wide, winding its way like a snake
from north to south as far as from New York City to Detroit, and
extending on both sides of the river. It is of irregular width, for in
some places the desert comes close to the river, while in others the
stream winds through black rocky hills which rise straight above it a
thousand feet. Farther on, one sees yellow sand, spotted with black
rocks, which show signs of volcanic origin, and then at a low bend in
the river the water may be conducted out over the sands and create
a cultivated patch three miles in width.
The Nile is so walled in by hills that its waters have to be lifted in
order to flow over any level place. This is done chiefly by the
sakiehs, of which there are something like four thousand on the
Nubian Nile. The great wheels, moving in cogs, can be seen high up
on the banks, with their strings of buckets hanging to them. As the
buckets descend, each dips into the water and carries to the top a
few quarts at a time. In some places men raise the water in baskets
or buckets, and in others, the river slopes at such an angle that they
carry it up by hand and water little patches twenty or thirty feet wide.
Every low place along the river is farmed, and when the Nile falls,
the sand banks and islands are planted to crops.
Wherever there is a stretch of cultivated land, a village of mud and
stone huts has grown up, and such villages spot the banks for
hundreds of miles. At times there is no green except between village
and river, and one wonders how men can be born and live and die
there. Nevertheless, there are more than one hundred thousand
people to whom this region is the centre of the world.
Though much of this Nile border is too narrow for profitable
cultivation, it is very fertile and raises excellent cotton. At present the
other chief crops are wheat, barley, and millet, and the chief fruit is
dates, which are sweeter and larger than those grown farther down
the Nile valley. Indeed, the date trees that one sees almost
everywhere along the banks are a source of revenue for the
government, which taxes them at the rate of ten cents per tree.
“On the Ibis we make about six miles an hour as our dusky Nubian pilot
corkscrews up the Nile. Fortunately we are almost free from the myriad flies, the
modern plague of Egypt.”
Though the Aswan Dam has been of inestimable benefit to Egypt, the whole
world shares regret that when the sluice gates are closed the water backs up and
submerges Pharaoh’s Bed and other ancient ruins on the Island of Philæ.