IJACSA Volume2No6

(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 2, No. 6, 2011

(i)
www.ijacsa.thesai.org
IJACSA Editorial
From the Desk of Managing Editor
It is a pleasure to present our readers with the June 2011 Issue of International Journal of Advanced Computer Science and
Applications (IJACSA).
IJACSA is one of the most prominent publications in the field, delivering up-to-date and authoritative coverage of advanced
computer science and applications.
The Journal aims to publish the highest quality material, both informative and scientific, on all aspects of Computer Science
research. It includes articles related to research findings, technical evaluations, and reviews. In addition it provides a forum for
the exchange of information on all aspects.
The Associate Editor Board consists of senior level researchers and/or practitioners, who constantly advise the Co-Editors-in-
Chief with respect to the content of and direction to be pursued by the Journal. They may also from time to time be called upon
to review manuscripts but that is not their primary role.
Some of the papers have an introductory character, some of them access highly desired extensions for a particular method, and
some of them even introduce completely new approaches to computer science research in a very efficient manner. This
diversity was strongly desired and should contribute to evoke a picture of this field at large. As a consequence only 32% of the
received articles have been finally accepted for publication.
The scope of the journal is fairly broad, although more defined than that of some other journals. Our focus is squarely on
experimental research, rather than on work that is largely descriptive.
By having in mind such future issues, we hope to establish a regular outlet for contributions and new findings in the field of
Computer science and applications. Therefore, IJACSA in general, could serve as a reliable resource for everybody loosely or
tightly attached to this field of science.
And if only a single young researcher is inspired by this issue to contribute in the future to solve some of the problems sketched
here or contribute to exiting methodologies and research work, the effort of all the contributors will be rewarded. In that sense
we would like to thank all the authors and reviewers that contributed to this issue for their efforts and their collaboration in this
project.
We hope to continue exploring the always diverse and often astonishing fields in Advanced Computer Science and Applications.
Thank You for Sharing Wisdom!

Managing Editor
IJACSA
Volume 2 Issue 6, June 2011
[email protected]
ISSN 2156-5570 (Online)
ISSN 2158-107X (Print)
2011 The Science and Information (SAI) Organization
Vol. 2, No. 6, 2011
(ii)
IJACSA Associate Editors
Dr. Zuqing Zhu
Service Provider Technology Group of Cisco Systems, San Jose
Domain of Research: Research and development of wideband access routers for hybrid fibre-
coaxial (HFC) cable networks and passive optical networks (PON)
Dr. Ka Lok Man
Department of Computer Science and Software Engineering at the Xi'an Jiaotong-Liverpool
University, China
Domain of Research: Design, analysis and tools for integrated circuits and systems; formal
methods; process algebras; real-time, hybrid systems and physical cyber systems;
communication and wireless sensor networks.
Dr. Sasan Adibi
Technical Staff Member of Advanced Research, Research In Motion (RIM), Canada
Domain of Research: Security of wireless systems, Quality of Service (QoS), Ad-Hoc Networks,
e-Health and m-Health (Mobile Health)
Dr. Sikha Bagui
Associate Professor in the Department of Computer Science at the University of West Florida,
Domain of Research: Database and Data Mining.
Dr. T. V. Prasad
Dean, Lingaya's University, India
Domain of Research: Bioinformatics, Natural Language Processing, Image Processing, Expert
Systems, Robotics
Dr. Bremananth R
Research Fellow, Nanyang Technological University, Singapore
Domain of Research: Acoustic Holography, Pattern Recognition, Computer Vision, Image
Processing, Biometrics, Multimedia and Soft Computing
Vol. 2, No. 6, 2011
(iii)
IJACSA Reviewer Board

A Kathirvel
Karpaga Vinayaka College of Engineering and Technology, India
Abbas Karimi
I.A.U_Arak Branch (Faculty Member) & Universiti Putra Malaysia
Dr. Abdul Wahid
Gautam Buddha University, India
Abdul Khader Jilani Saudagar
Al-Imam Muhammad Ibn Saud Islamic University
Abdur Rashid Khan
Gomal Unversity
Dr. Ahmed Nabih Zaki Rashed
Menoufia University, Egypt
Ahmed Sabah AL-Jumaili
Ahlia University
Md. Akbar Hossain
Aalborg University, Denmark and AIT, Greeceas
Albert Alexander
Kongu Engineering College,India
Prof. Alc-nia Zita Sampaio
Technical University of Lisbon
Amit Verma
Rayat & Bahra Engineering College, India
Ammar Mohammed Ammar
Department of Computer Science, University of Koblenz-Landau
Arash Habibi Lashakri
University Technology Malaysia (UTM), Malaysia
Asoke Nath
St. Xaviers College, India
B R SARATH KUMAR
Lenora College of Engineering, India
Binod Kumar
Lakshmi Narayan College of Technology, India
Dr.C.Suresh Gnana Dhas
Park College of Engineering and Technology, India
Vol. 2, No. 6, 2011
(iv)
Mr. Chakresh kumar
Manav Rachna International University, India
Chandra Mouli P.V.S.S.R
VIT University, India
Chandrashekhar Meshram
Shri Shankaracharya Engineering College, India
Prof. D. S. R. Murthy
SNIST, India.
Prof. Dhananjay R.Kalbande
Sardar Patel Institute of Technology, India
Dhirendra Mishra
SVKM's NMIMS University, India
Divya Prakash Shrivastava
EL JABAL AL GARBI UNIVERSITY, ZAWIA
Fokrul Alom Mazarbhuiya
King Khalid University
G. Sreedhar
Rashtriya Sanskrit University
Ghalem Belalem
University of Oran (Es Senia)
Hanumanthappa.J
University of Mangalore, India
Dr. Himanshu Aggarwal
Punjabi University, India
Dr. Jamaiah Haji Yahaya
Northern University of Malaysia (UUM), Malaysia
Prof. Jue-Sam Chou
Nanhua University, Taiwan
Dr. Juan Jos Martnez Castillo
Yacambu University, Venezuela
Dr. Jui-Pin Yang
Shih Chien University, Taiwan
Dr. K.PRASADH
Mets School of Engineering, India
Dr. Kamal Shah
St. Francis Institute of Technology, India
Kodge B. G.
S. V. College, India
Vol. 2, No. 6, 2011
(v)
Kunal Patel
Ingenuity Systems, USA
Lai Khin Wee
Technischen Universitt Ilmenau, Germany
Mr. Lijian Sun
Chinese Academy of Surveying and Mapping, China
Long Chen
Qualcomm Incorporated
M.V.Raghavendra
Swathi Institute of Technology & Sciences, India.
Madjid Khalilian
Islamic Azad University
Mahesh Chandra
B.I.T, India
Mahmoud M. A. Abd Ellatif
Mansoura University
Manpreet Singh Manna
SLIET University, Govt. of India
Marcellin Julius NKENLIFACK
University of Dschang
Md. Masud Rana
Khunla University of Engineering & Technology, Bangladesh
Md. Zia Ur Rahman
Narasaraopeta Engg. College, Narasaraopeta
Messaouda AZZOUZI
Ziane AChour University of Djelfa
Dr. Michael Watts
University of Adelaide, Australia
Mohammed Ali Hussain
Sri Sai Madhavi Institute of Science & Technology
Mohd Nazri Ismail
University of Kuala Lumpur (UniKL)
Mueen Malik
University Technology Malaysia (UTM)
Dr. N Murugesan
Government Arts College (Autonomous), India
Dr. Nitin Surajkishor
Vol. 2, No. 6, 2011
(vi)
NMIMS, India
Dr. Poonam Garg
Information Management and Technology Area, India
Rajesh Kumar
Malaviya National Institute of Technology (MNIT), INDIA
Rajesh K Shukla
Sagar Institute of Research & Technology- Excellence, India
Dr. Rajiv Dharaskar
GH Raisoni College of Engineering, India
Prof. Rakesh L
Vijetha Institute of Technology, India
Prof. Rashid Sheikh
Acropolis Institute of Technology and Research, India
Rongrong Ji
Columbia University
Dr. Ruchika Malhotra
Delhi Technological University, India
Dr.Sagarmay Deb
University Lecturer, Central Queensland University, Australia
Dr. Sana'a Wafa Al-Sayegh
University College of Applied Sciences UCAS-Palestine
Santosh Kumar
Graphic Era University, India
Shaidah Jusoh
Zarqa University
Dr. Smita Rajpal
ITM University Gurgaon,India
Suhas J Manangi
Microsoft India R&D Pvt Ltd
Sunil Taneja
Smt. Aruna Asaf Ali Government Post Graduate College, India
Dr. Suresh Sankaranarayanan
University of West Indies, Kingston, Jamaica
T V Narayana Rao
Hyderabad Institute of Technology and Management, India
Totok R. Biyanto
Industrial Technology Faculty, ITS Surabaya
Varun Kumar
Vol. 2, No. 6, 2011
(vii)
Institute of Technology and Management, India
Dr. V. U. K. Sastry
SreeNidhi Institute of Science and Technology (SNIST), Hyderabad, India.
Vinayak Bairagi
Sinhgad Academy of engineering, India
Vitus S.W. Lam
The University of Hong Kong
Vuda Sreenivasarao
St.Marys college of Engineering & Technology, Hyderabad, India
Mr.Zhao Zhang
City University of Hong Kong, Kowloon, Hong Kong
Zhixin Chen
ILX Lightwave Corporation
Vol. 2, No. 6, 2011
(viii)
CONTENTS

Paper 1: Successful Transmission Rate of Mobile Terminals with Agents in Segmented Ad Hoc Network
Authors: Kohei Arai, Lipur Sugiyanta
PAGE 1 12

Paper 2: Assessing 3-D Uncertain System Stability by Using MATLAB Convex Hull Functions
Authors: Mohammed Tawfik Hussein
PAGE 13 18

Paper 3: Automating the Collection of Object Relational Database Metrics
Authors: Samer M. Suleiman, Qasem A. Al-Radaideh, Bilal A. Abulhuda, Izzat M. AlSmadi
PAGE 19 27

Paper 4: A Decision Support Tool for Inferring Further Education Desires of Youth in Sri Lanka
Authors: M.F.M Firdhous, Ravindi Jayasundara
PAGE 28 35

Paper 5: An Authorization Mechanism for Access Control of Resources in the Web Services Paradigm
Authors: Rajender Nath, Gulshan Ahuja
PAGE 36 42

Paper 6: Design and Implementation on a Sub-band based Acoustic Echo Cancellation Approach
Authors: Zhixin Chen
PAGE 43 48

Paper 7: An Efficient Density based Improved K- Medoids Clustering algorithm
Authors: Raghuvira Pratap A, K Suvarna Vani, J Rama Devi, Dr.K Nageswara Rao
PAGE 49 54

Paper 8: Context Switching Semaphore with Data Security Issues using Self-healing Approach
Authors: M. Anand, Dr. S. Ravi, Kuldeep Chouhan, Syed Musthak Ahmed
PAGE 55 62

Paper 9: Mining Educational Data to Analyze Students Performance
Authors: Brijesh Kumar Baradwaj, Saurabh Pal
PAGE 63 69

Vol. 2, No. 6, 2011
(ix)
Paper 10: Comparison between Traditional Approach and Object-Oriented Approach in Software
Engineering Developmen
Authors: Nabil Mohammed Ali Munassar, Dr. A. Govardhan
PAGE 70 76

Paper 11: Estimation of Dynamic Background and Object Detection in Noisy Visual Surveillance
Authors: M.Sankari, C. Meena
PAGE 77 83

Paper 12: Improvement of Brain Tissue Segmentation Using Information Fusion Approach
Authors: Lamiche Chaabane, Moussaoui Abdelouahab
PAGE 84 90

Paper 13: Image Retrieval using DST and DST Wavelet Sectorization
Authors: Dr. H.B.Kekre, Dhirendra Mishra
PAGE 91 97

Paper 14: Performance Analysis of UMTS Cellular Network using Sectorization Based on Capacity and
Coverage
Authors: A.K.M Fazlul Haque, Mir Mohammad Abu Kyum, Md. Baitul Al Sadi, Mrinal Kar, Md. Fokhray
Hossain
PAGE 98 104

Paper 15: Architecture Aware Programming on Multi-Core Systems
Authors: M.R. Pimple, S.R. Sathe
PAGE 105 111

Paper 16: Implementation of ISS - IHAS (Information Security System Information Hiding in Audio
Signal) model with reference to proposed e-cipher Method
Authors: Prof. R. Venkateswaran, Dr. V. Sundaram, Director
PAGE 112 116

Paper 17: Image Compression using Approximate Matching and Run Length
Authors: Samir Kumar Bandyopadhyay, Tuhin Utsab Paul, Avishek Raychoudhury
PAGE 117 121

Paper 18: Interactive Intranet Portal for effective Management in Tertiary Institution
Authors: Idogho O. Philipa, Akpado Kenneth, James Agajo
PAGE 122 129

Vol. 2, No. 6, 2011
(x)
Paper 19: Iris Recognition Using Modified Fuzzy Hypersphere Neural Network with different Distance
Measures
Authors: S. S. Chowhan, U. V. Kulkarni, G. N. Shinde
PAGE 130 134

Paper 20: Multi-Agent System Testing: A Survey
Authors: Zina Houhamdi
PAGE 135 141

Paper 21: Video Compression by Memetic Algorithm
Authors: Pooja Nagpal, Seema Baghla
PAGE 142 145

Paper 22: Impact of Cloud Computing on ERP implementations in Higher Education
Authors: Ms. Shivani Goel, Dr Ravi Kiran , Dr Deepak Garg
PAGE 146 148

Paper 23: Algorithms for Content Distribution in Networks
Authors: Dr Deepak Garg
PAGE 149 151

Paper 24: Generation of Attributes for Bangla Words for Universal Networking Language(UNL)
Authors: Muhammad Firoz Mridha, Kamruddin Md. Nur, Manoj Banik, Mohammad Nurul Huda
PAGE 152 158

Paper 25: A MODIFIED METHOD FOR ORDER REDUCTION OF LARGE SCALE DISCRETE SYSTEMS
Authors: Dr. G.Saraswathi
PAGE 159 162
Vol. 2, No. 6, 2011
1 | P a g e
Successful Transmission Rate of Mobile Terminals
with Agents in Segmented Ad Hoc Network

Kohei Arai
Department of Information Science
Faculty of Science and Engineering, Saga University
Saga, Japan
Lipur Sugiyanta
Department of Electrical Engineering
Faculty of Engineering, State University of Jakarta
Jakarta, Indonesia

Abstract Mobile Wireless Ad-Hoc Network (MANET) is a
special kind of network, where all of the nodes move in time.
Random movement is the commonly used to model the mobility
in this network. Beside as a host, each node is intended to help
relaying packets of neighboring nodes using multi-hop routing
mechanism in order to solve problem of dead communication. In
recent years, a variety of routing protocols have been proposed
specially to accommodate this environment followed with its
performance evaluations, likes DSDV, DSR, AODV, and
extension OSPF. Researches in this network are mostly
simulation based. Research efforts haven't focused much in
evaluating network performance when applied to variable
number of nodes that involving agents and distributed over
several network areas. The paper performed simulations using
MANET simulator, which carried out based on the extension
OSPF routing protocol. The modification of OSPF protocol had
intended to suit the OSPF standard mechanisms for operating
with high performance over wireless networks characterized by a
broadcast-based transmission medium, in which the physical
layer connectivity is not fully meshed. Extensive simulation
scenarios have been conducted in the simulator with various
numbers of nodes, random and uniform movement direction, and
different agents quantity with different size of network areas. In
the performance evaluation of successful transmission (data)
packets, the OSPF protocol with throughput weighted metric will
be tested under different combination conditions of scenarios.

Keywords-MANET; agents; movement; packet; routing; network
area.
I. INTRODUCTION
Wireless Ad-Hoc Network consists of mobile nodes
platforms which are free to move in the area. One class of such
networks has been called mobile ad hoc networks (MANETs).
Node is referred to a mobile device which equipped with built-
in wireless communications devices attached and has capability
similar to autonomous router. The most important characteristic
of MANET is the dynamic topology, which is a consequence of
node mobility. Nodes can change position quite frequently,
which means that we need a routing protocol that quickly
adapts to topology changes. The node in an ad-hoc network can
be laptops and personal digital assistants which are often very
limited in resources such as CPU speed, storage capacity,
battery power and bandwidth, so the routing protocol should be
able to be reactive and capable of minimizing control traffic,
such as periodic update messages. To be effective, the routing
protocols should have keep the routing table up-to-date and
reasonably small, able to choose the best route for certain
destination (with weighted metrics in terms of number of hops,
reliability, throughput, and cost), and capable of converged
within an exchange of a small amount of messages.[8]
In MANET, mobile nodes connected with each other using
multi-hop wireless links. Because there is no static
infrastructure such as base stations, then if network move
arbitrarily thus network topology changes frequently and
unpredictably. Each node in the network forwards data packets
for other nodes. It makes routing protocol design much
difficult. The routing protocol which works well in fixed
networks does not show the same performance in MANET. In
these networks routing protocols should be more dynamic so
that they quickly respond to topological changes [14]. If two
nodes are not within proximity of its radio range, all packets
must pass through hop by hop by one or more intermediate
nodes.
To analyze MANET performance with hundreds or even
thousands of nodes, the network must be split into independent
layer 3 groups or domains. Smaller domains allow routing,
QoS and other network protocols to operate limitary on fewer
nodes, with cross-domain interaction only through overlapped
area nodes. This division has several key benefits. First, it
reduces overall protocol overhead. In most routing protocols,
for example, the aggregate route update overhead grows as
O(n
2
), where n is the number of nodes in a domain. Therefore,
using smaller domains with inter-domain interaction through an
overlapped area allows reducing the overall overhead. Second,
it made network life time longer. Nodes have salient feature of
energy-constrained devices. The battery of node is depleted by:
(i) computational processing and (ii) transmission/reception of
signal to maintain the signal-to-noise ratio above a certain
threshold. Although the energy consumption by computations
can be further reduced with new developments in low power
devices, the energy consumption by communications cannot be
overcome. Therefore, partitioning network into smaller domain
is essential to develop efficient networking algorithms and
protocols that are optimized for energy consumption. As a
consequence, when partitioning a network into domains, there
are some engineering rules that need to be taken into account.
For example, the partition should create balanced overlapped
areas; the partitions should minimize the expected traffic
among different areas so as to not overload the overlapped
Vol. 2, No. 6, 2011
2 | P a g e
area; there must be at least one path between every pair of
nodes in certain area and the nodes in overlapped area, and at
least one path must exist from overlapped area to nodes that
belong to other area for such transmission exchange-area to be
successfully take place; there must be at least one node that
existed in the overlapped area to act as an relay exchange-area
domain. Such nodes are called agents. Third, minimizing and
managing exchange-area traffic is desirable since it helps
ensuring that the backbone agents do not become congested
and maximizes packet forwarding via knowledgeable
exchange-area route paths. Agents ensure that no single area
suffers adversely from a disproportionately large volume of
overhead and data packets.

However, in this paper we focus on the impact of mobility
models on the performance of data packets transmission in the
large MANET that segmented into two domain areas. So our
observations regarding to discuss the effect of movement
mobility speed of the nodes to evaluate the performance of the
improved OSPF routing protocol with weighted throughput
metric, using self-made simulator that considering a dynamic
network size with varying number of movement speed at an
invariable pause time for mobile Ad-hoc network with
frequently and fast moving nodes, movement direction of
nodes and domain, either approaching agents or avoiding
agents, and different size of overlapped area which potentially
vary number of agents getting involved in the traffic at instance
of each cycles.
The remainder of the paper is organized as follows. Section
2 gives a brief description of the related work. Section 3
discusses some limitations of the simulation model and
provides the simulation model with mobility of our system and
describes the effect of some metrics, Section 4 presents the
evaluation performance network for data packets transmission,
and finally, we conclude the paper in Section 5.
II. RELATED WORK
Significant academic and industrial research has led to the
development of a variety of protocols, platforms and
architectures for reliable communication provided by the
MANET network. They are mostly centralized approaches that
assume that reliable communication can be provided by the
single uniform network.
Research approached in the field, like SANDMAN [13] and
DEAPSpace [12], is done by grouping nodes with similar
mobility patterns into clusters; in each cluster, one of the nodes
(called cluster head) stays awake permanently and answers
discovery requests. The rest of the nodes periodically wake up
to provide the actual services and also inform the cluster head
about their presence and services [1]. Other approaches take
into account the network partitioning and provide related
metrics and comparisons are briefly presented in the following
paragraphs.
A. Approach of Expanding MANET to a Large Network
There are currently no general approaches or methodologies
for the creation of domains that take into account the following
important engineering constraints: balanced domains, minimal
inter-domain traffic, and robust network design [3]. Most
existing work on domain generation, however, has used only
very limited local information. In fact, most approaches simply
elect a cluster-head within each subnet based on node
attributes like the lowest ID or highest degree [3][4]. Some
proposals use local metrics during cluster generation, but the
metrics are utilized just for the selection of cluster-heads [5],
[6].
Steffano Galli [3] viewed the network as a weighted
undirected graph G = (V, E), where each vertex v V
represents a network node and each edge eE represents a
viable (potential) communication link with an associated cost
we (the cost associated with every edge represents the
estimated link load). An example is shown in Figure 1. We
wish to partition graph G into N sub-graphs G
i
, i =1, 2, , N,
with the objective of satisfying a given set of constraints. It is
important to recognize that finding such sub-graphs, i.e.
creating network partitions, requires cutting a subset CE of
edges in G. The graph G in Figure 1 has been partitioned in N
= 4 sub-graphs G1, G2, G3, and G4. The set CE of cut edges
is shown in dotted lines.
The rationale for choosing the partition, i.e., for deciding
which edges to cut, is based on engineering rules given in the
Introduction which can be translated into a graph theoretic
framework as follows:
The partition is as balanced as possible: the number of
nodes in every sub-graph G
i
, i = 1, 2, , N, is the same
within the smallest possible tolerance.
The sum of the weights of all cut edges is minimized.
Each sub-graph G
i
, i = 1, 2, , N, is 1-connected.

Figure 1. Example of graph partitioning (four areas) of a network [3].
Based on the Galli proposed to solve the OSPF area design
problem for large networks using a two-step approach: a first
step consisting of an efficient Graph Partitioning Algorithm
that will create balanced areas (including a rough Area 0), and
a second step consisting of additional ad-hoc heuristics that are
able to ameliorate Area 0 design as well as factoring in OSPF
specific metrics.
B. OSPF MANET Extension
The Open Shortest Path First routing algorithm is widely
used for wired networks. As consequences of users
expectation to seamlessly coexist their mobile devices with

Vol. 2, No. 6, 2011
3 | P a g e
their wired counterparts, the modification of legacy protocol
must be designed. Due to a lack of central base station, nodes
in a MANET must form peering relationships to collectively
make routing decisions.
OSPF uses a link-state routing protocol to find the least-
cost path from a source router to a destination router within a
group of routers owned by an organization. As shown in Figure
2, a group of routers using the same routing protocol is
collectively referred to as an Autonomous system (AS). Upon
joining the AS, a router uses the Hello protocol to discover
neighboring routers. Then it forms adjacencies with its new
neighbors to exchange routing information; a process called as
Link-State Advertisements (LSAs). The exchange (flooding) of
routing information, allows routers to obtain complete view of
the network topology. Each router stored it as link-state
database (LSDB). Initially, LSDB represents the network
topology that is visible by router as a result of Dijkstra least-
cost path algorithm. In order to maintain and synchronize the
network convergence, each router periodically broadcast
updates about its current links state to their LSDBs.

Figure 2. Connected Autonomous systems (ASs)
Although OSPF was designed for wired networks where the
topology is predicable and the medium is relatively reliable, its
routing algorithm can be extended to support MANET. OSPF
is widely deployed and includes many innovative features. The
protocols specifications are in the public domain. It is used by
upper-tier Internet Service Providers to determine routes within
their networks. It supports multicast routing, multiple same-
cost paths, and the ability to organize a network as a hierarchy.
Nodes in a MANET randomly form peering relationships as
they move along its communication range. Moreover, packet
loss occurs frequently on a wireless medium due to path loss,
interferences, noise, shadowing, and multipath. Therefore,
OSPF needs to be modified to meet the routing requirements of
MANET, which include minimizing data exchange and control
overhead.
The OSPFv2 Wireless Interface Type was proposed to
support wireless networks. The enhanced protocol is not
limited to MANETs; it supports wireless, broadcast-capable,
multi-hop networks. A team of Boeing Company contributed
to the OSPFv2 Wireless Interface Type proposal [4]. Boeing
also made other contributions to the OSPF extension for
MANETs. They evaluated and analyzed the performance of
MANET extensions through simulation of OSPF MANET
Designated Routers (MDR) and Cisco's extension (Overlapping
Relay and Smart peering) [5][6].

Figure 3. The routers role in the OSPF protocol to reduces the packets
adjacencies.
Designated Routers (DR) algorithm provides a mechanism
to reduce packets adjacency for data exchanges. The Hello
protocol provides a mechanism for electing the network DR.
Besides its main function to find and maintain connections
among neighbors, Hello packets contain information about
router priority that is used to assign the DR. The first router
with the highest priority becomes the DR, and the router with
the second highest, the Backup Designated Router (BDR) [15].
The DR (router) is responsible to relay the packets to other
routers.
OSPF Multi-point Relaying (MPR) uses an MPR selector
set to reduce adjacencies. Unlike the OSPF-MDR extension,
OSPF-MPR makes not changes to Hello protocol. It minimizes
flooding overhead by choosing a flooding MPR set. Similar to
Cisco's extension, OSPF Multi-point Relaying (MPR) only
forms adjacencies with a limited subset of neighbors to
improve scalability and reduce control overhead (LSAs).
However, OSPF-MPR does not assign a MDR. A router forms
adjacencies with a subset of its one-hop neighbors called MPR
selector set.
Most routers only form adjacencies with its MPR selector
set, but some Synch routers also form adjacencies with
neighbors not in their selector set. A router becomes a synch
router only if it has highest router ID among its known
neighbors. Hello packets and LSA contain router ID of its
neighbors. At least one Synch router must exist in the MANET.
However, extra data is added in Hello packets in to facilitate
MPR selection.
III. SIMULATION MODEL
As foundation for this mobile environment, the core
algorithm is developed from static mode (e.g., sensor
networks). The enhancement algorithm for serving mobility
then detailed in support of topology development, topology
maintenance, and routing maintenance.
The model is initiated from broadcast mechanism and
propagated through node-to-node based routing metrics
approach. Each source injects single big packet which
fragmented into multiple packets in the network, which
traverse through the network until those reach the final
destination. Packets are queued at each node in its path where it
waits for an opportunity to be transmitted. This model is not
only applicable in direct communication (one hop
transmission) but it can also work in multi-hop transmission.

Vol. 2, No. 6, 2011
4 | P a g e
In this situation, when the source and final destination
nodes are located outside its clustered network area, source
node is capable to discover multiple hop route lead to agents
thus maintaining the connectivity required in comparison to
standard flooding based ad hoc routing designs.
It is square of Cartesian model area with 200x200 areas and
one overlapped section. We consider the case where all nodes
in the network are similar, i.e., assuming a homogeneous
infrastructure. Inside areas, nodes are deployed uniformly,
distributed at random position in the both areas. This
deployment produces a connected topology under some
assumptions; sometimes a completely connected topology is
built and sometimes topology is not fully connected.
Simulation build a large connected component quickly using a
communications radius considerably smaller than the radius
needed to have the entire network connected. Agents are
located in such place to facilitate communication between
wireless areas and minimize the number of hops to achieve
optimum throughput of mesh clients which communicate each
other. Agents have static position located in the overlapped
between two network areas.

(a)

(b)
Figure 4. Development of network topology within two areas. (a) With
Agents involved in delivery of packets. (b) Without Agents involved in
delivery of packets.
This mode of messenger can be made clear from the Figure
4. At Figure 4(a), if the source (node 4 mark with red color)
coming from certain clustered network area and having
destinations (e.g. node 33) located at different (adjacent)
clustered network area, the paths are able to meet at a common
Agents (node 1000 or node 1001) simultaneously then only any
one of the agents will be sufficient to carry all the messages to
the proper destination. Thus our routing scheme utilizes the
agents capability where a collection of independent request
transmission come together for the purpose of cooperative task
behaviors and maintaining these connectivity among pairs of
sources - destinations. All these tasks essentially work directly
through node - agent communication. The entire algorithm
works on the fact that agents need to know the existence of
each nodes at each clustered areas. On the other side, at Figure
4(b), predefined source (node 4) start the topology
development protocol by sending (broadcast) an initial Hello
Message. With receive-transmit subsequent routine, the process
continues to all reachable nodes. Not every node will be
selected to be part of the tree, and those which were not
selected will keep silent (in the propagation of packet). Without
any knowledge of destination topology (which is located in
other network area), then the packets must travel to each vertex
to reach the destination node. In both pictures, if there are more
than one source nodes starts to transmit packets
simultaneously, then several trees may be built in parallel.
A. The Model
Simulation consists of multi clustered network environment
of homogeneous nodes that communicate with each other using
the broadcast services of IEEE 802.11. There are nodes with
different roles simulated in this simulation, namely initiator
node/source node, receiver node, sender node, destination
node, and final destination node. Initiator node/source node is
node that initiates transmission of packet. Packet can be either
route discovery or data transmission. Like other nodes, initiator
is always moving with random direction, speed, and distance.
At the time it is moving, initiator node is always sensing its
neighbor to maintain connectivity. Receiver node is node that
can be reached by source/sender node. Nodes are defined as
neighbors if it located within its distance radius range. At initial
time, node senses its neighbors before packet data is required to
be transmitted. Coverage neighbor nodes always receive
packets that are broadcasted from sender. Destination node is
selected receiver node in multi hop transmission that should
relay packets to the next receiver node. Final destination node
is node that became the finish destination of packets.
The layered concept of networking was developed to
accommodate changes in local layer protocol mechanism. Each
layer is responsible for a different function of the network. It
will pass information up and down to the next subsequent layer
as data is processed. Among the seven layers in the OSI
reference model, the link layer, network layer, and transport
layer are 3 main layers of network. The framework is
configured in those layers. Genuine packets are initiated at
Protocol layer, and then delivered sequentially to next layer as
assumed that fragmented packets to be randomly distributed.
Simulation models each layer owned with finite buffers.
Limited buffer makes packets are queued up according to the
drop tail queuing principle. When a node has packets to
transmit, they are queued up provide the queue contains less
than K elements (K 1). To increase the randomization of the
simulation process, simulation introduces some delay on some
common processes in the network, like message transmission
delay, processing delay, time out, etc. This behavior will result
that at each instance of a simulation would produce different
results. The packets exchanged between sender and receiver is
of a fixed rate transmission based on a Poisson distribution.
Nodes that have packet queued are able to transmit it out using
in each available bi-directional link channel.
Vol. 2, No. 6, 2011
5 | P a g e
Our work extends the chance of contacting of the agents
that arranged at the fixed cluster from nodes which are
distributed randomly within the network. Agents navigating
through the network for delivering messages must understand
these clustered nodes whenever they are initiating a new route
request and thereby increasing the degree of spatial
coordination (agents must be on the same place). The temporal
coordination has been enhanced with the introduction of a short
waiting delay offered to each nodes/agents by the clustered
overheads packets. This node agent coordination will reduce
the number of hops and waiting time in spite of further increase
the overheads packets hang around with agents and head to
highly the agent-chasing problem. The place hosted for the
clustered agents can be called as the overlapped area within the
network and the detainment period by the agents can be called
agent periods.
This overlapped area actually offers a temporary space to
be used by all agents for sharing network knowledge, and
exchanging messages. Thus when an agent come a fresh it can
exploit all other agents who are currently experiencing their
agent periods. Here the traveling agents are allowed to carry
the information of already visited clusters along with them. The
idea behind this is to capture and share the partial network
information present with roaming agents. The integration of all
such partial information at a common overlapped area helps
cooperative tasks like taking the decision for next destination,
suitable exchange of messages between agents, getting up-to-
date knowledge of the network and reducing unnecessary
redundant overhead packets used to visit nodes.
Thus the model of coordination clustered network area
where the autonomous agents will be able to deliver messages
within a large network with the cooperative communication
between them at suitable overlapped area is necessary. There is
need of knowing the routes proactively or reactively where part
of the network capacity is used for exchanging chunks of
routing table data.
We built network simulator to evaluate this proposed
algorithm. The simulator supports physical, link and routing
layers for single/multi hop ad-hoc networks. We assume that
IEEE 802.11 Distributed Coordination Function (DCF) or
MAC protocol which uses Channel Sense Multiple Access with
Collision Avoidance (CSMA/CA) already deployed.
Successfully received packet by receivers interface is packet
whose SNR is above a certain minimum value otherwise the
packets cannot be distinguished from background
noise/interference. Packets are transmitting through physical
layer in accordance with Poisson distribution. Communication
between two nodes in IEEE 802.11 uses TCP signaling before
the actual data transmission takes place. Simulation simulates
this with random hearing to links condition. If link allow
packets to be sent, then sender executes some packets already
queued. To execute preferential event in sequentially
distributed events, we used a simple approach that consists of
applying a different time-event execution by means of the
triggering event sequences action. The lower and upper bound
of the queuing interval are set such that they do not interfere
with predefined timers used by the other events for layers and
modification events.
B. Nodes
Nodes are equipped with antenna module installed as
capable of dynamically adjusting the transmission energy used
to communicate with other nodes. Industrial standard of
antenna module supports a management for controlling this
energy consumption. The energy consumption required to
transmit packet between nodes A and B is similar to that
energy required between nodes B and A if and only if the
distance and the size of packet are same. The coverage distance
range of the nodes is a perfect symmetric unit disk (omni-
directional). If d
x,y
r
x
x and y can see each other. This
assumption may be acceptable in the condition that interference
in both directions is similar in space and time; which is not
always the case. Usually interference-free Media Access
Control (MAC) protocol such as Channel Sense Multiple
Access (CSMA) may exist. In addition, wireless link channel is
assumed to have no physical noise; i.e., the errors in packet
reception due to fading and other external interferences are not
considered as a serious problem. Packets from sender to
receiver will be transmitted as long as the bandwidth capacity
is sufficient and the received signal to noise ratio (SNR) is
above a certain minimum value. Thus every packet
successfully received is acknowledged at the link layer and de-
encapsulate at the higher layer. Each node is capable of
measuring the received SNR by analyzing overheard packet. A
constant bit error rate (BER) is defined for the whole network.
Whenever a packet is going to be sent, a random number is
generated and compared to the packets CRC. If the random
number is greater, the message is received, otherwise it is lost.
The default value for the BER is 0, which means there is no
packet loss due to physical link error.
Energy is power kept in each node. Heinzelman et al.
assumed that the radio dissipates E
elec
= 50 nJ/bit to run the
transmitter or receiver circuitry and
amp
= 100 pJ/bit/m
2
for the
transmit amplifier [7]. The radio model is shown in the Figure
5 below.

Figure 5. The radio model.
Thus, to transmit a k-bit message a distance d using this
radio model, the radio expends:
( )
()
( )
( )

(1)
and to receive this message, the radio expends:
()
()
()

(2)
Let E
min_i
is the minimum energy ratio of node I at which a
node can still receive, process, and transmit packets. Node j
finds out the energy level of neighbor node I through analyzing
of received reply packet from node I as it responded the
previous transmitted Hello packets. The computation of E
min_i

is done through two-step propagations. The use of two-steps

Vol. 2, No. 6, 2011
6 | P a g e
propagation model is to simulate interactive propagation in the
operation of the protocol in dynamic environment. As a future
research, the appropriate propagation model that best matches
to this environment should replace the simple two-steps model
presented here [9][10]. The two-steps propagation model is
appropriate for outdoor environments where a line of sight
communication existed between the transmitter and receiver
nodes and when the antennas are omni-directional. The two-
steps propagation model assumes there are two main signal
components. The first component is the signal traveling on the
line of sight to reached neighbors along with its reply from
neighbors and the second component is a confirmation packet
transmitted to selected neighbors.
C. Agents
In this paper we introduce mobile agents to hop around the
networks for delivering packets .The agents are allowed to
meet with other agents at fixed overlapped places as has been
mentioned. The cooperation among agents will mutually
benefit each other by cooperating in delivering packets. In this
current flexible and decentralized framework any autonomous
node can send message to any other node at any instant within
the network by just issuing a mobile agent. The agent then
communicates with other agents to determine the proper one to
carries the message to the corresponding destination node. The
selected agent then becomes responsible for delivering the
message to proper destination. Analog to the real life, these
agents actually play the role of messengers and the selected
agent play the role of post offices in the ad hoc wireless
scenario. Such cooperation among agents scheme has been
explicitly designed to reduce the agent traffic in the network.
The unnecessary redundant node visits made by the agents to
reach destination node has been avoided by sharing and
merging with other agents. These agents take the responsibility
of providing communication services and improvement of
overall traffic in the network.
While delivering messages, an agent will maintain the path
records of all the visited nodes in both clustered networks and
its corresponding topology. Carrying this network information
provide coordination and share the updated network knowledge
with other agents. This information field carries topology tree
along with numeric values of this membership list (of nodes)
collected from each clustered network. In few cases the length
of the information list carried by an agent gets longer due to the
course of journey made from agent to reach the proper
destination. If this happened, the list has been restricted using
the hop count limit in order to avoid a huge series of data to be
carried along by an agent and subsequent nodes. No packet is
allowed to travel forward further whenever its hop count has
got exhausted and it is compelled to move back to its
originating source node/agent. Thus whenever an agent has
finished its forward journey it will eventually follow the same
path back to the source node.
The objective of the navigation procedure is to minimize
the hops between the agent at overlapped location (current
location of the node where the agent is residing) and the source
and destination nodes location. This criterion would enable an
agent to select a neighbor of its current location and take out
the packets to the destination nodes. If there is no neighbor
available at that instant of time satisfying the above-mentioned
criterion, the agent waits for a pre specified amount of time
(randomly) and tries to communicate with other agents (any
agents can be reached) to get its knowledge. Such contacted
agents will respond the request. Through intensive
communication among agents, the best agent can be selected to
take responsibility for delivering received packets.
In the simulation, when a source node within the ad hoc
network wants to send some packets, it immediately senses
whether an agent is needed. Each such agent attaches with
itself a topology bag to accommodate the request (certain)
destination node. This bag is of a given capacity, which can be
made full or can be made empty. The source node after
initiating the agent puts the packets to (appropriate) agent. The
agents are able to exchange these packets with other agents on
having suitable coordination with them, which will have the
best route path with minimum hops to reach destination node.
These agents will be inactive automatically when there is no
more packets need to be delivered.
Because of high degree of mobility, the topology will
change and it is assumed that the agent will eventually succeed
to migrate [2][11]. Whenever an agent wants to leave its
current location for delivering packets to some unknown
networks it will collect the topology list information from the
nodes and will try to reach for a boundary node through which
it may get an exit point. The node lying at the boundary will
have neighbors from two or more different agents and can act
as gateway nodes to other clustered network area. Thus if an
agent can reach such a cluster area boundary, it can start
visiting a fresh (other) agents. As the location of the agents can
be made available from any node of that same clustered
network area it can easily track the new nodes there, which has
been compulsory. Though the order of cluster visits take place
in a random manner still the redundancy in the path visit has
been avoided by maintain the path visit list (using BFS
function). The agents are free to roam among the overlapped
area within the network in a random manner. This agents
capability will be presented on the next paper.
IV. PERFORMANCE EVALUATION
In this section, we evaluate the performance of the routing
scheme for packets delivery within clustered ad hoc wireless
network areas. We first describe our simulation
implementation, performance metrics and methodology, and
then we evaluate the agents initiated packets delivery scheme.
The results confirm that the agents and their coordination
routing algorithm at overlapped area are very efficient in
delivering packets under high traffic. The result will also show
that living agents in the system as there are packets spread
around the clustered network areas; through suitable exchange
and message sharing then the number of successful traffics get
considerably increased with time.
We started to consider network composed of 60 nodes
distributed into two area domain and 2 agents that are located
within overlapped area between network areas. The network
was set 400x200 areas, with overlapped region width of 20.
Fig.6 showed this view of simulation. The movement of the
nodes is made random. Nodes are allowed to move at average
speeds of 20m/sec. Simulation is run for 100 cycles with
Vol. 2, No. 6, 2011
7 | P a g e
movements parameter was set random (i.e. direction, distance,
and speed). Nodes are restricted to move only inside in its area.
The following simulation with nodes movement to other
network area, including overlap area will be presented. In the
subsequent simulations, the number of nodes is raised up along
with combinations of different agents and various width of
overlapped area.

Figure 6. Initial network topology.
The performance of the agents capability is evaluated
using the following two criteria: i) Number of packets whose
data contained in the propagation mechanism i.e. the total
amount of packet traffic destined at agent to relay packets
issued by the nodes in the network. ii) The time period selected
as RTT period for successful packets delivery through network;
both between source agent and agent - destination. Network
evaluation is intended to investigate the successful transmission
rate in the hops between source and final destination nodes and
then followed by several network performance analysis that are
marked as scenario 1, scenario 2, scenario 3, and scenario 4. At
each of scenario, network with similar initiator nodes are
loaded with different packets length. We analyze these
variables to determine their effects on the minimizing number
of hops and average end-to-end delay for packets delivery. The
simulation computes different hops/relay connectivity for the
packets flow in the delivering packets from source to final
destination.
As showed at Fig. 7(a), MANET tends to discover more
successful data packets transmission in densely mobile
environment (due to moving direction that getting closer to
agents). It has been described in section III(b) and III(c), the
network topology structure actually consists of two sub
protocols, i.e. node topology protocol and agent topology
protocol. In the case that many nodes located adjacently each
other, thus every node periodically broadcasts Hello packets to
denote its presence. Agents do the same sequences to adapt
domain environments and to update network topologies for
both of proximity areas. This extensive broadcast leads to
unavoidable redundant packets for nodes located near
overlapped zone. This is depicted in Fig.7(c) where high traffic
is identified in the subsequent cycles. These agents, however,
are intended to avoid global flooding and save network
resources, but in such discovered case, agents are suffered in
servicing network.

(a)

(b)

(c)
Figure 7. Simulation with nodes movement direction staticly closer to
Agents for network with 2 agents and 60 nodes in the two network areas. (a)
Network topology at final cycle. (b) Accumulated hops and distances of
successful data transmission for different cycle. (c) Broadcasted packets
during simulation.
In the highly mobile environments (due to node mobility),
more long-successful data packets delivery can be discovered.
This is explained by the fact that when the nodes are highly
mobile, paths are difficult to be maintained and hence far-away
transmission tend to last for a very short amount of time since
the probability for a path break is larger when nodes move
faster.

198
228
218
305
320
369
348
402
459
11
1010
9 9 9 9
8 8
9
8 8
7 7 7
6 6 6 6
5 5
4 4 4 4
3 3 3 3
2 2
0
2
4
6
8
10
12
0
100
200
300
400
500
600
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
#Hops
Accumulated
Distance
Cycle
Dist_Hop 1
Dist_Hop 2
Dist_Hop 3
Dist_Hop 4
Dist_Hop 5
Dist_Hop 6
Dist_Hop 7
Dist_Hop 8
Dist_Hop 9
Dist_Hop 10
Dist_Hop 11
#Hop
4
5
1
7
2
6
.
9
2
5
6
3
2
.
6
1
1
7
6
.
5
1
6
2
7
.
5
2
1
6
1
.
9
5
2
7
4
2
.
6
7
5
3
1
3
6
.
9
5
3
8
1
9
.
2
5
4
2
7
6
.
7
5
3
9
7
6 4
6
2
3
.
5
2
5
3
9
5
4
.
3
5
1
1
7
5
5
9
8
.
5
2
5
6
7
3
5
.
4
5
1
4
7
.
0
5
6
5
5
3
.
0
5
5
2
6
0
.
7
5
5
6
0
3
.
2
7
5
4
7
7
8
.
0
2
5
7
0
1
0
.
6
5
8
9
2
3
.
6
2
5
7
8
6
4
.
2
2
5
8
2
6
8
.
0
2
5
8
7
4
6
.
0
7
5
8
6
9
2
.
7
7
5
7
9
2
8
.
8
5 9
1
2
5
.
1
5
1
1
2
9
5
.
7
7
5
9
0
8
8
.
3
2
5
7
7
3
1
.
4
7
5
6
9
0
3
.
7
5
5
7
6
1
.
2
5
9
6
6
.
4
7
5
6
8
5
1
.
7
7
2
2
1
.
5
1
5
9
8
.
1
7
5
1
9
5
0
.
1
7
5
2
1
7
.
4
4
2
0
2
.
3
4
2
1
3
.
5
7
5
1
7
0
.
5
2
0
500
1000
1500
2000
2500
3000
3500
0
2000
4000
6000
8000
10000
12000
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39
#
P
a
c
k
e
t
s
T
h
o
u
s
a
n
d
s
#
L
e
n
g
h
t
o
f

p
a
c
k
e
t
s
T
h
o
u
s
a
n
d
s
Cycle
#lenght_packets_overhead[succ] #packets_overheads[succ]
#packets_DATA[succ] #lenght_packets_DATA[succ]
Vol. 2, No. 6, 2011
8 | P a g e
When nodes move slower these paths tend to be more
stable and hence services tend to be available for a longer time.
Such of this occurrence is explained in the Fig. 8.

Figure 8. Hops of successful data transmission for different speed of
simulation with movement direction closer to Agents for network with 2
agents and 60 nodes in the two network areas.
It is obvious from Fig. 7 that maximum successful data
packets transmission can be achieved at low mobility when
nodes are in overlapped area. The values of average successful
data transmission over various mobility speeds are presented in
Fig. 8. It is evident from this figure that the average successful
data transmission actually decreases when speed increases and
its direction are far from agents.
In order to evaluate the movement direction to network
traffic, we put aware of the average duration of successful data
packet transmission between a node and final destination
through various movement directions. As previously predicted,
the topology protocol would perform better in low mobility
speed and direction of approaching agents. This is explained by
the fact that the data transmission services discovered in that
setting would be adequate for nodes to complete their
transaction. The inverse would be true for low average
successful data packet transmission duration, where a setting
consists of high mobility speed and direction of away from
agents is used for the topology protocol. Such situation can be
concluded from Fig. 7, Fig. 8, and Fig. 9.

(a)

(b)

(c)
Figure 9. (a) Network topology at final cycle. (b) Accumulated hops and
distances of successful data transmission for different cycle of simulation with
movement direction away to Agents for network with 2 agents and 60 nodes
in the two network areas. (c) Broadcasted packets during simulation.
Simulation was further modified in order to evaluate
agents service information in routing every relayed packet.
Each node that is broadcasting this topology update packets
sets the TTL (Time to Live) field in these packets equal so that
they will be dropped at agents at overlapped area. Every node
whose instance location is in overlapped area is set to as agents.
Agents listen to information gathered from topology update
packets, modify its table and then periodically broadcast its
topology table back to nodes in the network. In the situation,
where more nodes are located in the overlapped area, overhead
traffic among agents leads to significantly higher overhead
packets interaction (approximately 125% in average) and at the
same time it fail to manage the transmission as no other nodes
are discovered nor almost the nodes cannot reach agents. The
following figures depict the behavior of network at different
movement directions which is grouped as the final situation of
network topology (in the (a)), the distance of successful data
packets transmission at each hop (in the (b)), and the correlated
broadcasted packets during simulation.
11
10 10
9 9 9 9
8 8
9
8 8
7 7 7
6 6 6 6
5 5
4 4 4 4
3 3 3 3
2 2
0
2
4
6
8
10
12
14
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
Hops
Cycle
#Hop[speed 20kmh] #Hop[speed 60kmh]
66
75
85
2 2 2
0
0.5
1
1.5
2
2.5
0
20
40
60
80
100
120
1 2 3
#Hops
Accumulated
Distance
Cycle
Dist_Hop 1
Dist_Hop 2
#Hop
77.903
56.182
20.374
23.064
-5
5
15
25
35
45
55
65
75
85
0
50
100
150
200
250
300
350
400
1 2 3 4
#
P
a
c
k
e
t
s
T
h
o
u
s
a
n
d
s
#
L
e
n
g
h
t
o
f

p
a
c
k
e
t
s
T
h
o
u
s
a
n
d
s
Cycle
#lenght_packets_overhead[succ] #lenght_packets_DATA[succ]
#packets_overheads[succ] #packets_DATA[succ]
Vol. 2, No. 6, 2011
9 | P a g e

(a)

(b)

(c)
Figure 10. Simulation with nodes movement direction staticly to East for
network with 2 agents and 60 nodes in the two network areas. (a) Network
topology at final cycle. (b) Accumulated hops and distances of successful data
transmission for different cycle. (c) Broadcasted packets during simulation.

(a)

(b)

(c)
Figure 11. Simulation with nodes movement direction staticly to West for
440 439
462
12
14
12
-1
1
3
5
7
9
11
13
15
0
100
200
300
400
500
600
1 2 3
#Hops
Accumulated
Distance
Cycle
Dist_Hop 1
Dist_Hop 2
Dist_Hop 3
Dist_Hop 4
Dist_Hop 5
Dist_Hop 6
Dist_Hop 7
Dist_Hop 8
Dist_Hop 9
Dist_Hop 10
Dist_Hop 11
Dist_Hop 12
Dist_Hop 13
Dist_Hop 14
#Hop
2
6
4
.
9
2
5
3
.
3
5
2
9
7
.
5
2
5
3
1
6
.
7
3
5
0
.
5
2
5
3
6
8
.
2
7
5
4
3
6
.
0
5
4
1
0
.
3
7
5
4
1
5
.
6
5
3
7
2
.
2
2
5
4
1
8
.
2
5
4
1
6
.
4
7
5
5
0
7
.
6
5
5
8
.
9
7
5
3
4
5
.
9
7
5
3
5
8
.
3
7
5
3
8
3
.
8
5
4
2
1
.
8
2
5
4
2
6
.
5
5
4
3
7
.
9
4
8
1
.
3
5
6
5
9
.
0
2
5
7
6
1
.
1
5
7
5
2
.
5
5
5
1
0
.
2
5
5
0
7
.
7
7
5
5
2
9
.
8
7
2
2
.
8
9
1
4
.
3
8
4
4
.
2
5
8
6
6
.
4
5
9
3
7
.
9
5
9
3
5
.
1
5
9
7
4
.
8
5
1
0
4
1
.
8
5
9
6
1
.
9
8
7
7
.
9
7
5
2
7
7
.
0
5
2
7
7
.
0
5
2
7
7
.
0
5
1
6
6
.
1
1
8
4
.
2
2
1
5
4
.
0
2
0 0
0
100
200
300
400
500
600
700
800
900
1000
0
200
400
600
800
1000
1200
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39
#
P
a
c
k
e
t
s
T
h
o
u
s
a
n
d
s
#
L
e
n
g
h
t
o
f

p
a
c
k
e
t
s
T
h
o
u
s
a
n
d
s
Cycle
393 392
418
421
11
12
11 11
-1
1
3
5
7
9
11
13
15
0
100
200
300
400
500
600
1 2 3 4
#Hops
Accumulated
Distance
Cycle
Dist_Hop 1
Dist_Hop 2
Dist_Hop 3
Dist_Hop 4
Dist_Hop 5
Dist_Hop 6
Dist_Hop 7
Dist_Hop 8
Dist_Hop 9
Dist_Hop 10
Dist_Hop 11
Dist_Hop 12
#Hop
3
5
6
.
8
2
5
4
2
3
.
3
2
5
4
7
5
.
5
5
1
6
.
9
5
8
2
.
7
7
5
5
3
2
.
1
6
4
5
.
1
7
5
5
2
3
.
4
5
8
5
.
5
7
5
6
2
1
.
2
5
7
5
2
.
6
2
5
8
5
5
.
5
5
9
0
1
.
0
7
5
9
2
6
.
3
5
8
9
8
.
6
6
3
4
.
9
7
1
7
.
7
2
5
1
0
3
9
.
9
5 1
3
0
2
.
6
1
3
4
3
.
5
5
1
2
8
5
.
4
5 1
5
6
3
.
1
2
5
1
6
5
5
.
7
1
9
3
0
.
6
5
2
1
3
8
.
3
5
2
0
4
0
.
6
2
5
1
8
4
8
.
1
5 2
0
9
9
.
0
5
2
2
1
4
.
8
2
5
9
4
5
.
4
7
5
4
5
1
.
8
5
2
5
.
7
4
6
6
6
3
1
.
4
2
5
7
4
8
.
8
7
8
9
.
9
7
5
5
7
9
.
8
5
5
7
9
.
8
5
6
7
1
.
1
1
7
2
.
1
4
1
6
9
.
1
2
1
6
6
.
1
1
4
7
.
9
8
0
100
200
300
400
500
600
700
800
900
1000
0
500
1000
1500
2000
2500
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39
#
P
a
c
k
e
t
s
T
h
o
u
s
a
n
d
s
#
L
e
n
g
h
t
o
f

p
a
c
k
e
t
s
T
h
o
u
s
a
n
d
s
Cycle
Vol. 2, No. 6, 2011
10 | P a g e
In the Fig. 10(c) and 11(c), with the network movement
direction is set to west and east, the broadcasted packets are
significantly high at around cycle 28 to 30. During those
cycles, almost nodes in one area are located in the overlapped
area and acted as agents. Communication among (more) agents
and (less) nodes lead to flood the network with overhead
packets. After such situation, the traffic tends to decline where
nodes and agents are come to a stop at border and overlapped
areas.

(a)

(b)

(c)
Figure 12. Simulation with nodes movement direction staticly to South for
Subsequently, the scenario which every node has speed and
direction that is determined of south and north is given below
at Fig. 12(c) and the following Fig. 13(c). In all the two
simulation models we did this simulation for 50 cycles with 60
nodes, as used previously for Fig. 10 and Fig. 11. Readings
were taken for mobility of (speed) 20 km/hour. From the
results it is evident that as the nodes coming near the border
area increase; the traffic of both figures decline. Low packets
nature suggests that more nodes stops looking for agents when
no route path founded to reach them in the overlapped area. We
simulated static agents in all figures.

(a)

(b)

(c)
Figure 13. Simulation with nodes movement direction staticly to North for
451
417
448 446
434 436
427
452
437
506
499
487
476
496
484
491
12 12 12 12
14 14
13 13
14 14 14
12 12
11 11 11 11
-1
1
3
5
7
9
11
13
15
0
100
200
300
400
500
600
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
#Hops
Accumulated
Distance
Cycle
Dist_Hop 1
Dist_Hop 2
Dist_Hop 3
Dist_Hop 4
Dist_Hop 5
Dist_Hop 6
Dist_Hop 7
Dist_Hop 8
Dist_Hop 9
Dist_Hop 10
Dist_Hop 11
Dist_Hop 12
Dist_Hop 13
Dist_Hop 14
#Hop
3
0
6
.
0
7
5
2
3
8
.
6
2
3
5
.
7
3
1
0
.
0
5
3
1
4
.
2
5
3
0
1
.
9
2
3
1
.
8
7
5
2
1
7
.
8
5
2
1
9
.
5
5
2
0
3
.
8
2
5
2
0
4
.
3
7
5
1
8
2
.
3
5
1
8
4
.
3
2
5
1
7
3
.
3
7
5
1
5
5
1
0
9
.
6
5
1
0
4
.
4
7
5
6
2
.
1
2
5
6
2
.
1
2
5
5
9
.
0
7
5
2
7
.
1
2
5
2
6
.
8
2
5
2
6
.
8
2
5 5
5
.
1
2
1
.
3
5
2
3
.
9
5
1
5
7
.
0
4
1
6
0
.
0
6
1
5
7
.
0
4
1
8
2
.
8
2
5
2
1
7
.
4
4
2
1
7
.
4
4
2
0
2
.
3
4
1
9
6
.
3
2
1
1
.
4
1
9
6
.
3
1
9
9
.
3
2
1
7
2
.
1
4
1
7
2
.
1
4
1
7
6
.
6
1
1
6
4
.
4
3
1
7
5
.
1
6
1
7
5
.
1
6
0
0
10
20
30
40
50
60
70
80
0
50
100
150
200
250
300
350
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
#
P
a
c
k
e
t
s
T
h
o
u
s
a
n
d
s
#
L
e
n
g
h
t
o
f

p
a
c
k
e
t
s
T
h
o
u
s
a
n
d
s
Cycle
435
444
453 454
474
498 495
517
12 12 12 12
13
12
14
13
-1
1
3
5
7
9
11
13
15
0
100
200
300
400
500
600
1 2 3 4 5 6 7 8
#Hops
Accumulated
Distance
Cycle
Dist_Hop 1
Dist_Hop 2
Dist_Hop 3
Dist_Hop 4
Dist_Hop 5
Dist_Hop 6
Dist_Hop 7
Dist_Hop 8
Dist_Hop 9
Dist_Hop 10
Dist_Hop 11
Dist_Hop 12
Dist_Hop 13
Dist_Hop 14
#Hop
3
1
1
.
4
5
3
9
8
.
9
7
5
3
1
8
.
3
5
3
6
3
.
7
5
3
7
1
.
3
7
5
3
2
1
.
5
3
1
2
.
0
7
5
3
7
9
.
7
7
5
2
8
2
.
3
5
2
8
2
.
7
5
2
0
7
.
8
1
7
7
.
5
2
5
1
7
4
.
7
7
5
1
8
4
.
7
5
1
5
0
.
7
2
5
9
1
.
3
8
4
.
7
5
7
.
0
7
5
5
3
.
3
5
3
.
2
7
5
5
7
.
0
5
5
3
.
7
2
5
4
1
.
6
3
1
.
5
5
3
1
.
6
0
.
3
5
2
2
8
.
9
2
5
2
2
3
.
7
6
2
1
0
.
0
5
5
2
1
7
.
1
7
2
7
0
.
0
1
2
4
0
.
5
3
2
7
4
.
0
2
5
2
9
1
.
8
6
5
0 0 0 0 0 0 0 0 0 0
0
10
20
30
40
50
60
70
80
90
100
0
50
100
150
200
250
300
350
400
450
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
#
P
a
c
k
e
t
s
T
h
o
u
s
a
n
d
s
#
L
e
n
g
h
t
o
f

p
a
c
k
e
t
s
T
h
o
u
s
a
n
d
s
Cycle
Vol. 2, No. 6, 2011
11 | P a g e
Related to density, we simulated two scenarios. The first
scenario included 120 nodes that are moving on two areas of
total size 400x400 meters, following the random waypoint
model with average speed 20m/s (minimum speed is still
0m/s). The second scenario was identical to the previous one
but involved only 60 nodes. Both scenarios had duration of 50
step movements each. The results are confirmed with previous
scenarios. Different network density showed the same pattern
of data packets transmission duration distributions, but the
number of successful transmission in the sparsely scenarios is
lower on the average of the others found in the first scenario.
This is due to the fact that re-discoveries network topology is
more frequent in a dense environment. Other fact is also
obtained that overhead routing packets are marginally
increased in size in order to encapsulate transmission
information when density is increasing. The size of overhead
packets plays a significant role under high-density cases where
congestion is present. Hence, an increase in node density would
lead to an increase in overhead packets and such as loss of data
packets transmission due to congestion would be discovered.
V. CONCLUSION
In this paper, we evaluate the successful data packets
delivery during packets transmission of MANETs with multi-
hop routing. With network whose topology always changed all
the time, then successful of data packets transmission is mainly
depend on the situation of nodes to others. In the densely
populated network, this is may not be the issues. On the
contrary, if nodes stand apart from other and with their moving
directions as such they cannot reach others, then no packets can
be transmitted. In the simulation, agents have cascading effects.
With their inherent advance capabilities, e.g. relay received
packets, to examine the network topologies, determine the next
destination of received packet, and guide the route path of
packets transmission to reach the destination, agents are mainly
to achieve optimum transmission while maintaining
connectivity among areas. They can reduce the route of large
number of unnecessary packets, particularly in the transmission
of packets destined to other area. Except these excellent
capability, communication among agents must be take into
consideration. Nodes movement direction, as key connectivity
of nodes to form MANET topology, is evaluated to show the
importance of agents. In the dynamic mobile environment,
since there are more alternative paths for agents through which
node can reach agents and also more alternative agents
available in the overlapped area, hence a failure of one or more
paths doesnt necessarily mean that the node cannot access the
agents in order to deliver packets to reach final destination.
Simulation results presented in section IV show that this is not
always true, particularly, when nodes are moving away from
agents. In such situation, the density increases, despite the
existence of multiple paths and agents, the average successful
data packets transmission is decreased. This is explained by the
fact that pair of source and final destination nodes failed to
maintain its connectivity as its distance getting out of agents
proximity. On the contrary, more nodes whose moving
direction is getting closer to agents will create more contention
for accessing the channel and transmitting service
advertisements. Hence, more packet collisions occur and large
energy is consumed. The total number of successful data
packets transmission however is higher in dense environment
with this movement direction. This means that high density
may increase the number of successful data packets delivery
but it decreases their quality in terms of availability. In nearly
future work, we plan to add agents capabilities to cover the
nodes which move through each network areas. Another further
investigation need to be conducted on dynamic routing
advantages and factors which affect routing mode, e.g., flow
type, delay, throughput/delay/reliability tradeoffs between
wireless network areas, etc.
ACKNOWLEDGMENT
The authors would like to thank the anonymous reviewers
for the helpful comments and suggestions.
REFERENCES
[1] Christopher N. Ververidis and George C. Polyzos, Impact of Node
Mobility and Network Density on Service Availability in MANETs,
Proc. 5th IST Mobile Summit, Dresden, German, June 2005
[2] Romit RoyChoudhury, Krishna Paul, and S. Bandyopadhyay, MARP:
A Multi-Agent Routing Protocol for Mobile Wireless Ad Hoc
Networks Autonomous Agents and Multi-Agent Systems (Kluwer), 8
(1): 47-68.
[3] Galli et al., A Novel Approach to OSPF-Area Design for Large
Wireless Ad-Hoc Networks, IEEE ICC05, Seoul, Korea, May 16-20,
2005
[4] Thomas R.H., Phillip A. S., Guangyu P., Evaluation of OSPF MANET
Extensions, Boeing Technical Report: D950-10897-1, The Boeing
Company, Seattle, July 21, 2005
[5] Henderson, T., et al., Evaluation of OSPF MANETExtensions, Boeing
Technical Report D950-10915-1, March 3, 2006,
[6] Spagnolo, P., Comparison of Proposed OSPF MANETExtensions,
The Boeing Company, October 23, 2006
[7] Heinzelman, W., Chandrakasan, A., Balakrishnan, H. Energy-efficient
communication protocol for wireless microsensor networks. In:
Proceedings of the 33rd International Conference on System Sciences
(HICSS): 110, 2000.
[8] M. S. Corson, S. Batsell, and J. Macker, Architecture consideration for
mobile mesh networking, Proceedings of the IEEE Military
Communications Conference (MILCOM), vol. 1, pp. 225-229, 21-24
Oct. 1996.
[9] Chang-Woo Ahn, Sang-Hwa Chung, Tae-Hun Kim, and Su-Young
Kang. A Node-Disjoint Multipath Routing Protocol Based on AODV in
Mobile Adhoc Networks. Proceeding of Seventh International
Conference of Information Technology ITNG2010, pp. 828-833, April
2010.
[10] Zheniqiang Ye, Strikanth V. Krishnamurthy and Satish K. Tripathi, A
Framework for Reliable Routing in Mobile Ad HocNetworks, IEEE
INFOCOM, 2003.
[11] Romit RoyChouldhury, S. Bandyopadhyay and Krishna Paul, A Mobile
Agent Based Mechanism to Discover Geographical Positions of Nodes
in Ad Hoc Wireless Networks, Accepted in the 6th Asia-Pacific
Conference on Communications (APCC2000) , Seoul, Korea, 30 Oct. - 2
Nov. 2000.
[12] M. Nidd, Service Discovery in DEAPspace, IEEE Personal
Communications, August 2001, pp. 39-45.
[13] G. Schiele, C. Becker and K. Rothermel, Energy-Efficient Cluster-
based Service Discovery for Ubiquitous Computing, Proc. ACM
SIGOPS European Workshop, Leuven, Belgium, Sept. 2004.
[14] N. H. Vaidya, Mobile Ad hoc networks routing, mac and transport
issues, Proceedings of the IEEE International Conference on Computer
Communication INFOCOM, 2004.
[15] Cisco, OSPF Design Guide, Document ID: 7039, Aug 10, 2005.
[16] Indukuri, R. K. R. (2011). Dominating Sets and Spanning Tree based
Clustering Algorithms for Mobile Ad hoc Networks. IJACSA -
Vol. 2, No. 6, 2011
12 | P a g e
International Journal of Advanced Computer Science and Applications,
2(2). Retrieved from http://ijacsa.thesai.org.
[17] Division, I. S. (2011). A Novel approach for Implementing Security over
Vehicular Ad hoc network using Signcryption through Network Grid.
IJACSA - International Journal of Advanced Computer Science and
Applications, 2(4), 44-48.

AUTHORS PROFILE
Kohei Arai
Prof K. Arai was born in Tokyo, Japan in 1949.
Prof K. Arais major research concern is in the
field of human computer interaction, computer
vision, optimization theory, pattern recognition,
image understanding, modeling and simulation,
radiative transfer and remote sensing. Education
background:
BS degree in Electronics Engineering
from Nihon University Japan, in March 1972,
MS degree in Electronics Engineering
from Nihon University Japan, in March 1974, and
PhD degree in Information Science from Nihon University Japan, in June
1982.
He is now Professor at Department of Information Science of Saga
University, Adjunct Prof. of the University of Arizona, USA since 1998 and
also Vice Chairman of the Commission of ICSU/COSPAR since 2008. Some
of his publications are Routing Protocol Based on Minimizing Throughput for
Virtual Private Network among Earth Observation Satellite Data Distribution
Centers (together with H. Etoh, Journal of Photogrammetory and Remote
Sensing Society of Japan, Vol.38, No.1, 11-16, Jan.1998) and The Protocol
for Inter-operable for Earth Observation Data Retrievals (together with
S.Sobue and O.Ochiai, Journal of Information Processing Society of Japan,
Vol.39, No.3, 222-228, Mar.1998).
Prof Arai is a member of Remote Sensing Society of Japan, Japanese
Society of Information Processing, etc. He was awarded with, i.e. Kajii Prize
from Nihon Telephone and Telegram Public Corporation in 1970, Excellent
Paper Award from the Remote Sensing Society of Japan in 1999, and
Excellent presentation award from the Visualization Society of Japan in 2009.

Lipur Sugiyanta
Lipur Sugiyanta was born in Indonesia at December 29, 1976. Major field
of research is computer network, routing protocol, and information security.
Education background:
Bachelor degree in Electrical Engineering from Gadjah Mada University
of Indonesia, in February 2000
Magister in Computer Science from
University of Indonesia, in August 2003.
He is now lecturer in Jakarta State University
in Indonesia. Since 2008, he has been taking part
as a PhD student in Saga University Japan under
supervision of Prof K. Arai.

Vol. 2, No. 6, 2011
13 | P a g e
Assessing 3-D Uncertain System Stability by Using
MATLAB Convex Hull Functions

Mohammed Tawfik Hussein
Electrical Engineering Department, Faculty of Engineering
Islamic University of Gaza
Gaza City, Gaza Strip

Abstract This paper is dealing with the robust stability of an
uncertain three dimensional (3-D) system using existence
MATLAB convex hull functions. Hence, the uncertain model of
plant will be simulated by INTLAB Toolbox; furthermore, the
root loci of the characteristic polynomials of the convex hull are
obtained to judge whether the uncertain system is stable or not. A
design third order example for uncertain parameters is given to
validate the proposed approach.
Keywords- Algorithm; 3-D convex hull; uncertainty; robust
stability; root locus
I. INTRODUCTION
Dealing with higher order system can be considered a
challenge and a difficult problem, therefore the contribution of
this paper is to the utilization of existence built- in MATLAB
Convex Hull algorithm and functions to handle such control
problems with less time consuming as will be illustrated
throughout this research paper.
A. Motivation and objectives
This paper is dealing with the robust stability of an interval
or uncertain system. Developing an algorithm that checks
robust stability of third order uncertain system such systems
will be an efficient and helpful tool for control systems
engineers.
B. Literature Review
The problem of an interval matrices was first presented in
1966 by Ramon E. Moore, who defined an interval number to
be an ordered pair of real numbers [a,b], with a b [1- 2].
This research is an extension and contribution of previous
publications and ongoing research of the author [3]-[9].
C. Paper approach
Three dimension (3-D) convex hull approaches is utilized
within MATLAB novel codes that is developed to assess 3-D
uncertain system stability, and the algorithm associated is
discussed and presented in this paper.
II. UNCERTAIN SYSTEMS AND ROBUST STABILITY
Due to the changes in system parameters due to many
reasons, such as aging of main components and environmental
changes, this present an uncertain threat to the system,
therefore such a system need special type of control system
called Robust to grantee the stability to the perturbed
parameters. For instances , in recent research robust stability
and stabilization of linear switched systems with dwell
time[10], as well stability of unfalsified adaptive switching
control in noisy environments [11] were discussed.
A. Robust D-stability
Letting D(p,q) denote the uncertain denominator
polynomial, then the roots of D(p,q) lie in a region D as shown
in Fig. 1 , then we can say that the system has a certain robust
D-stability properly.

Figure 1. D-region
Definition 1: (D-stability)
Let D _ C and take P(s) to be a fixed polynomial, then
P(s) is said to be D-stable if and only if all its roots lie in the
region D.
Definition 2: (Robust-D-stability)
A family of polynomials P={p(.,q):qeQ} is said to be
robustly D-stable, if all qeQ, p(.,q) is D-stable, i.e. all roots
of p(.,q)lie in D region. For special case when D is the open
unit disc, P is said to be robustly schur stable.
B. Edge Theorem
A polytope of a polynomial with invariant degree p(s, q) is
robustly D-stable if and only if all the polynomials lying along
the edges of the polytopic type are D-stable, the edge theorem
gives an elegant solution to the problem of determining the
root space of polytopic system [12], [13].
It establishes the fundamental property that the root space
boundary of a polytypic family polynomial is contained in the
root locus evaluated along the exposed edges, so after we
Vol. 2, No. 6, 2011
14 | P a g e
generate the set of all segments of polynomials we obtain the
root locus for all the segments as a direct location for the edge
theorem.
C. Uncertain 3x3 systems
Third order uncertain systems can take the following
general form:
| | | | | |
| | | | | |
| | | | | |
=
33 33 32 32 31 31
23 23 22 22 21 21
13 13 12 12 11 11
, , ,
, , ,
, , ,
a a a a a a
a a a a a a
a a a a a a
A

It has nine (9) elements which mean 2
9
possible
combination of matrix family if all elements were uncertain.
Generally, we have 2
n
possible combinations of an uncertain
system where n is number of uncertain elements in the system.
Characteristic equations for a general 3x3 matrix can be
calculated as shown below in equation (1):
] [
] [ ] [ ) (
22 13 31 23 12 31 32 13 21 33 12 21 32 23 11 33 22 11
13 31 12 21 22 11 33 11 32 23 33 22
2
33 22 11
3
a a a a a a a a a a a a a a a a a a
s a a a a a a a a a a a a s a a a s s P
+ + +
+ + + + + + = (1)

The aim of this paper is to calculate the family of all
possible combinations for a 3x3 uncertain matrix and so
family of possible characteristic equations can be calculated.
Then, using convex hull algorithm we will find exposed edges
of calculated polynomials and so the roots of exposed edges to
determine region of eigenvalues space for studied system.
D. Computing the Convex Hull of the Vertices
The convex-hull problem is one of the most important
problems from computational geometry. For a set S of points
in space the task is to find the smallest convex polygon
containing all points [14].
Definition 1: A set S is convex if whenever two points P and
Q are inside S, then the whole line segment PQ is also in S.
Definition 2: A set S is convex if it is exactly equal to the
intersection of all the half planes containing it.
Definition 3: The convex hull of a finite point set S = {P} is
the smallest 2D polygon O that contains S.
III. METHODOLOGY AND ALGORITHM
The main goal of this research is to provide a simple and
efficient algorithm to determine the bounds of an interval
matrix that represent three dimensional problems, hence assess
the stability of such an uncertain system by generating a
MATLAB Algorithm for three by three interval matrix.
Therefore the methodology and algorithm associated with will
be discussed and presented in the following sections.
A. Input data and program call
The developed program takes the nine elements of 3x3
uncertain matrix in a vector form and is called in MATLAB
command, and these elements can be either real number for
specific elements or interval for uncertain matrix entries.
B. Calculating family of possible matrices
To obtain the family of all possible matrices then the
following steps are performed within the function
<afamilynew4.m>:
- Check for size of each input element to find position
of uncertain elements.
- Declare input vector of 18 elements containing upper
and lower values of elements, coeff, if element is
specific then upper and lower values are equal.
- Calculate number of uncertain elements in matrix,
ss.
- Calculate number of possible combinations, 2
ss
.
- Weighting the coefficients vector by use of the
function <weig2.m> that assign, at each
combination, upper or lower values of elements by
making use of the idea if binary numbers in
combinations. This is done for all the 512 (2
9
)
possible combinations.
- Calculate family of 512 possible matrices, A.
- Check for repeated matrices and delete it and make
sure that remaining matrices are the 2
ss
unique
matrices, AA.
- Calculate 2
ss
possible characteristic polynomials,
polypointsn, according to equation 2.1. Note that
polypointsn is a matrix and is expected to have the
size of (2
ss
,3.
C. Find the 3-D convex hull of polynomials
For this purpose, the existence QuickerHull algorithm for
convex hulls is utilized and incorporated in the main
MATLAB Program [15], [6].
This algorithm has the advantage of being quicker than
convhulln, the built-in code in MATLAB, as illustrated in
Fig. 2.

Figure 2. Comparison of processing time in 3D between normal and quicker
hull
Then, 3D convex hull of system under study is plotted
according to the MATLAB QuickerHull code shown below.
4 4.2 4.4 4.6 4.8 5 5.2 5.4 5.6 5.8 6
0
0.5
1
1.5
2
2.5
log10 Numbers of points
T
i
m
e

[
s
]
Time comparison in 3D Space

Normal
Quicker
Vol. 2, No. 6, 2011
15 | P a g e
function tess=QuickerHull(P)
% QuickerHull N-D Convex hull.
% K = QuickerHull(X) returns the indices K of the points in
X that
% comprise the facets of the convex hull of X. X is an m-
by-n array
% representing m points in n-D space. If the convex hull has
p facets
% then K is p-by-n.
%
% CONVHULLN first attempts to clear points that cannot
be part of the
% convex hullthan uses Qhull. For dimensions higher than
six and
% points number less than 1000 no filtering is done.
%
% Example:
%
% X=rand (1000, 3);
% tess=QuickerHull(X);
% See also convhulln, convhull, qhull, delaunayn, voronoin,
% tsearchn, dsearchn.
%erroro check

if nargin>1
error('only one input supported')
end

[N,dim]=size(P);

if dim>1
if dim>5 || N<1000
%run the normal convhull for high dimensions and a few
points
tess=convhulln(P);
return
end

else
error('Dimension of points must be>1');
end

%%Filtering points

ncomb=2^dim;%number of combinations among the
dimensions
comb=ones(ncomb/2,dim);%preallocate combination
forbregion=zeros(2^dim,1);

%get all combinations using binary logic
for i=2:dim
c=2^(dim-i);
comb(:,i)=repmat([ones(c,1);-ones(c,1)],2^(i-2),1);
end
comb=[comb;-comb];%use combinations simmetry

%for each combination get forbidden region point
for i=1:ncomb/2
vect=zeros(N,1);
for j=1:dim
vect=vect+P(:,j)*comb(i,j);
end
[foo,forbregion(i)]=max(vect);
[foo,forbregion(i+ncomb/2)]=min(vect);
end

%get the simplyfied forbidden region
%for each dimension get upper and lower limit

deleted=true(N,1);
for i=1:dim

%get combination with positive dimension
index=comb(:,i)>0;

%upper limit
simplregion=P(forbregion(index),i);
upper=min(simplregion);

%lower limit
simplregion=P(forbregion(~index),i);
lower=max(simplregion);
deleted=deleted & P(:,i)<upper & P(:,i)>lower;
end

%delete points id that cannot be part of the convhull
index=1:N;
index(deleted)=[];

%Run QuickHull with the survivors
tess=convhulln(P(~deleted,:));
%reindex
tess=index(tess);
end
D. Calculate and sort roots of exposed edges of the convex
hull
First we calculate roots of all exposed edges, and plot the
convex hull of all roots of possible characteristic equations.
In order to encircle only imaginary components of roots,
we need a special sorting algorithm. We sort values according
to real and imaginary axes. Sorting may lead into "zero"
values in the imaginary matrix if there are real distinct or
repeated roots. So, a process of searching for "zeros" in the
imaginary matrix is held by replacing any zero vectors in the
imaginary by the following one.
IV. PROGRAM OUTPUTS AND RESULTS
For testing research papers program, an example of 3x3
uncertain system of a printer belt-drive system is presented
and simulated to validate the proposed approach.
Vol. 2, No. 6, 2011
16 | P a g e
A. Design Numerical Simulation Example
The method proposed in this paper will now be
demonstrated using printer belt-drive system that is described
mathematically by equation (2) and shown below in Fig. 3.

Figure 3. Printer belt-drive system
The system has the following model, see [17]-[19].
d
m
T
J
x
J
b
JR
k k K
J
kr
m
k
r
x
=
1
0
0
2
0 0
2
1 0
2 1
(2)

Where;
r: radius of pulleys [m].
k:spring constant of the belt.
k
1
:light sensor constant [V/m].
b: internal friction of the motor [Nms/rad].
R: coil resistance of motor [].
K
m
: motor constant [Nm/A].
J: total inertia of motor and pulleys [kgm
2
].
T
d
: disturbance torque [Nm].

Values of some parameters of this model can vary in the
following manner as shown in Table 1:
TABLE I. MODEL PARAMETERS
Parameters Values
m 0.2
k1 1
k2 0.1
r 0.15
b [0.1 0.25]
R [1.5 2.5]
Km 2
J 0.01

Then the A matrix is obtained with both lower and upper
parameters values as given below:

=
] 10 25 [ ] 8 33 . 13 [ ] 30 1200 [
0 0 ] 400 10 [
15 . 0 1 0
A

B. Source code
With four uncertain elements as shown in matrix A; thus,
we expect a family of 16 matrices. By calling the main
MATLAB program with the incorporated Quicker Hull
functions as presented below:

function main4(a11,a12,a13,a21,a22,a23,a31,a32,a33)

polypointsn =
afamilynew4(a11,a12,a13,a21,a22,a23,a31,a32,a33);
size(polypointsn)
tess=QuickerHull(polypointsn)
figure(4)
trisurf(tess,polypointsn(:,1),polypointsn(:,2),polypointsn(:,3))
r=zeros(length(tess),3);
for n=1:length(tess)
r(n,:)=(roots([1 tess(n,:)]))';
if imag(r(n,1))~=0
temp=r(n,3);
r(n,2:3)=r(n,1:2);r(n,1)=temp;
end
end
r
x=real(r)
y=imag(r)

figure(5)
plot(x,y,'b+')
hold on;
k=convhull(x,y);
plot(x(k),y(k),'-r')
grid on

q=length(r)
v=0;
z=zeros(1,3);
for i=1:q-1
if y(i,:)==z
v=v+1
z
for n=i:q-1
y(n,:)=y(n+1,:);
x(n,:)=x(n+1,:);
end
end
v1=v;
y;
if i==q-1-v
break
end
for s=1:v
if y(i,:)==z
Vol. 2, No. 6, 2011
17 | P a g e
v=v+1
for n=i:q-1
y(n,:)=y(n+1,:);
x(n,:)=x(n+1,:);
end
end
end
end

yn=y(1:q-v,:)
xn=x(1:q-v,:)
v
length(yn)
length(xn)

xxp=xn(:,3);
yyp=yn(:,3);
xxn=xn(:,2);
yyn=yn(:,2);

kp=convhull(xxp,yyp);
kn=convhull(xxn,yyn);
figure(6)
plot(x,y,'b+')
hold on;
plot(xxp(kp),yyp(kp),'-r'), plot(xxn(kn),yyn(kn),'-r')
hold off
grid on

figure(7)
plot(xxp(kp),yyp(kp),'-r'),hold on, plot(xxn(kn),yyn(kn),'-r')
axis equal
grid on
%plot(real(r),imag(r))
%rtess=QuicherHull(r)
end

C. Program output
Our proposed program is supposed to show all of family of
possible matrices in addition to roots values of characteristic
equations.
Four figures are generated while processing our algorithm,
Fig. 4 shows 3D convex hull of polynomials. Fig. 5 shows
roots locations on s-plane and encircles them by a convex hull.
While Fig. 6 shows encircling only imaginary parts of roots,
i.e. an identical polygons are generated on and below real axis
and other roots are shown also. Fig. 7 focuses on identical
polygons encircle imaginary parts of roots.

Figure 4. 3D convex hull of system polytopes

Figure 5. Convex hull of characteristic polynomials roots

Figure 6. Convex hull of imaginary part of characteristic polynomials roots

10
15
20
25
0
200
400
600
0
2000
4000
6000
8000
10000
12000
-14 -12 -10 -8 -6 -4 -2 0 2
-4
-3
-2
-1
0
1
2
3
4
-14 -12 -10 -8 -6 -4 -2 0 2
-4
-3
-2
-1
0
1
2
3
4
Vol. 2, No. 6, 2011
18 | P a g e

Figure 7. Convex hull of imaginary roots in focus
V. CONCLUSION AND FUTURE WORK
In this paper, the stability of uncertain system using
convex hull algorithm was tested. And we use MATLAB and
INTLAB toolbox to write program that can plot the 3D-
convex hull, root loci, step response, and frequency response
for any uncertain system. This paper tested the robust stability
of an interval 3x3 matrix by the implementation of Printer
Belt-Drive System. An efficient and enhanced algorithm was
introduced and improved for this purpose.
This algorithm can be easily extended to deal with higher
order matrices (n-dimensional system) without a very large
increase of processing time.
ACKNOWLEDGMENT
The author wishes to thank his graduate students who
enrolled in his course titled Uncertain Control Systems
during spring 2010 for their inputs, simulations and
contribution to this research paper, especially in selecting,
testing and running practical engineering examples.
REFERENCES
[1] R.E. Moore, Interval Analysis. Englewood Cliffs, N.J.: Prentice Hall,
Inc., 1966.
[2] L.V. Kolev, Interval Mathematics Algorithms for Tolerance Analysis,
IEEE Trans. On Circuits and Systems, 35, pp. 967-975, 1988.
[3] M. T. Hussein," An Efficient Computational Method For Bounds of
Eigenvalues of Interval System Using A Convex Hull Algorithm", The
Arabian Journal for Science and Engineering, Volume 35, Number 1B,
pages 249-263, April 2010,
[4] M.T. Hussein, Lecture Notes for Control of Uncertain Systems Course
Faculty of Engineering, IUG, Jan- May 2010.
[5] M.T. Hussein, Using a Convex Hull Algorithm in Assessing Stability
of a DC Motor with Uncertain Parameters: A computational Approach,
IJCSSEIT (ISSN:0974-5807), PP 23-34, Vol. 1, No. 1, January- June
2008.
[6] M.T. Hussein, A novel Algorithm to Compute all Vertex Matrices of an
Interval Matrix: Computational Approach, International Journal of
Computing and Information Sciences, IJCIS (ISSN 1708-0460), 2(2)
(2005), pp.137-142.
[7] M.T. Hussein, A Method to Enhance Tolerance Frequency Analysis of
Linear Circuits: Computational Approach, Al Azher University
Engineering Journal, 8(11) (, pp. 373-386, 2005
[8] M. T. Hussein, A Method for Determination of an Eigenvalue Bounds
for an Infinite Family of Interval Matrices: A Computational Approach,
the First International Conference of Science and Development (ICSD-
I), March 1-2, 2005.
[9] M. T. Hussein, Response Analysis of RLC Circuit with an Interval
Parameters, Engineering Week Conference, Gaza, Palestine, December
2004.
[10] L. I. Allerhand and U. Shaked, Robust Stability and Stabilization of
Linear Switched Systems with Dwell Time, IEEE Trans. on Automatic
Control, 56 (2), pp.381-386, 2011.
[11] G. Battistelli, E. Mosca, M. G. Safonov, and P. Tesi, Stability of
Unflasified Adabtive Switching Control in Noisy Envrionments,
IEEE Trans. on Automatic Control,55 (10) ,pp.2424-2429 , 2010.
[12] S.P. Bhattacharyya, H. Chapellat, and L.H. Keel, Robust Control: The
Parametric Approach. Prentice Hall, 1995.
[13] S.R. Ross and B.R. Barmish, A sharp Estimate for the Probability of
Stability for Polynomials with Multilinear Uncertainty Structure, IEEE
Trans. on Automatic Control, 53(2), pp.601-606, 2008.
[14] Ketan Mulmuley, Computational Geometry, an Introduction through
Randomized Algorithms, Prentice hall, 1994.
[15] INTLAB toolbox Users Guide Version 3 by math works,
http://www.ti3.tu-[harburg.de/rump/intlab/, last visit 20/5/2010.
[16] Luigi Giaccari, http://giaccariluigi.altervista.org/blog/,
Created:04/01/2009, last update 08/012009.
[17] C.Bradford Barber, David P. Dobkin, Hannuhuhdanpaa," The Quickhull
Algorithm for Convex Hulls", ACM Transactions on Mathematical
Software, Vol. 22, No. 4, Pages 469483, December 1996.
[18] R. C. Dorf, R. H. Bishop, Modern Control Systems, 10th ed., Prentice
Hall, Upper Saddle River, NJ 07458, 2005.
[19] K. Ogata, Modern Control Engineering, 4th ed., Prentice Hall, Upper
Saddle River, N.J., 2002.
AUTHORS PROFILE
Dr. Mohammed T. Hussein, an Associate Dean of Engineering for Research
and Development and an associate Professor of Electrical Engineering joined
the department of Electrical and Computer engineering at Islamic university
on August 2003. Dr. Hussein was named Director of e-Learning Center on
November 1, 2003. Prior to this appointment he served as a department Head
of Engineering Technology in College of Engineering at Prairie View A&M
University, Texas. Dr. Hussein earned a Ph.D. degree in electrical engineering
from Texas A&M University, College Station, Texas, USA. Dr. Hussein is a
registered professional engineer (P.E.) in the State of Texas. Dr. Hussein
worked for Motorola Inc., in Tempe, Az., and Oak Ridge National Laboratory
in state of Tennessee. His research interests include robust control systems,
computer algorithms and applications, and e-Learning. Dr. Hussein holds
scientific and professional memberships in IEEE(SM), Eta Kappa Nu, and
Tau Beta Pi. He is the recipient of numerous national, state, university,
college, and departmental awards including Who's Who among Americas
Best Teachers on 2000, Marquis Whos who among World Leaders on
2010, and Teaching Award in the College of Engineering. Dr. Hussein was
nominated and selected on 2003 as an evaluator for Accreditation board for
Engineering and Technology (ABET), USA. Dr. Hussein spent summer 2008
as a DAAD visiting Professor at Berlin Technical University, Germany, and
on 2009 was selected as academy Fellow, Palestine Academy for Science and
Technology.

-5 -4 -3 -2 -1 0 1 2
-3
-2
-1
0
1
2
3
Vol. 2, No. 6, 2011
19 | P a g e
Automating the Collection of Object Relational
Database Metrics

Samer M. Suleiman
Department of Computer Science, Faculty of Computer
and Information Technology,
Jordan University of Science and Technology,
Irbid, Jordan
Qasem A. Al-Radaideh, Bilal A. Abulhuda, Izzat M.
AlSmadi
Department of Computer Information Systems, Faculty of
Information Technology and Computer Sciences,
Yarmouk University,
Irbid, Jordan

Abstract The quality of software systems is the most important
factor to consider when designing and using these systems. The
quality of the database or the database management system is
particularly important as it is the backbone for all types of
systems that it holds their data. Many researches argued that
software with high quality will lead to an effective and secure
system. Software quality can be assessed by using software
measurements or metrics. Typically, metrics have several
problems such as: having no specific standards, sometimes they
are hard to measure, while at the same time they are time and
resource consuming. Metrics need also to be continuously
updated. A possible solution to some of those problems is to
automate the process of gathering and assessing those metrics. In
this research the metrics that evaluate the complexity of Object
Oriented Relational Database (ORDB) are composed of the
object oriented metrics and relational database metrics. This
research is based on common theoretical calculations and
formulations of ORDB metrics proposed by database experts. A
tool is developed that takes the ORDB schema as an input and
then collects several database structural metrics. Based on those
proposed and gathered metrics, a study is conducted and showed
that such metrics assessment can be very useful in assessing the
database complexity.
Keywords- Object Oriented Relational Database; Metrics; Software
Quality.
I. INTRODUCTION
The need to store and retrieve data efficiently relative to
traditional file structure retrieving was the basic motivation to
introduce the relational databases, which is basically the
process of representing the data on the form of a collection of
related relations (i.e. tables) [1].
Due to the increasing demand on more efficient techniques
to store, retrieve and represent complex and huge data types
such as images, a new data model is presented with the
inspiration of object oriented programming languages. Object
Oriented Database (OODB) emerged to meet these demands.
OODB is the process of representing the data in a form of
complex columns (i.e. objects) that contain attributes and
operations to access them [1] [2] [3] [4].
Object Relational Databases (ORDB) have recently evolved
for two reasons: the first is the limitation of the traditional
relational databases against the increasing demands of the huge
applications for storage and fast retrieval of data. The second is
the great complexity of the pure OODB [5] [2]. The integration
between the relational and the object oriented methodologies
could overcome some of the drawbacks that are known in the
relational databases, as well as, enable developers to utilize
the powerful features of the relational and object oriented
databases such as simplicity and usability [5] [6].
Generally, an ORDB system has two main natures: (1) the
dynamic nature which reflects the external quality of the
system that can be collected from the system at runtime (i.e.
dynamic or runtime metrics) (2) the static nature of the system
which reflects the internal quality and that can be measured at
the design time (i.e. static metrics) [7].
In this scope, metrics are tools to show indications that can
help software management in several aspects. For example it
can help facilitating the maintenance effort of the schema and
hence improve the quality and reduce the complexity of the
resulting schema [5]. Controlling the quality of the database
system in the design phase may help in preventing the whole
system from collapsing in the later phases (e.g. the
implementation phase). It also saves the cost and time for the
development process in general [8]. The assumption here is that
these metrics are standardized and formulated in order to be
measured as numbers, and thus facilitates the automation
process.
The main metrics for ORDB are: Table Size (TS),
Complexity of Weighted Methods (CWM), Cohesion Between
Methods (CBM), Coupling Between Objects (CBO), Number
of Inherited Properties (NIP), Referential Degree (RD), and
Depth in the Relational Tree (DRT) [9] [5] [8] [10].
Several researchers deduced that the value of automation
process comes from making the collection and the evaluation
process for software and system metrics easier in comparison
with manual techniques. In an example in this direction,
Stojanovic and El-Emam [11] constructed an object oriented
prediction model that can detect the faulty classes based on
previous data. They described an open source tool for C++
source code that can calculate the object oriented metrics from
the interface specifications at the design phase; these metrics
are size, coupling, and inheritance.
Vol. 2, No. 6, 2011
20 | P a g e
There are several challenges facing the evaluation of the
metrics. One of the main challenges is the ambiguity in the
definitions and formulation of these metrics in addition to the
nature of the process of collecting, processing and evaluating
those metrics [9] [5] [6]. According to [8] , most of the
developed software metrics are programs oriented and they are
not dedicated to database systems.
Al-Ghamdi et al. [11] described three tools for the
collection; analysis and evaluation of object oriented metrics.
The tools are: (1) Brooks and Buells: which contains: a parser,
a query engine and a model class hierarchy. (2) The second tool
for analyzing C++, and (3) The third tool for gathering OO
metrics. In addition, they built their own tool which collects
and measures inheritance coupling in object oriented systems.
They compared their tool with the three other tools in terms of
differences and main features.
Scotto et al. [12] mentioned that there is no standardization
for software measure and metrics where they suggested using
an intermediate abstraction layer to handle the frequent changes
on the extraction process for these metrics. They used an
automated tool to collect several web metrics. They
recommend to separate the two primary activities for any tool
that is supposed to measure the metrics, these two activities are
the process of extraction and storing the information from the
source code, and the second one is the process of analyzing
these information and get some conclusion from it.
AL-Shanag and Mustafa [13] proposed and built a tool to
facilitate the maintenance and understanding effort for C#
source code. The assumption is that software maintenance
process can benefit indirectly from software metrics through
predicting complexity and possible areas of problems in the
software code. Such metrics can help the developer and
maintenance software engineers in understanding the source
code of programs especially those that have no or little
documentation. The proposed tool collects several code
elements such as: interfaces, classes, member data and methods
from the source code. The tool can also collect the following
code metrics: Weighted methods per Class, Depth of
inheritance tree, Lines of code, Number of public methods, and
Data access metric.
As for this research, the main objective is to build a tool
that can collect and evaluate ORDB metrics that will enable
designers to calibrate and tune the database schemas to increase
usability, maintainability and quality for schemas. The
automation of this process becomes essential to overcome the
complexity of the evaluation process. Based on the most
common definitions of these metrics and units, this research
automates the collection of these metrics and realizes them by
units scale based on some formal definitions for these metrics.
For testing purposes, the proposed automation tool assumes
that the input schema is syntactically correct with respect to a
standard MS SQL database (particularly MS SQL 2003).
II. RESEARCH BACKGROUND
The challenge in OORD metrics is to find suitable
definitions and formulas in order to measure these metrics.
These measures are assumed to facilitate the process of
controlling the quality of the schema which will enhance the
overall performance of the associated information systems [9].
The evaluation of the quality of the database schema must
be validated formally. These metrics have to be validated
formally through both theoretical and practical approaches.
Piattini et al. [8] stated that the concentration is on the practical
approaches through developing practical experiments. They
developed an experiment to validate the ORDB metrics in
order to ensure the benefits of these metrics. They proof the
formality and validity of these metrics by repeating the same
experiments twice in CRIM center in Canada and University of
Castilla-La Mancha in Spain and get similar results. These
results showed that Table Size (TS) and Depth in the Relational
Tree (DRT) can be used as indicators for the maintainability of
database tables.
In ORDB, the table consists of two types of columns: the
first is the Standard Column (SC) which is defined as integer or
dynamic string data types, and the other one is the Complex
Column (CC) which is the User Defined Type (UDT) column.
According to this categorization, the metrics also are classified
as table related metrics. Those can be applied to a table which
includes: TS, DRT, RD, and Percentage of Complex Columns
of tables (PCC), Number of Involved Classes (NIC), and
Number of shared classes (NSC). Other metrics are applied to
schema which includes the DRT, RD, PCC, NIC, and NSC [8].
ORDB schema requires an extra metric due to additional
capabilities which come from the object oriented features in
order to ensure its internal quality which will be reflected on its
external quality in terms of understandability, usability and
reliability [9].
A. Metrics of ORDB
Justus and Iyakutti [9] defined and formulated the metrics
of ORDB based on three schemas. These metrics are:
Table Size (TS): It represents the summation of the size
of both the simple columns (SC) which includes the
traditional attributes data types, such as integer and
varchar, and the complex columns (CC) which represents
the User Defined Types (UDTs). The larger the value of
this metric leads to a higher maintenance cost [9] [8]. It is
calculated as follows:

(1)
Where is a metric function, SC represents the Simple
Columns and n represents their numbers, CC is complex
columns and ms are their numbers.
Complexity of Weighted Methods (CWM): It represents
the summation of the whole complexities for each
weighted method in the table [9] [14]. It is calculated as
follows:

(2)
Where C
i
is the complexity of method i.
Cohesion Between Methods (CBM): It measures the
connectivity between two or more methods and it is
measured as the proportion between the similar used
attributes in the methods of the class to the total number
of attributes. High cohesion is desired which indicates that
we are grouping together related methods. Low cohesion
Vol. 2, No. 6, 2011
21 | P a g e
should have a negative impact on maintenance [9] [7] [14]
[3]. It is calculated as follows.

(3)
Where I is the instance of the attributes for the class, and
V is the total number of variables.
Coupling Between Objects (CBO): It represents the
dependency or the connectivity between two methods that
exist in two different classes. While high cohesion is
desired, high coupling is not as it complicates the design
and will complicate maintenance effort, update and reuse
[9][7]. It is calculated as follows:
(4)
Where Min v
j
is the number of methods called and the
average number of arguments involved in each
invocation.
Number of Inherited Properties (NIP): It represents the
number of the properties that have been inherited by the
child and thus its value determines high coupling
complexity as a trade off from its reuse [9]. It is
calculated as follows:

(5)
Where M
j
= [No. of types * CWM] + [No. of methods *
CBO].
Referential Degree (RD): It represents the number of
foreign and reverse refernce keys in the database schema
[9] [15] [8]. It is calculated as follows:

(6)
Where FK is the foreign key and Rvref is the inverse
reference.
Depth in the Relational Tree (DRT): It represents the
longest referential path between the tables in or of a
database schema [9] [10]. It is calculated as follows:

(7)
Where d is the distance between the related tables.

B. Metrics Units
In an attempt to unitize the ORDB metrics, Justus S and K
Iyakutti [9] proposed to formulate and calibrate some of these
metrics. This is because there is a great need for a standard
scale for these ORDB metrics to be correlated with software
code metrics. Justus and Iyakutti [9] proposed some units for
this purpose based on some experimental studies. The units are:
Column Complexity (clm) unit for the TS metric. Figure 1
shows the relation between TS and the cost in term of clm
unit.
If the value is in one of the scaled ranges then we can
conclude its cost and complexity. For example if the TS
value is 50 clm, this means that the maintenance table cost
is between optimal and low scale.

Low Optimal High

31-45 46-60 61-75 76-90 91-115
Figure 1: The Table Size unit scale.
Number of interactions per variables set (intr/vs) unit. It is
used to measure the cohesion metric as calculated in
equation (3).
Number of messages imported or exported per interaction
(msgs/intr) unit. It is used to measure the coupling metric
as shown in equation (4).

For both intr/vs and msgs/intr, Figure 2 shows the relation
between these two measures against reusability and
maintainability where high intr/vs indicates a high class
reusability. However, the situation with msgs/intr is the
opposite which means that when msgs/intr is high, this
indicates a less class reusability.

Low High

0 1
Figure 2: The COM and CBO Units Scale.
The laxity unit is used to measure the reusability metric
(NIP). The reusability denotes the usage of the same
class-type another time in another class or type. Figure 3
shows the relationship between the reusability of the
class-type and this unit. The higher the value of laxity the
more the probability for the class-type to be reused.

Low High

4.5 - 6 6.1-7.5 7.6 9 9.1-10.5 10.6 12
Figure 3: The NIP laxity unit scale.
III. THE DEVELOPED TOOL
The goal of developing this tool is to automatically collect and
evaluate ORDB metrics. The proposed tool should enable
designers to calibrate and tune the database schemas to
increase usability, maintainability and quality for schemas.
The proposed tool consists of three main modules: The
tokenizer, the lexical analyzer, and the metrics calculator. The
lexical analyzer is considered to be the main module of the
tool. The architecture of the developed tool is illustrated in
Figure 4 and the basic data model for the tool is illustrated in
Figure 5.

Vol. 2, No. 6, 2011
22 | P a g e

Figure 4: The Lexical Analyzer Architecture
A. Tokenizer
The tokenizer aims to facilitate the evaluation process by
reading the stream of the text file (i.e. the schema) and tokenize
it by recognizing each word individually. In this phase, there is
no need to build the relationships between these tokens where
the analyzer will handle this job. Tokens will be stored in the
database and tagged in order to be used later. Some of these
tokens will not be stored in the database for efficiency reasons
and because it will not produce any relevant information to the
process such as punctuation marks and other language related
symbols.
The tokenizer also stores the schema as an entry to the
database and links it with other objects and artifacts, in order to
create a dataset and accomplish related statistics to obtain a
design trend for each dataset. It is worth to mention here that
some metrics could be calculated directly such as Lines of
Code (LOC). As a summery for this part: the input for the
tokenizer is a schema text, and the output is a set of tokens.
B. Analyzer
The analyzer is the main part of the tool. It actually
accomplishes a great percentage of the evaluation process for
the schema. It acts as the bridge between the other parts of the
tool. The analyzer starts by reading the tokens from the
tokenizer and loads them onto the computer memory in order
to be processed and classified into the basic objects for the
evaluations. These objects are: the tables, the complex columns
and their methods. The challenge here is to reserve the
relational structure of the database that will be used
subsequently.
The analyzer will check each token against a set of
constraints. It will check if the token is an SQL reserved word
and if this word is related to the measures or not. For example
the table name which can be known if it appears after the two
reserved words:create table or at least the word, table, hence
will be stored in the database and connect all the following
distinguished tokens with this entry until the next table name
appears.
The tool will store all the information that the metrics
formulas mentioned earlier may need to be collected in
efficient and smooth manners. The design of the tool is based
on separating the analyzer from the rest of other modules. This
may help for any future changes on the metrics equation and
guarantee that these changes will not affect the analyses
process.
The analyzer distributes the tokens in different connected
relational tables and ensures that all the tokens have been
stored in the exact place. The equation metrics can easily be
applied on these values that we can get from the tables. To
summarize: the analyzer inputs are stored tokens, the output
from the analyzer categorizes tokens are stored in related
tables.
C. The Calculator Engine
The calculator engine module uses the information stored in
tables to calculate the different metrics presented in section 2.
The results are saved in the database associated with each
schema for later possible revisions. The calculator module has
two sub-modules: the scalar and the evaluator. The scalar role
is to map the result of each equation against the scales
presented in section II.
The scalar gives meaning to these numbers by assigning a
suitable unit to each one of them. The Evaluator evaluates the
overall quality of the schema based on the quantized numbers
that have been already obtained from the previous units.

Vol. 2, No. 6, 2011
23 | P a g e

Figure 5: The Tool Data Model
IV. EXPERIMENTS AND EVALUATION
In order to evaluate the tool we will provide a complete
example that shows how the tool is used to accomplish the
tasks. Figure 6 represents a small part of a schema definition
adopted from [9] that contains several tables. Table 1
represents the analysis for the tables tokens. Figure 7 shows a
sample script for creating a person_t type for the same
schema definition.

CREATE TABLE Tab_Staff
( emp_no varchar(4),
person_info person_t,
do_joining date ,
working_department department_t ,
work_for varchar ( 7 ) FK Tab_Student ( roll_no ) inverse ref ) ;
CREATE TABLE Tab_Student
( roll_no varchar(7),
person_info person_t,
in_department department_t,
ward_for varchar(4) );
Figure 6: A Sample Schema Definition Script (Justus and Iyakutti, 2008)
As mentioned earlier, the main input for the tool is the
schema script (written in SQL 2003 standards) and it is
assumed to have object-oriented and relational features that are
syntactically correct. The schema is read by the tool as a stream
of words or tokens. The main two segments are: the schema
definition segment and the implementation segment. They are
separated using a special tag or flag.
TABLE 1: TABLES OF THE SAMPLE IN FIGURE 6
tabID tabName schID No_fk No_rev
1 Tab_Staff 1 1 1
2 Tab_Student 1 0 0

CREATE TYPE person_t
( Name varchar(20) ,
Gender varchar(1) ,
Birth_date date,
Address_info address_t,
MEMBER FUNCTION set_values ( ) RETURN person_t,
MEMBER PROCEDURE print_person ( ) );

Figure 7: A Sample Schema definition for UDT
The tool analyzes this UDT table and relates each member
to this new defined type:
Simple attributes: those represent the standard types just like
the simple columns each recognized one is stored in
SimpleAttributes table and relates to its entry in the CC
table. Table 2 represents the record Instances for the
SimpleAttributes table after the tool reads the script.

TABLE 2. SIMPLE ATTRIBUTES INSTANCES
aID aName typeID
1 Name 1
2 Gender 1
3 Birth_date 1
Complex Attributes: the other possible member items for the
UDT is the complex column which is either stored on the
Types table or need to be treated as we did with the same
item member in table type. It can be related to one of the CC
entries. Table 3 represents the record Instances for the
ComplexAttributes table after the tool reads the script.

TABLE 3. COMPLEX ATTRIBUTES INSTANCES
caID caName typeID
1 Address_info 1

Vol. 2, No. 6, 2011
24 | P a g e

Member methods. These members are recognized by the
keywords PROCEDURE and FUNCTION, so once the tool
has read one of these tokens it will consider the next one as
a member function and insert it to the Members table and
relates it to its entry in the CC table. Table 4 represents the
record Instances for the ComplexAttributes table after the
tool reads the script.

TABLE 4: MEMBERS INSTANCES
mID mName typeID
1 set_values 1
2 print_person 1

After this analysis is accomplished, all the necessary
information for the schema definition are already stored in the
database tables. The following subsection describes the
calculation for the Tables Size, RD and DRT metrics, by using
the stored data.
A. Calculations
The following sub sections present the calculations process
for Table size metrics, Referential Degree, and Depth of
Referential Tree
The Calculation of TS
In order to calculate the table size, the following
information has to be retrieved from relevant tables:
Tables: choose every record which represents a unique
table, and retrieve the tabid, in order to be used to retrieve
the related records from the other tables.
SimpleColumns: the tool retrieves the related simple
columns records from the table for a specific tabid value
through the FK_Tables_SimpleColumns relation (see
Figure 8). The number of the retrieved records is stored
temporarily in order to be used later.
ComplexCoulmns the tool retrieves the related complex
columns records from the table for a specific tabid value
through the FK_Tables_ComplexColumns relation. The
number of the retrieved records is stored temporarily in
order to be used later.
The tool checks some other tables which are:
SimpleAttributies, ComplexAttributies, and
Members. It retrieves all the related records for the type of
the complex columns and stores the count sequentially.
The retrieving process is making use of the following
relation between related tables:
FK_Types_ComplexColumns,
FK_ComplexColumns_SimpleAttributes,
FK_ComplexColumns_Members
FK_ComplexColumns_ComplexAttributes.

The result of this activity is a number that represents the
size of the specific type that is mentioned in the table
definition. The calculation is based on equation (1) presented in
section 2 and the extracted size is stored on the size field of
the Types table. Table 5 illustrates these calculations.

TABLE 5: UDT INSTANCES WITH SIZES
typeID typeName Size
1 person_t 14
2 department_t 16
3 address_t 7

By summation of all the retrieved values from these tables,
the tool can calculate the size for each table (i.e. number of
records from the SimpleCoulmns table + size of the existing
UDT type(number of records from SimpleAttributies table +
number of records from ComplexCoulmns table + number
of records from the Memebers table). This is illustrated in
Table 6.
TABLE 6: TABLE SIZES
tabID tabName schID No_fk Size
1 Tab_Staff 1 1 22
2 Tab_Student 1 0 23

To understand these numbers the tool will check them
against the proposed scale as in Figure 1. This indicates that
tables in this schema fall in the first scale (<= 31-45) and they
have low complexity level. Comparing this automatic
measurement process with the manual calculations presented
by Justus and Iyakutti [9] for TS metric for the same sample
schema, the tool shows the same results.
The Calculation of RD metric
The calculation of this metric depends on equation (6)
presented in section 2. The tool looks for the existence of the
keywords FK which stands for foreign key and for inverse
ref which stands for inverse reference that means the relation
is bidirectional between the two subject tables.
The tool counts the frequency for FK token in the schema
definition and stores the summation on the no_fk field for
each table as illustrated in Table 1. The same process is applied
on the inverse reference to get the number that represents their
counts. In order to calculate the value for this metric, the tool
adds the two summations (i.e. sum of FKs and sum of inverse
refs) for each table. Thus we can conclude the complexity of
the schema according to the value for this metric. The higher
the value of RD means the higher the level of complexity for
the schema. Comparing results with [9] for this metric for the
same sample schema, the tool gets the same results as they had.
The Calculation of DRT metric
The calculation of this metric depends on equation (7), in
which the tool stores the referential path between tables by
analyzing the foreign key constraint. It stores the id for the
referencing and referenced table. It compares the frequency of
the same tabid value in both columns and then counts this
frequency to get the depth of the referential tree.
The tool gets the length of DRT as a single number that
represents the number of tables associated with this relation.
Each time it counters the FK keyword, it stores both tables
name in the Tree table. The first table name after FK
Vol. 2, No. 6, 2011
25 | P a g e
keyword is stored in the DetailTab field where as the current
processing table is stored in MasterTable field. Table 7
illustrates these findings:
TABLE 7: DRT FOR THE SAMPLE SCHEMA
MasterTable DetailTab
Tab_Student Tab_Staff
Tab_Staff Exam

The tool then performs an SQL Statement that counts the
existence frequency of each table in both columns for each
record in the Tree table.
B. Tool Evaluation
Table 8 summarizes the results of this research and the
output for the developed tool compared with Justus and
Iyakutti [9] manual work to calculate the table size. The results
in Table 8 show that the tool gets the same results in addition to
storing each object in a relational manner in order to be
retrieved when they are required. Justus and Iyakutti [9]
calculated neither the Exam table size nor the subject-t
type size. Our tool calculates them and compared with the
results of manual calculations which are the same.
The same comparison is made between the two approaches
with respect to the remaining calculated metrics; the
Referential degree (RD) and DRT. The result is illustrated in
Table 9, which shows that the automation process facilitates the
evaluations for these metrics since the manual approach
requires human memorization and tracing.
Figure 8 presents a screen shot of the analysis process that
appears for the Get Tokens button. All the analyzed tokens
appear in the targeted grid view.
The evaluation process for the software system quality in
general is said to be complex especially for the ORDB metrics.
This is due to different aspects such as the lack of formal
definitions and standard evaluation. In addition to the lack of
such tools that are capable to perform this evaluation.
However, one key point that was investigated in this
research is the need to separate the analysis process from the
evaluation and calculations processes regarding to these
metrics. This separation makes the two main processes
independent from one another and handles, partially, the
problem of changing the metrics evaluation or definition.
The developed tool performs the analysis process on the
schema definition, separates different ORDB artifacts and
stores them separately. These artifacts include: tables, their
simple columns, their new defined types for the complex
columns and the objects created for the complex columns
which includes the simple attributes and the members of these
objects (i.e. procedures or functions).
The automation process facilitates metrics gathering and
evaluation and gives the designers and developers more
capabilities and perspectives to ensure the quality of the ORDB
systems. It performs the same equations proposed by [9] for
TS, DRT, and RD metrics and gets the same results in terms of
accuracy compared with manual metrics calculation while
improving the performance through calculating those metrics
automatically.
The tool is adaptable to the changes of the metrics
equations since it separates the analysis process from the
evaluation process and this may be useful for the
standardization effort by tuning only the evaluation parts of this
tool. The analysis may not have to be changed or it may require
a little modification. This can help in the continuous evolution
of metrics formulas construction and assessment.
In future, the automation process should be extended to
include the remaining proposed metrics such as: (COM) and
(CWM). It should be also extended to include more metrics
that will be further investigated through looking at different
database systems. Another future issue to deal with is that some
proposed metrics have no formal scale that enables the
designers to conclude the quality of the schema in terms of its
complexity and maintainability, and thus future work may
define a formal scale for these kinds of metrics.
It is recommended to extend the automation process to
include different schemas written in different databases and
formats such as: Oracle and MySql. Once these modifications
are implemented, there is a need to build a dataset to test them
and calibrate their results. The tool may be extended to have
the ability to obtain the schema from an existing database and
perform the same process as it did with the text schema format.

Vol. 2, No. 6, 2011
26 | P a g e
TABLE 8: COMPARISON BETWEEN MANUAL AND AUTOMATE TABLE SIZE CALCULATIONS

TABLE 9: A COMPARISON BETWEEN MANUAL AND AUTOMATE RD AND DRT CALCULATIONS
Table Name
Foreign Key
Statement
Manual Automation
RD DRT RD DRT
Foreign
Key
inverse ref
Memorize master,
detail
Foreign
Key
inverse
ref
Store Master, Detail
Tab_Staff
work_forvarchar (7)
FK Tab_Student
(roll_no) inverse ref
1 1
counterFK +
counterINVREF=2
Tab_Student 0 0 0
Exam
Staff_chargevarchar
(4) FK Tab_Staff
(emp_no) inverse ref
1 1
Memorize master,
detail
counterFK +
counterINVREF=2
Store Master, Detail

Trace for detail
for each
memorized master

Select count(*) where
master=detail from tree=1

Vol. 2, No. 6, 2011
27 | P a g e

Figure 8: Screen Shot of Results
REFERENCES
[1] Elmasri R. and Navathe S.,"Fundamental of Database Systems", 5th
edition, 2007.
[2] Behm A., Geppert A., and Dittrich K.,On The Migration Of Relational
Schemas and Data to Object-Oriented Database Systems In Proc. of the
5th International Conference on Re-Technologies in Information
Systems, Klagenfurt, Austria, December, 1997.
[3] Vaishnavi V., Purao S. and Liegle J. Object-oriented product metrics: A
generic framework, Information Sciences 177 (2007) pp. 587606.
[4] Van Zyl P. and Boake A., Comparing the Performance of Object
Databases and ORM Tools in Proceedings of the 2006 annual research
conference of SAICSIT, 2006.
[5] Baroni A.L., Coral C., Brito e Abreu1 F., and Piattini M., Object-
Relational Database Metrics Formalization, Proceedings of the Sixth
International Conference on Quality Software (QSIC'06), 2006.
[6] Justus S. and Iyakutti K., Measurement Units and Scales for Object-
Relational Database Metrics, First International Conference on
Emerging Trends in Engineering and Technology, 2007(a).
[7] Justus S. and Iyakutti K., Assessing the Object-level behavioral
complexity in Object Relational Databases, International Conference on
Software Science, Technology and Engineering, 2007(b).
[8] Piattini M., Genero M., Coral C. and ALARCOS G., Data Model
Metrics In Handbook of Software Engineering and Knowledge
Engineering: Emerging Technologies, World Scientific, 2002.
[9] Justus S. and Iyakutti K., A Formal Suite of Object Relational Database
Metrics, International Journal of Information Technology 4:4, pp. 251-
260, 2008.
[10] Stojanovic M. and El Emam K. ES1: A Tool for Collecting Object-
Oriented Design Metrics, 2001.
[11] Al-Ghamdi J., Elish M. and Ahmed M., A tool for measuring
inheritance coupling in object-oriented systems, Information
SciencesInformatics and Computer Science: An International Journal,
v.140 n.3, p.217-227, 2002.
[12] Scotto M., Sillitti A., Succi G., and Vernazz T., A Relational Approach
to Software Metrics, 2004.
[13] Al-Shanag F. and Mustafa S., Clustering Data Retrieved from C#
Source Code To Support Software Maintenance, Master graduation
project, Department of computer and information system, Yarmouk
university, 2009.
[14] Marinescu R.,Using Object-Oriented Metrics for Automatic Design
Flaws Detection in Large Scale Systems, Object-Oriented Technology
(ECOOP98 Workshop Reader), LNCS 1543, 1998
[15] Mao C., "DBViewer: A Tool for Visualizing and Measuring Dependency
Relations between Tables in Database," wcse, vol. 4, pp.13-17, World
Congress on Software Engineering, 2009.

Vol. 2, No. 6, 2011
28 | P a g e
A Decision Support Tool for Inferring Further
Education Desires of Youth in Sri Lanka

M.F.M Firdhous*
Department of Information Technology
Faculty of Information Technology
University of Moratuwa
Moratuwa
Sri Lanka

Ravindi Jayasundara
Department of Mathematics
Faculty of Engineering
University of Moratuwa
Moratuwa
Sri Lanka
*corresponding author

Abstract This paper presents the results of a study carried out
to identify the factors that influence the further education desires
of Sri Lankan youth. Statistical modeling has been initially used
to infer the desires of the youth and then a decision support tool
has been developed based on the statistical model developed. In
order to carry out the analysis and the development of the model,
data collected as part of the National Youth Survey has been
used. The accuracy of the model and the decision support tool
has been tested by using a random data sets and the accuracy
was found to be well above 80 percent, which is sufficient for any
policy related decision making.
Keywords Educational Desires of Youth; Univariate Analysis;
Logit Model; Data Mining.
I. INTRODUCTION
Sri Lanka has witnessed several incidents of youth unrest in
the recent past. Out of these insurgencies, two insurgencies
involved the youth of the south while the other one involved
the youth of the north. There have been many discussions and
debates about youth unrest and the increasingly violent and
intolerant nature of their conflicts. Since these discussions have
been rather impressionistic there has always been the need for
systematic studies to obtain information on Sri Lankan youth
and their background and desires [1]. In order to collect up to
date information, targeting to explore facts on Sri Lankan
youth and their perceptions, an island wide national youth
survey has been carried out. This has been conducted as a joint
undertaking involving the United Nations Development
programme (UNDP) and other six Sri Lankan and German
institutions in the turn of the century. In this survey they have
considered four main segments of youth, that is, their politics,
conflicts, employment and education. Further Education
Desires of youth have been selected to be studied further in this
research. The relationship between the types of further
education desire of youth in Sri Lanka had been studied with
relation to other social factors.
Education domain consists of many different areas but
presently in Sri Lanka only a few areas are catered by the
national educational institutes [1]. By finding out the
educational desires of the youth, it will be possible to design
and develop educational and professional programmes and
institutes which can be readily accepted by the youth and give
better results than that can be achieved by only pursuing
traditional programmes. Data Mining which is a powerful tool
that can recognize and unearth significant facts, relationships,
trends and patterns can be employed to discover this
information [2]. In this project, a data mining model has been
developed to predict the educational desire of youths at an
early stage from other social data.
II. THEORETICAL BACKGROUND
Descriptive statistics are used to describe the basic features
of the data in a study. They provide simple summaries about
the sample and the measures. Together with simple graphics
analysis, they form the basis of virtually every quantitative
analysis of data [3]. Univariate analysis is the simplest form of
quantitative (statistical) analysis.
The analysis is carried out with the description of a single
variable and its attributes of the applicable unit of analysis.
Univariate analysis contrasts with bivariate analysis the
analysis of two variables simultaneously or multivariate
analysis the analysis of multiple variables simultaneously.
Univariate analysis is also used primarily for descriptive
purposes, while bivariate and multivariate analysis are geared
more towards explanatory purposes [4]. Univariate analysis is
commonly used in the first stages of research, in analyzing the
data at hand, before being supplemented by more advance,
inferential bivariate or multivariate analysis [5]. Pearson's chi-
square test is the best-known of several chi-square tests
statistical procedures whose results are evaluated by reference
to the chi-square distribution [6].
With large samples, a chi-square test can be used. However,
the significance value it provides is only an approximation,
because the sampling distribution of the test statistic that is
calculated is only approximately equal to the theoretical chi-
squared distribution. The approximation is inadequate when
sample sizes are small, or the data are very unequally
distributed among the cells of the table, resulting in the cell
counts predicted on the null hypothesis (the "expected values")
Vol. 2, No. 6, June 2011

29 | P a g e
being low. The usual rule of thumb for deciding whether the
chi-squared approximation is good enough is that the chi-
squared test is not suitable when the expected values in any of
the cells of a contingency table are below 5 or below 10 when
there is only one degree of freedom [7]. In contrast the Fisher
exact test is, as its name states, exact, and it can therefore be
used regardless of the sample characteristics [8]. For hand
calculations, the test is only feasible in the case of a 22
contingency table. However the principle of the test can be
extended to the general case of an mn table [9].
Logistic regression is most frequently employed to model
the relationship between a dichotomous (binary) outcome
variable and a set of covariants, but with a few modifications it
may also be used when the outcome variable is polytomous
[10].
The extension of the model and the methods from a binary
outcome variable to a polytomous outcome variable can be
easily illustrated when the outcome variable has three
categories. Further generalization to an outcome variable with
more than three categories is more of a notation problem than a
conceptual one [11]. Hence, it will be considered only the
situation when the outcome variable has three categories.
Main objective of fitting this statistical model is to find out
the sequence of variables being significant to the model, so that
the sequence of variables, as a whole or a subsequence starting
from the first variable, will be used as necessary in constructing
a decision tree. In this study we make use of this statistical
model not for interpretations but only for doing a comparison
with the outcome of a Data Mining approach in decision
making.
III. ANALYSIS
Univariate analysis is carried out with the purpose of
analyzing each variable independently from other variables.
Therefore each of the categorical variables measured as a factor
is cross tabularized with the dependent variable Type of
Further Educational Desires calculating percentages of
respondents belonging to each combination of levels, and the
Chi-Square Test is performed in order to measure the strength
of association between factors and the response of interest. A
tolerance rate of 20 percent has been fixed as the significance
level for further analysis. Table 1 shows the results of the
analysis.
Table 1: Results of Univariate Analysis at 20% Tolerance Level
Explanatory
Variable
Pearsons
Chi-square
Value
Degrees
of
Freedom
Asymptotic
Level of
Significance
Significant
Variables
at 20%
level
Age group 208.0 6 0.0000
Gender 68.751 3 0.0000
Educational
level
198.1 6 0.0000
Ethnicity 53.032 12 0.0000
Province 198.1 24 0.0000
Sector 6.882 6 0.2020
Social class
(self definition)
77.154 9 0.0000
Present
financial
25.455 6 0.0000
situation
Financial
situation in past
10.391 6 0.0810
Whether school
provide good
education
7.723 9 0.4860
Major problems
with education
95.518 21 0.0000
Access to
educational
facilities
90.419 6 0.0000
Type of activity 1066.0 18 0.0000

From the results shown in Table 1, the variables with their
p-values less than 0.2 have been detected as significant.
A. Fitting a Statistical Model
The main purpose of this part of analysis is to determine the
factors, which affect or are associated with having different
types of Further Educational Desires in youth. Though several
variables have been identified as significant factors, where each
could independently build a significant effect on developing
different wishes on education among youth, due to the
confounding nature of these factors, it is not easy to conclude
on their corporative influence on making Further Education
Desires different in people.
Therefore this modeling approach can be very much useful
in detecting the genuine effect of these factors when adjusted
for some other factors as well.
Since the response variable is Multinomial and the scale of
response levels are Nominal, it was decided to work out a
logit link in regression modeling. Therefore a Generalized
Logit Model will be fitted to accomplish the objective.
B. Fitting the best fitted Generalized Logit Model
The Forward Selection procedure is used in selecting
variables to the model. In assessing the fit of the terms to the
model, the difference in deviance of the two models compared,
which is distributed as Chi-Squared has been used at the 5%
significance level. However the terms will be selected to the
model, as they do the best representation of all the data.
The results obtained in following the steps of fitting a
Generalized Logit Model using the procedure CATMOD in
SAS package, are tabularized in the body of the analysis.
Let the Null Model
be,
;
where
is the probability that a respondent has the

type of Further Education Desire, f , and type F is the
Further Education Desire category No Desire.
Fitting Main Effects to the Model
Step 1: Null Model vs One Variable Model (Model 1)
Table 2 shows sample of data used in devising Model 1.
Vol. 2, No. 6, June 2011

30 | P a g e
Table 2: Adding the 1
st
Most Significant Variable to the Model

From Table 2, the lowest p-value is associated with the
variable Type of activity. The selection procedure of the most
significant variable requires that Type of activity be added to
the Null Model as the first step of developing a model where
the Type of Further Education Desire being the response
variable.
The explanatory variables in the model: Type of activity
Model 1:

where
is the probability that a respondent in Type of

activity category i has the
type of Further Education

Desire, f , and type F is the Further Education Desire
category No Desire.
Thus two logits are modeled for each activity type: the logit
comparing Technical/Vocational Education to No Desire and
the logit comparing University/Higher Education to No Desire.
Model 1: for Type of activity i
logit(response1/response3)i=1 models the probability of
response category 1 relative to the response category 3.
logit(response2/response3)i=2 models the probability of
response category 2 relative to the response category 3.
There are separate sets of intercept parameters
and
regression parameter
for each logit and the matrix xi is the

explanatory variable for the i
th
population.
Step 2: Model 1 vs Two Variable Model (Model 2)
nd
Model 2
(terms added)
Raw
Deviance
Difference
in
Difference
in df
p-value
(-2Log
likelihood)
Deviance
Model 1 3450.942 0 - -
Age group 3437.718 13.2236 4 0.01023338
Gender 3434.514 16.4276 2 0.000270889
Educational
level
3288.146 162.7954 4 3.67573E-34
Ethnicity 3427.465 23.4763 7 0.001407636
Province 3316.505 134.4366 16 8.79825E-21
Social class 3420.852 30.0896 6 3.77963E-05
Present
financial
situation
3432.545 18.3971 4 0.001031951
Financial
situation in
past
3448.478 2.4638 4 0.65112949
Major
problems with
education
3427.67 23.2713 14 0.055995893
Access to
educational
facilities
3437.028 13.9133 4 0.00757697

Since the variable Educational Level has the lowest p-
value, it was brought into the model that has been adjusted for
Type of activity.
The explanatory variables in the model: Type of activity,
Educational Level
Model 2:

where the matrix
is the set of explanatory variables for

the ij
th
population.
Step 3: Model 2 vs Three Variable Model (Model 3)
rd
Model 3
(terms added)
Raw
Deviance
(-2Log
likelihood)
Difference
in
Deviance
Difference
in df
p-value
Model 2 3288.146 0 - -
Age group 3276.062 12.0842 4 0.016735994
Gender 3268.712 19.4339 2 6.02535E-05
Ethnicity 3267.805 20.3416 7 0.00487722
Province 3155.297 132.8494 16 1.79279E-20
Social class 3266.442 21.7041 6 0.001369785
Present
financial
situation
3276.61 11.536 4 0.02115679
Financial
situation in
past
3285.776 2.3706 4 0.667946717
Major
problems with
education
3267.495 20.6513 14 0.110907989
Access to
educational
facilities
3281.422 6.7239 4 0.151218299

Since the variable Province has the lowest p-value, it was
brought into the model already adjusted for Type of activity
and Educational Level.
Model 1
(terms added)
Raw
Deviance
(-2Log
likelihood)
Difference
in
Deviance
Difference
in df
p-value
Null Model 4540.468 0 - -
Age group 4332.934 207.5343 4 9.00999E-44
Gender 4474.573 65.8957 2 4.90829E-15
Educational
level
4359.197 181.2715 4 3.97611E-38
Ethnicity 4490.291 50.1769 7 1.33346E-08
Province 4401.342 139.1265 16 1.06808E-21
Social class 4477.97 62.4984 6 1.39659E-11
Present
financial
situation
4517.147 23.3216 4 0.000109204
Financial
situation in
past
4532.136 8.332 4 0.080146321
Major
problems with
education
4452.271 88.1975 14 8.30242E-13
Access to
educational
facilities
4450.442 90.0261 4 1.30006E-18
Type of
activity
3450.942 1089.527 10 9.5521E-228
Vol. 2, No. 6, June 2011

31 | P a g e
Educational Level, Province
Model 3:

where the matrix

the ijk
th
population.
Step 4: Model 3 vs Four Variable Model (Model 4)
th
Model 4
(terms added)
Raw
Deviance
(-2Log
likelihood)
Difference
in
Deviance
Difference
in df
p-value
Model 3 3155.297 0 - -
Age group 3145.734 9.5632 4 0.048464702
Gender 3133.988 21.3087 2 2.3598E-05
Ethnicity 3152.673 2.6242 7 0.917458009
Social class 3128.48 26.8167 6 0.000156716
Present
financial
situation
3146.391 8.9055 4 0.063505426
Financial
situation in
past
3152.922 2.3747 4 0.667204138
Major
problems with
education
3138.381 16.9154 14 0.260714267
Access to
educational
facilities
3140.936 14.3609 4 0.006227982
In Step 4, Gender was found significant and brought into
the model.
Educational Level, Province, Gender
Model 4:

where the matrix

the ijkl
th
population.
Step 5: Model 4 vs Five Variable Model (Model 5)
th
Model 5
(terms added)
Raw
Deviance
(-2Log
likelihood)
Difference
in
Deviance
Difference
in df
p-value
Model 4 3133.988 0 - -
Age group 3124.457 9.5309 4 0.049116179
Ethnicity 3131.679 2.3095 7 0.940746391
Social class 3108.071 25.9174 6 0.00023067
Present
financial
situation
3125.265 8.7235 4 0.068394724
Financial
situation in
past
3131.218 2.7702 4 0.596987596
Major
problems with
3117.493 16.495 14 0.284088473
education
Access to
educational
facilities
3118.975 15.0128 4 0.004674743

In Step 5, Social class was found significant and brought
into the model.
Educational Level, Province, Gender, Social class
Model 5:

where the matrix
is the set of explanatory variables

for the ijklm
th
population.
Step 6: Model 5 vs Six Variable Model (Model 6)
th
Model 6
(terms added)
Raw
Deviance
(-2Log
likelihood)
Difference
in
Deviance
Difference
in df
p-value
Model 5 3133.988 0 - -
Age group 3124.457 9.5309 4 0.049116179
Ethnicity 3131.679 2.3095 7 0.940746391
Present
financial
situation
3125.265 8.7235 4 0.068394724
Financial
situation in
past
3131.218 2.7702 4 0.596987596
Major
problems with
education
3117.493 16.495 14 0.284088473
Access to
educational
facilities
3118.975 15.0128 4 0.004674743

In Step 6, Age group was found significant and brought
into the model.
Educational Level, Province, Gender, Social class, Age group
Model 6:

where the matrix

for the ijklmn
th
population.
It was observed that addition of the remaining variables did
not improve the results. Hence the Model 6 has been identified
as the best main effect model.
C. Improving the Model
It was further investigated to determine if the addition of
two way interaction terms improved the model. The
importance of an interaction term was assessed by checking the
impact of the difference in deviance of the model.
Step 7: Model 6 Vs Model 7
Vol. 2, No. 6, June 2011

32 | P a g e
st
Most Significant 2-Way Variable to the Model
Model 7
(interaction
term added)
Raw
Deviance
Difference
in
Deviance
Difference
in df
p-value
Model 6 3108.0707 0 0 -
acti*class 3077.8665 30.2042 23 0.143629043
acti*age 3073.7177 34.353 17 0.007557411
edu*pro 2975.2098 132.8609 3 1.30766E-28
edu*gen 3098.4356 9.6351 2 0.008086575
pro*class 3072.2071 35.8636 34 0.381101733
pro*age 3076.8779 31.1928 33 0.557285717
gen*class 3095.0509 13.0198 6 0.042722597
gen*age 3098.6612 9.4095 8 0.308936646
class*age 3093.4829 14.5878 16 0.555009994
eth*mprob 3055.5054 52.5653 45 0.204371948
eth*faci 3094.1994 13.8713 21 0.875063689
fin *mprob 3041.6852 66.3855 42 0.009620228
fin *faci 3079.7823 28.2884 16 0.029200335
finp*mprob 3056.0176 52.0531 43 0.162103659
finp*faci 3084.6956 23.3751 16 0.104069216
mprob *age 3051.2768 56.7939 42 0.063390185
mprob*faci 3052.4179 55.6528 44 0.111883885
faci*gen 3093.6093 14.4614 8 0.070504003
faci*class 3084.5775 23.4932 14 0.052702948
faci*age 3082.1476 25.9231 16 0.055119929

The interaction between Education Level and Province
has been found to have an effect on the model and hence added
to the model.
Educational Level, Province, Gender, Social class, Age group,
Edu*Pro
Model 7:

where the matrix

for the ijklmn
th
population and
is the interaction effect of

Edu*Pro.
nd
Model 8
(interaction
term added)
Raw
Deviance
Difference
in
Deviance
Difference
in df
p-value
Model 8 2975.2098 0 0 -
acti*age 2942.6119 32.5979 21 0.05087
edu*gen 2970.8828 4.327 2 0.114922
edu*class 2942.4092 32.8006 3 3.55E-07
pro*gen 2962.5509 12.6589 11 0.316203
pro*class 2940.9991 34.2107 27 0.160007
gen*class 2960.1934 15.0164 4 0.004667
gen*age 2961.1957 14.0141 8 0.081399
class*age 2954.729 20.4808 15 0.154254
eth*mprob 2922.7682 52.4416 42 0.129663
eth*faci 2960.932 14.2778 18 0.710811
fin *mprob 2906.8217 68.3881 40 0.003421
fin *faci 2945.8832 29.3266 14 0.009437
finp*age 2952.1219 23.0879 15 0.082291
finp*mprob 2852.9844 122.2254 42 9.18E-10
finp*faci 2952.4005 22.8093 15 0.088274
mprob *age 2913.2762 61.9336 42 0.024235
mprob*faci 2923.8426 51.3672 47 0.306598
faci*class 2955.1103 20.0995 13 0.092757
faci*age 2944.8296 30.3802 15 0.010622

In this step, the interaction between Financial Situation in
Past (Finp) and Major Problems with Education (Mprob) has
been found to have an effect on the model and hence added to
the model.
Edu*Pro, Finp*Mprob
Model 8:

where the matrix

for the ijklmn
th
population, the matrix
is the interaction
effect of Edu*Pro and the matrix
is the interaction effect

of Finp*Mprob.
rd
Model 9
(interaction
term added)
Raw
Deviance
Difference
in
Deviance
Difference
in df
p-value
Model 9 4027.772
acti*age 2909.589 19.7647 17 0.286441
edu*gen 2925.853 3.5005 5 0.623312
edu*class 2894.889 34.4644 15 0.002929
pro*class 2896.654 32.6994 32 0.432465
pro*age 2908.125 21.2286 31 0.905749
gen*class 2913.901 15.4524 6 0.017015
gen*age 2928.941 0.413 4 0.981399
class*age 2922.754 6.6001 12 0.882871
eth*acti 2899.069 30.2848 19 0.048284
eth*edu 2915.022 14.3318 3 0.002487
fin*acti 2907.075 22.2787 6 0.001078
fin *edu 2918.808 10.5456 11 0.482078
fin *pro 2892.44 36.9141 33 0.292752
finp*acti 2909.158 20.1959 16 0.211527
finp*edu 2919.167 10.1871 14 0.74838
mprob *edu 2892.521 36.8324 33 0.295967
mprob *pro 2788.718 140.6361 87 0.00024
mprob *gen 2895.19 34.164 34 0.459862
faci*class 2908.176 21.1779 13 0.069484
faci*age 2909.906 19.4477 11 0.053516
In this step, the interaction between Major Problems with
Education (Mprob) and Province (Pro) has been found to
have an effect on the model and hence added to the model.
Edu*Pro, Finp*Mprob, Mprob*Pro
Model 9:

Vol. 2, No. 6, June 2011

33 | P a g e
where the matrix

the ijklmnth population, the matrix
is the interaction effect

of Edu*Pro, the matrix

Finp*Mprob and the matrix

Mprob*Pro.
After Step 9 no more significant two-way terms were
revealed.
D. Checking the Adequacy of the Best 2-Way Interaction
Model
Goodness of Fit: Hypothesis Testing with Deviance
Statistics
H0: No lack of fit
H1: There is some lack of fit
Table 11: Goodness of Fit Testing
Chi-Square df p-value
Deviance 2145.881 2930 1.000

According to the p-value, it can be concluded that the Null
Hypothesis is not rejected. Hence, the test is the proof that
there is no lack of fit of the model or the model developed fits
the data well.
E. Classification Table
The classification table was used to evaluate the predictive
accuracy of the regression model.
Table 12: Classification Table for Model 9
OBSERVED
PREDICTED
NO
DESIRE
TECHNICAL/
VOCATIONAL
EDUCATION
UNIVERSITY/
HIGHER
EDUCATION
%
CORRECT
NO DESIRE 641 141 51 77.0%
TECHNICAL/
VOCATIONAL
EDUCATION
249 304 89 47.4%
UNIVERSITY
/HIGHER
EDUCATION
46 45 519 85.1%
OVERALL
PERCENTAGE
44.9% 23.5% 31.6% 70.2%

From the results shown in Table 12, it can be seen that
Model 9 has the accuracy of more than 70 percent.
F. Implementation of Data Mining Techniques
Finally the model (Model 9) developed using Univariate
analysis was used to develop a data mining model. This data
mining model can predict the Further Educational Desire in a
youth from other attributes discussed above.
Construction of Decision Tree
A decision tree was constructed using the attributes
identified as significant in the Univariate analysis. The ordering
of attributes in the decision tree was also as determined in the
statistical analysis. Table 13 shows the attributes and the no. of
levels of each attribute. Type of Further Education Desire is
the class attribute.
Table 13: List of Attributes used in the Decision Tree
Seq. No Attribute Name No. of Levels
1 Type of Activity 7
2 Educational Level 3
3 Province 9
4 Gender 2
5 Social Class 4
6 Age group 3
7 Financial Situation in Past 3
8 Major Problems with Education 8
9 Type of Further Education Desire 3

Figure 1 shows the portion of the decision tree thus
constructed.

Figure 1: Portion of the Decision Tree
Figure 2 shows a sample classification rule set developed
using the Decision Tree shown in Figure 1.

Vol. 2, No. 6, June 2011

34 | P a g e
Rule 1: Type of Activity=Permanently Employed ^ Educational
Level=No Schooling/Grade 1-5 ^ Province=Western ^
Gender=Male ^ Social Class=Middle Class ^ Age Group=20-24
yrs No Desire
Level=No Schooling/Grade 1-5 ^ Province=Western ^
Gender=Male ^ Social Class=Working Class ^ Age Group=15-
19 yrs No Desire
Level=No Schooling/Grade 1-5 ^ Province=Central ^
Gender=Male ^ Social Class=Working Class ^ Age Group=20-
24 yrs No Desire
Level=No Schooling/Grade 1-5 ^ Province=Central ^
Gender=Female ^ Social Class=Working Class ^ Age
Group=20-24 yrs No Desire
Level=No Schooling/Grade 1-5 ^ Province=Uva ^ Gender=Male
^ Social Class=Working Class ^ Age Group=24-29 yrs
Technical/Vocational Education
Figure 2: Sample Rule Set
G. Implementation
A software tool was developed using Visual Basic to
implement the rule set developed above. Figure 3 shows the
interface of the software tool developed.

Figure 3: Interface of the Software Tool Developed
H. Evaluation
The system developed was tested using a random set of
data containing 485 test records. The selected random set was
divided into four test sets each test set containing around 125
records. Table 14 shows the classification table obtained by
inputting data into the application.
Table 14: Classification Table Obtained from Test
OBSERVED
PREDICTED
TECHNICAL /
VOCATIONAL
EDUCATION
UNIVERSITY/
HIGHER
EDUCATION
NO DESIRE

TECHNICAL/
VOCATIONAL
EDUCATION
21 13 19 22 3 5 5 7 13 17 16 7
UNIVERSITY
/HIGHER
EDUCATION
0 4 3 7 30 23 35 26 3 5 2 2
NO DESIRE

10 5 11 8 4 7 2 0 32 51 27 40

Table 15 shows the Confusion Matrices for various
adjusted outcomes.
Table 15 (a): Confusion Matrix Adjusted for No Desire
Predicted
No Desire
Have a
Desire
Observed
Data Set 1
No Desire 36 14
Have a Desire 16 54
Data Set 2
No Desire 51 12
Have a Desire 22 45
Data Set 3
No Desire 27 13
Have a Desire 18 62
Data Set 4
No Desire 40 8
Have a Desire 9 62
Table 15 (b): Confusion Matrix Adjusted for Uni/Higher Education
Predicted
Uni/Higher Not So
Observed
Data Set 1 Uni/Higher 30 3
Not So 7 76
Not So 12 86
Not So
7 73
Not So 7 77
Table 15 (c): Confusion Matrix Adjusted for Tech/Voc. Education
Predicted
Tech/Voc Not So
Observed
Data Set 1
Tech/Voc 21 16
Not So 10 69
Data Set 2
Tech/Voc 13 22
Not So 9 86
Data Set 3
Tech/Voc 19 21
Not So 14 66
Data Set 4
Tech/Voc 22 14
Not So 15 68

Table 16 shows the resulting measures obtained from the
tests.
Table 16: Measures Obtained from Tests
Tech/Voc Uni/High No
Desire
Avg.
Data Set 1

TPR 0.567568 0.909091 0. 72 0. 7322
FPR 0.126582 0.084337 0.228571 0.1464
Accuracy 0.775862 0.913793 0. 75 0.8132
Data Set 2
TPR 0.371429 0. 71875 0.809524 0.6332
FPR 0.094737 0.122449 0.328358 0.1818
Accuracy 0.761538 0.838462 0. 738462 0. 7794

Data Set 3
TPR 0.475 0.875 0.675 0.675
FPR 0.175 0.0875 0.225 0.1625
Accuracy 0. 708333 0.9 0. 741667 0. 7833
Data Set 4
TPR 0.611111 0. 742857 0.833333 0. 7291
FPR 0.180723 0.083333 0.126761 0.1302
Accuracy 0. 756303 0.865546 0.857143 0.8263
Overall
TPR 0.50628 0.81142 0. 75946 0.6923
FPR 0.14426 0.0944 0.22717 0.1552
Accuracy 0. 75051 0.87945 0. 77182 0.8005

From Table 16, it can be seen that the overall accuracy of
the system is above 80 percent. Figure 7 shows the Receiver-
Operator Characteristic (ROC) curve drawn from the data in
Figure 4.
Vol. 2, No. 6, June 2011

35 | P a g e

Figure 4: Receiver Operator Characteristic Curve for Model
From the ROC curve, it can be seen that the results
concentrate towards the upper left hand corner. This is the
proof that the accuracy of the Data Mining model is acceptable
as pure random guess would lie along the diagonal line.
IV. CONCLUSIONS
This paper presented the results of the research carried out
to find the factors on which the Educational Desires of Sri
Lankan Youth depends. The research found that the
Educational Desires of the Youth could be predicated through
the combination of several social factors. The findings of the
research were finally used to design a data mining model for
the predication of the Educational Desires of the Youth. This
model can be used by decision makers in dealing with issues
concerning youth especially their further educational
requirements.
REFERENCES
[1] Hettige ST, Mayer M, Sri Lankan Youth Challenges and Responses,
Friedrich Ebert Stiftung, Colombo, Sri Lanka, (2002).
[2] Fernando HC, Tissera WMR, Athauda RI, Gaining Insights to the Tea
Industry of Sri Lanka using Data Mining, Proceedings of the
International Multi Conference of Engineers and Computer Scientists
(IMECS 2008) Vol I, Hong Kong, (2008).
[3] Research Methods Knowledge Base, Descriptive Statistics,
http://www.socialresearchmethods.net/kb/ statdesc.php [Accessed
on 08/04/2010]
[4] Babbie ER, The Practice of Social Research", 12
th
edition, Wadsworth
Publishing, (2009).
[5] Cooper RA, Weekes AJ, Data, Models, and Statistical Analysis,
Rowman & Littlefield Publishers, Inc., USA, (1983).
[6] Wikipedia, Pearson's Chi-Square Test,
http://en.wikipedia.org/wiki/Pearson's_chi-square_test [Accessed on
08/04/2010]
[7] Kinley L, "Small-sample comparisons of exact levels for chi-squared
goodness-of-fit statistics", Journal of the American Statistical
Association 73 (362): pp. 253263. (1978).
[8] Mehta CR, Patel NR, Tsiatis AA, "Exact Significance Testing to
Establish Treatment Equivalence with Ordered Categorical Data".
Biometrics 40 (3): pp. 819825, (1984).
[9] Weisstein EW, "Fisher's Exact Test." From MathWorld--A Wolfram
Web Resource. http://mathworld.wolfram.com/FishersExactTest.html
[Accessed on 08/04/2010]
[10] Sarkar SK, Midi H, Optimization Techniques for Variable Selection in
Binary Logistic Regression Model Applied to Desire for Children Data
Journal of Mathematics and Statistics 5 (4): pp. 387-394, (2009).
[11] Pahwa P, Karunanayake CP, Modeling of Longitudinal Polytomous
Outcome from Complex Survey Data - Application to Investigate an
Association between Mental Distress and Non-Malignant Respiratory
Diseases, BMC Medical Research Methodology, Vol 9, (2009).
AUTHORS PROFILE

Mohamed Fazil Mohamed Firdhous is a senior lecturer
attached to the Faculty of Information Technology of the
University of Moratuwa, Sri Lanka. He received his BSc
Eng., MSc and MBA degrees from the University of
Moratuwa, Sri Lanka, Nanyang Technological University,
Singapore and University of Colombo Sri Lanka
respectively. In addition to his academic qualifications, he is
a Chartered Engineer and a Corporate Member of the Institution of Engineers,
Sri Lanka, the Institution of Engineering and Technology, United Kingdom and
the International Association of Engineers. Mohamed Firdhous has several
years of industry, academic and research experience in Sri Lanka, Singapore
and the United States of America.

Ravindi Jayasundara is a lecturer attached to the
Department of Mathematics, Faculty of Engineering of
the University of Sri Lanka. She completed the BSc
(Hons) degree in Statistics with Second Class (Upper
Division) at the University of Colombo, Sri Lanka and
MSc in Operations Research at the University of
Moratuwa, Sri Lanka. Ravindis research interests are in
statistics, operations research and data mining.
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
T
P
R

FPR
Vol. 2, No. 6, 2011
36 | P a g e
An Authorization Mechanism for Access Control of
Resources in the Web Services Paradigm

Rajender Nath
Department of Computer Science & Applications,
Kurukshetra University, Kurukshetra,
Haryana, India
Gulshan Ahuja
Tilak Raj Chadha Institute of Management. &
Technology,
Yamunanagar,
Haryana, India

Abstract With the increase in web based enterprise services,
there is an increasing trend among business enterprises to
migrate to web services platform. Web services paradigm poses a
number of new security challenges, which can only be realized by
developing effective access control models. The fact that the
enterprises allow access to the resources through web services
requires development of access control models that can capture
relevant information about a service requester at the time of
access request and incorporate this information for making
effective access control decisions. Researchers have addressed
many issues related to authentication and authorization of web
services requests for accessing resources, but the issues related to
authorization work and identity based access are still poorly
addressed. Authors of this paper focus on providing an extended
approach to capture relevant information about a service
requester and establish a certain level of trust so that amount of
authorization work required for accessing any resource is
reduced and service requests are served in an efficient manner.
Compared with existing access control mechanisms, the proposed
mechanism has reduced the amount of authorization work
required for accessing resources across varied domains.
Keywords- Web Services; Access Control; Authentication;
Authorization.
I. INTRODUCTION
Web services are loosely coupled applications which use
protocols like SOAP (Simple Object Access Protocol) for
message exchange, WSDL (Web Services Description
Language) for interface description, and UDDI (Universal
Description, Discovery, and Integration) for service discovery
and communication across different security domains in a
distributed environment. Currently, the most of the business
enterprises try to align their business processes with the IT
infrastructure support by migrating towards web service
oriented architecture (WSOA). The use of identity certificates
issued by trusted third parties to assign roles to users has
already been addressed. Looking at the available specifications
in the web service security areas like WS-Security, WS-Trust,
SAML, XACML, it is understandable that developers give
more emphasis on securing web services and often ignore
access to the resources. In web services paradigm, a web
service operation must be granted access based on the
permissions and by also considering the web service invocation
parameters. The aforementioned models consider an action to
be a limited set of operations like read, write, execute etc. and
do not consider use of invocation parameters for making access
control decisions. The problem with such mechanisms is that a
significant amount of authorization work is required towards
verifying the credentials for each service request. Every time, a
service requester sends a request to the service provider for a
desired service, its identity certificates issued by the third party
are verified and access control policies are evaluated before
granting access to any resource. This makes the task of
enforcement of access control burdensome and time
consuming. To address the above problems, we propose an
authorization mechanism for access control of resources in web
services paradigm. The proposed mechanism considers service
request invocation parameters to derive new set of parameters
and uses access percentile to access any web service. Our
approach presents a new means to realize the authorization in
an efficient manner. The efficiency of proposed mechanism is
considered in terms of reduced authorization work and time to
serve any request. Our approach makes use of generalized
parameters, a subset of which have to be supplied by the
service requester at the time of making a request to access any
resource using a web service. Another subset of parameters can
be set by system administrator as per the requirements of
domain in which the access is to be granted. These parameters
are used to check whether access to required web service can
be granted or not. The proposed approach also deals with the
concept of identity tracking so that the service requests
belonging to a particular service requester can be logged and
the subsequent requests from the same service requester can be
served with less frequency of validation checks.
This paper has been organized into IX sections. Section II
discusses the state of access control for resources in web
services environment. Section III discusses the related work.
Section IV discusses problem formulation. Section V proposes
an authorization mechanism for access control of resources in
web services paradigm. Section VI presents a n example case.
Section VII presents a contrast of our approach vs. existing
access control approaches. Section VIII presents performance
evaluation results. Section IX concludes the paper and presents
future scope and improvements.
II. AN OVERVIEW OF ACCESS CONTROL FOR RESOURCES

Vol. 2, No. 6, 2011
37 | P a g e
In web services environment, service providers willing to
provide services publish their web services interfaces in public
registries like UDDI. Service requesters must search and locate
the web services from this publicly accessible database. Once
an appropriate service has been found, the service requester can
send the request to service provider for the service by
encapsulating the requesters information in SOAP message
using HTTP protocol. The service provider draws out the
invoking parameters from SOAP messages and initiates the
process to execute the appropriate methods for granting access
to the required resources. Service requester and service
provider may not belong to the same domain. In this case,
every time service requester needs to access a resource in
different domain, it has to provide its credentials for
authentication and authorization. As the number of available
resources and requesting entities increases, the amount of
authorization work also significantly increases. The existing
role based access control (RBAC) models solve this problem
by providing mapping between users and roles, permissions
and roles. In order to guarantee more security for resource
access, researchers have developed techniques like attribute
certificates (AC) based on privilege management infrastructure
(PMI) [1]. PMI allows including and revoking attributes and
can contain information about the privileges or roles of a user.
Every time a service requester needs to access a resource, it
must provide its identity and attribute certificates, which must
be validated before granting access permission. The access to
resources can be made more secure by establishing certain
level of trust and the amount of authorization work can be
reduced by using the established trust level.
III. RELATED WORK-ACCESS CONTROL MODELS
In 1969, the basics about access control were described
formally for the first time [3] and the concepts of subjects and
objects were introduced. To control the access to objects,
security policies were introduced. To access any object, the
security level of subject has to be more than the security level
of the specified object. These types of models are termed as
lattice based access control (LBAC). LBAC models are not
scalable and their use is restricted in specific scenarios. To
overcome these limitations, Sandhu et al. [4] in 1996 proposed
the notion of Role-Based Access Control (RBAC) for
enterprise applications as a powerful and generalized approach
for access of enterprise resources. Later, Sanhu presented RBAC
[5], which defines four reference models for RBAC as follows:
RBAC0 model defines a basic RBAC system, RBAC1 model
augments RBAC0 with role hierarchies, RBAC2 adds constraints
to RBAC0, while RBAC3 combines RBAC1 and RBAC2.
ARBAC97 (administrative RBAC97) [6] have three
components: 1) URA97, which is concerned with userrole
assignment; 2) PRA97, which focuses on permission role
assignment and is a dual of URA97; 3) RRA97, which deals with
role-role assignment
With the advent of web services [7, 8], the access control
mechanisms became more complex. The most significant work
in relation to access control in the web services was proposed
by Damiani et al. [9].
Freudenthal et al. [10] presented a distributed role-based
access control model (dRBAC) for dynamic coalition
environments. dRBAC presents a scalable, decentralized trust-
management and access control mechanism for systems that
span multiple administrative domains, which utilizes PKI
identities to define trust domains, roles to define controlled
activities, and role delegation across domains to represent
permissions to these activities.
Xu et al. [11] proposed a context aware access control
model for web services. Miao Liu et al. [12] proposed an
attribute and role based access control model for web services.
Yizho Zhao et al. [13] proposed a model a flexible role and
resource based access control model (RRBAC) which can
support open and distributed environments. RRBAC is suitable
for multiple security domains with different applications.
RRBAC adopts role directory structure instead of role
hierarchy but do not consider about authorization of requests
using services.
Xu Feng et al.[14] proposed a security architecture for web
services and employs a SOAP proxy for processing of SOAP
messages. R Bhatti et al.[15] highlights a trust based context
aware access control model for web services. Christian Emig et
al. [16] proposed an access control metamodel for web service
oriented architecture.
IV. PROBLEM FORMULATION
A significant amount of research has been carried towards
access control in web services paradigm. We have based our
research on RBAC because it brings much convenience to
system administrators and service requesters by introducing the
concepts of role in the system. However, RBAC do not fit well
for multi security domains and cannot support trust
management (TM) [17] and trust negotiation (TN) [18]. Many
papers have proposed frameworks to integrate RBAC with web
services and addressed issues like security, context based
access, attribute based access, trust establishment etc. With the
increase in number of service requesters and available
resources, the amount of authorization work also increases.
Security architecture [14] for web services employs a
SOAP proxy for processing of SOAP messages. The
architecture considers only parsing of messages and realization
of a role for service access. It lacks to consider about the
amount of authorization work required for web service access.
The work presented by R Bhatti et al.[15] highlights a trust
based context aware access control model for web services but
do not consider how trust level can be established and used for
access control. Christian Emig et al. [16] proposed an access
control metamodel for web service oriented architecture. Their
approach proposes to use an authorization verification service
but do not highlight that how service requesters parameters can
be utilized for reduced authorization effort.
We derive a motivation here to introduce the mechanism to
use service invocation parameters like requesters credentials -
identity certificates, digital signatures, object identifier etc. to
establish a level of trust and conserve identity of service
requests for subsequent accesses. The proposed approach uses
the parameters supplied during service request and derives a
new set of parameters for making access decision.
Vol. 2, No. 6, 2011
38 | P a g e
V. THE PROPOSED WORK: AN AUTHORIZATION
MECHANISM FOR ACCESS CONTROL OF RESOURCES IN WEB
SERVICES PARADIGM
Fig. 1, presents a model for implementing the proposed
approach. The authorization approach defined by the authors of
this paper uses a number of modules. The purpose and use of
each module is defined as follows
A. Details about modules
1) Service Initiator Module (SIM)
The model comprises of Service Initiator Module (SIM),
which is the main module for initiating the process for
authorization and allowing access to the required web service.
2) Identity Tracker and Verifier (IDTV)
Identity Tracker and Verifier (IDTV) module is responsible
for computation and assignment of access percentile and
verification of the attribute certificates. Service Provider (SP)
maintains a local database of Service Requesters for
maintaining information about their identity.
3) Role Based Access Control (RBAC) processor
This module is composed of Policy Evaluator (PE) for
evaluation of policies and Role Mapper (RM) for mapping
service requester to assigned role. RBAC module checks
service requests against specified policies from the policy store
and invokes appropriate role for accessing a resource. There are
three functions to perform the mapping:
(service requester SRi, role Ri): to map a service requester
SRi to a role Ri.
(role Ri, resource RSi): role assignment, to map a role to a
certain resource, i.e., relating the role Ri with resource RSi
(resource RSi, permission set PSi): resource assignment, to
map a resource to a certain permission set, i.e.,
corresponding the resource RSi , with the permission set
PSi.
4) History Buffer (HB)
Is used for temporary conservation of information about
past accesses to the resources.
B. Implementation Details for Authorization Mechanism
Service Provider (SP) maintains a local database of Service
Requesters for maintaining information about their identity. To
access any resource through a web service, the service
requester sends a SOAP message, which includes various
pieces of information about service requester to the service
provider. On service providers end, Service Initiator Module
(SIM) processes it before passing it to the required web service
for allowing access to the resource. Each web service is
assigned an access percentile which specifies a value that must
be possessed by any service request for accessing a resource
through a web service.
The access percentile for each web service request is
computed based on number of parameters supplied during
service request. These parameters in the form of records are
also maintained in history buffer. Table 2 represents a set of
records for calculation of access percentile.

Fig. 1. Authorization architecture for access of resources using web services.

TABLE 2. RECORDS FOR ACCESS PERCENTILE
Access
Parameters

Service
Requesters
IP_Addr Id _Cert Service
_ Info
Time_
to_
Live
SR
1

192.168.0.
15
AXZAB--
---YPZ
WS
1
,
WS
3

TTL
1

SR
2

192.157.1
2.17
QPOST---
--ORP
WS
6
,
WS
7,

WS
9

TTL
2

Columns in the table 2 represent access parameters
IP_Addr, Id_cert, Service_Info, Time_to_Live fields. IP_Addr
field maintains information about IP address from where the
previous requests had originated. Id_Cert field maintains
information about id certificate of service requester.
Service_Info field maintains information about which all
services have already been accessed by a service requester.
Time_to_Live field states the time period for which the record
in the history buffer remains alive.
The set of parameters P
r
S is chosen as a generalized set,
which can be changed as per requirements. Based upon these
access parameters, the access percentile is computed as
described below.
Each individual parameter is assigned a weight W
i,
which is
assigned by the system administrator.
IP_Addr field is assigned a weight W
IP
according to the type
of IP. For e.g static IP address may be assigned more weight
than dynamic IP address.
ID_Cert field is assigned weight W
ID
as per the trust and
reputation of the CA.
Service_Info field is assigned a weight W
IS.
This field keeps
track of which all services have already been accessed by a
service requester in a given time frame. More is the number of

Service Provider
Service Requester (SR)

Service Initiator Module
(SIM)

CA
Identity
Tracker (IDTV)
Policy
Evaluator(PE) RBAC
PRO
CESSOR

Role Mapper
(RM)
WS1

WS2
:
:
WSn
History
Buffer
H
B
Local
Registry
Policy
Store
Vol. 2, No. 6, 2011
39 | P a g e
times a given service WS
i
has been accessed the greater is the
value of weight W
IS.
Time_to_Live field is assigned a weight W
TTL.
The value of
TTL field decides that for how long the record will remain
in the history buffer. The value for the weight W
TTL
is
computed as follows:
W
TTL
= RTETTL
Where RTE is the remaining time within which the record
will expire in the History Buffer (HB) and TTL is the total time
to live in the HB. The value of weight will be more for
frequently accessed services and will decrease if a service is
accessed after some time gap.
Now, the access percentile for any service requester
SR
i
,
Access_Percentile
can be calculated as

SR
i
,
Access_Percentile
= X
i,,j
* W
IP
+ X
i,,j
* W
IP
+ X
i,,j
* W
IS
+ X
i,,j
*
W
TTL.

X
i,,j
is a presence matrix where i represents any service
requester SR
i
value of j ranges as 1 j N
AP,
where N
AP
is the
number of access parameters stored in HB.
SR
i
<IP_Addr, Id _Cert, Service_Info, Time_ to_ Live>

<0,0,0,0> iff SR
i
does not exist in HB.

<1,1,1,1> iff SR
i
exists in HB.

The SOAP message is checked for identity certificates of
the service requester. The identity certificates can be obtained
from a certified authority (CA) who is a trusted server in the
domain, which generates the X.509 certificates. Each domain
can have its own CA or if two or more domains trust each other
then they can use a single shared CA. The primary
responsibilities of CA include: Identification of clients
requesting for issue of certificates, to Issue, archive and delete
the certificates, to maintain a namespace of unique names for
the certificate owners.
Reference Fig. 1., If SOAP message does not contain
identity certificates the request is rejected otherwise local
database of service requesters is checked to find out that
requesting service requester already exist in the registry or not.
If service requester exists in the registry, its access percentile is
computed using stored parameters from HB, to check whether
it can call the required web service for accessing a resource.
Any service requester serving the request for the very first time
will not be having any record in the history buffer. In this case,
the service requester has to go through a complete cycle of
validation process for granting access to any web resource
through a web service.
Once a service requester has been authenticated and
authorized, information about its service request parameters is
recorded in the history buffer and in the local registry of
registered users. The next time, if the same service requester
needs to access a resource; its information is checked in the
local registry, because the same service requester had already
been authenticated and authorized, its record will be found in
the local registry, after that the history buffer is accessed to
compute access percentile based on available parameters.
If the service requesters access percentile is equal or more
than the permitted access percentile, the request is passed on to
RM for role mapping, otherwise the request is passed to PE for
policy evaluation to cross verify against specified policies and
thereafter the required resource can be accessed using the
called web service. In case the computed access percentile is
less than the permitted level of access percentile for required
web service, the access is denied.
C. The algorithms for the proposed mechanism are described
in Fig. 2.1, Fig. 2.1(a) and Fig. 2.2
The Service Provider (SP) will identify the decryption key
and the elements to decrypt. Thereafter, each encrypted
element is decrypted. If decryption fails a fault message is sent
to the service requester. SIM carries the validity of SOAP
message using the following method.
InitiateRequest( )
Input: Parameters from service requester
Output: allow,deny
{
If SR
IP
Permitted
IP
{
if SR
i
,
Time_Stamp
Permitted
Time_Stamp
{
if ( VerifySignature() )
{
If ( ProcessRequest() )
return true;
else
return false;
}
else
return false;
}
else return false;
}
else return false;
}
Fig. 2.1 Validation of SOAP message

VerifySignature( )
Input: - Parameters from service requester
Output:- true,false
{
Use public key (SR
PK
) of Service Requester

Digital Signature (SR
PK
) message digest

Apply hash message digest.

Compare and match
}
Vol. 2, No. 6, 2011
40 | P a g e
Fig. 2.1(a) Verification of digital signature
Verification is carried to check whether the SOAP message
and enclosed credentials have come from a valid service
requester or not. When service provider receives a signed
SOAP message, it will initiate the verification process. The
public key of the service requester is used to decrypt the digital
signature and retrieve the message digest. The hash algorithm
is applied again to the digital contents to generate another
message digest. These two message digests are compared and
if they match verification is successful. If there were any
changes in the digital contents the resultant message digest
would differ from the original one and the verification would
fail.
ProcessRequest( )
Input:- Parameters from HB, Records from Local Database,
Parameters from service requester
Output:-true,false
{
If SR
i
,
IDCert
exist in SOAP message
Check Local registry of SR
i
where 1 i n
If SR
i

found
SR
i
,
Access_Percentile
=ComputeAccessPercentile(SR
i
)
If (SR
i
,
Access Percentile
> WS
i
,
Allowed_Percentile
)
Pass SOAP message RM
return true;
end if
elseif (ValidateCertificate(SR
i
,
IDCert
))
Pass SOAP message RBAC Processor
Update HB
else return false;
else return false;
}
Fig. 2.2 Processing of Service Requesters Information
Once the digital signatures have been verified, the SIM
proceeds to check whether identity certificates are contained
within message or not.
VI. AN EXAMPLE CASE
Suppose, SecDom, be a security domain which offers
online services. O
1
is the owner of a database resource system,
which can provide storage service for public. O
1
registers itself
for providing access to its resource through SecDom domain.
After the registration, the database storage system becomes a
valid resource in SecDom and the registered resource is made
available through a web service. O
1
may specify many roles
which are associated with different privileges. The roles
include user, guest, associate_partner, admin etc. For instance,
an associate partner can own the privileges of {access, read,
upgrade, downgrade, delete}. Suppose a service requester
wants to access a resource in SecDom domain, he authenticates
himself into the domain and sends a SOAP message to the
service provider, which includes service invocation parameters
like requesters credentials - identity certificates and digital
signatures, object identifier, security token. On service
providers end, the service request is processed for verification
of supplied credentials. After that, the identity certificates are
checked and validation of the certificates is carried. This is
done only incase of new users or users whose request is made
after a long time gap. For all those users who make request
within a specific time interval, the validation of credentials can
be skipped. In this way the time required to grant access to the
specified resource is reduced and it also results in better
bandwidth utilization.
ACL description
ACL offers a way to store the access control information.
The format of ACL is chosen as follows:
acl = (role) : (resource) : (permission set)

Table 1. ACL table at service providers end
Assigned Role Resource Type Permission Set
User database storage
system
{access, read,
downgrade,upgrade}
Guest database storage
system
{access}
associate_partner database storage
system
{access, read, upgrade,
downgrade, delete}
Admin database storage
system
{access, read, write,
delete, upgrade,
downgrade, create
directory, delete
directory, create file,
delete file }
VII. COMPARISON WITH EXISTING ACCESS CONTROL
APPROACHES
In this section, we present a comparison of our approach
with Existing Access Control models based on Web Services.
As per Table 3, we can find characteristics and features
supported in our approach.
Table 3. Proposed Approach vs. Access Control Models for Web Services
Sr.
No.
Feature/Characteristic Web
Services
& Role
Based
Access
Models
Our
Approach
1. Support for distributed
environment

2. Support for web service
based access

3. Resource request
management

4. Trust level assignment to
web services

5. Identity tracking &
Management

6. Reduced authorization work
for access requests.

The above comparison reflects that our approach uses a
mechanism to assign access percentile to web services and uses
it to make access control decisions. The concept of identity
tracking is used so that subsequent requests from same service
requester can be served efficiently. The number of times the
validation is to be carried for granting access to resources is
reduced which results in reduced authorization work during
serving of access requests.
Vol. 2, No. 6, 2011
41 | P a g e
VIII. PERFORMANCE EVALUATION
We consider two different scenarios for evaluation of
performance of our approach.
In first scenario, we calculate time required to grant access
to any resource. The measurement is carried in two ways. 1)
For users who have had already requested for a resource and 2)
For those users who requested for a resource for the first time.
In both the cases, without our approach the time required for n
service requesters would be
Following terms are defined and used for evaluation.
Let T
REQ
the time required to service any request made
by any service requester SR
i.
The time T
REQ,
includes the round
trip communication time i.e. from service requester (SR)
service provider (SP) and vice versa.
Let T
D, CERT
Time required to validate credentials of
service requester from CA.
T
D, PE
Time required to evaluate associated policies.
T
D, RM
Time required to map to an appropriate role.
T
SRi
Total time required to serve one service request
from any service requester SR
i.
Let N be the maximum number of service requests which
can be made by any service requester SR
i.
TA
SRi
Total time required serve N service requests from
same service requester SR
i

T
GT
Grand Total time required for granting access to all
service requesters can be defined as
Let n be the maximum number of service requesters who
can make a request for a resource.
We compute grand total time for first scenario as
T
SRi
= T
D, CERT
+ T
D, PE
+ T
D, RM

TA
SRi
= N * T
SRi

T
GT
= TA
SRi
whereSR
i
1in
In second scenario, we calculate time and effort required to
grant access to any resource using our approach. The
measurement is carried for users who have had already
requested for a resource and for those users who requested for a
resource for the first time.
Following terms are defined further and used for
evaluation.
Let P be the number of times access percentile of any SR
i
is
sufficient to grant access and Q be the remaining attempts for
which complete validation is required.
Also P+Q N and where P Q for all service requests
made within allowed time period.
So the time required to service P service requests from
i
within a specified time frame would
be
TP
SRi =
P*T
D, RM
And the time required to service Q service requests from
i
would be
TQ
SRi =
Q* T
SRi
Grand total time for second scenario would be
T
GT
=

(TP
SRi
+ TQ
SRi
) whereSR
i
1in
The above calculations reveal that once a service requester
has been validated, for subsequent requests the amount of
authorization work will be reduced considerably.
IX. CONCLUSION
In this paper, an approach has been proposed to handle
authorization of service requests in web services paradigm.
The approach makes use of the concept of identity tracking and
access percentile for invoking any service. The paper has
analyzed how the service request invocation parameters can be
utilized for making efficient authorization verification. The
performance of the proposed approach has been tested by
taking a user case scenario. The results have indicated a good
improvement in the performance during authorization of the
request. The legacy systems that depend upon composition of
services for accomplishing a particular task may require
wrapping of these services through a common service interface,
which may enable business process integration in an easy
manner.
REFERENCES
[1] Jordi Forne, M. Francisca Hinarejos,Webbased Authorization based on
X.509 Privilege Management Infrastructure, IEEE 2003, pp 565-568.
[2] David F. Ferraiolo, Janet A. Cugini, D. Richard Kuhn, Role Based
Access Control: Features and Motivations, Proceedings of 11th Annual
Conferences, December 1995, pp 241-248.
[3] B.W. Lampson,Dynamic protection structures,AFIPS conference
proceedings, 1969, pp. 27-38.
[4] Ravi S. Sandhu, Edward J. Coyne, Role Based Access Control
Models, IEEE Computer, Feb 1996, pp. 38-47.
[5] Ravi S. Sandhu, Role Hierarchies and Constraints for Lattice Based
Access Controls, Proceedings of the 4th European Symposium on
Research in Computing Security, September 1996.
[6] R. S. Sandhu, V. Bhamidiparti, Q. Munawer. The ARBAC97 model for
role-based administrator of roles. ACM Transactions on Information and
System Security. Vol. 2, No. 1, Feb, 1999, pp: 105-135.
[7] Web Services Architecture, http://www.w3.org/TR/ws-arch/
[8] Web Services Activity, http://www.w3.org/2002/ws/
[9] E. Damiani, S.D.C. Vimercati, S. Paraboschi and P. Samarati, Fine
Grained Access Control for SOAP e-services, Proceedings of 10th
International Conference on World Wide Web , Hong Kong, 2001, pp.
504-513.
[10] E. Freudenthal, T. Pesin, L. Port, E. Keenan, V. Karamcheti. dRBAC:
disttibuted Role-Based Access Control for dynamic coalition
environments. In proceeding of 21th International Conference on
Distributed Computing Systems. IEEE Press, 2002, pp. 128-137.
[11] Xu F., Xie J., Huang H., and Xie L., Context Aware Role Based Access
Control Model for Web Services, Proceedings of International
Workshop on Information Security and Survivability for Grid, Wuhan,
China, October 21-24, 2004.
[12] Miao Liu, He Qing Guo, Jin Dian Su, An Attribute and Role Based
Access Control Model for Web Services, Proceedings of the Fourt
International Conference on Machine Learning Cybernetics, Guangzhou,
August 18-21, 2005 pp. 1302 1306.
Vol. 2, No. 6, 2011
42 | P a g e
[13] Yizhu Zhao, Yanhua Zhao, Hongwei Lu, A Flexible Role and Resource
based Access Control Model, International Colloqium on Computing,
Vommunication, Control and Management, 2008, pp. 75-79.
[14] Xu Feng, Lin Guoyuan, Huang Hao, Xie Li, Role Based Access
Control System for Web Services, Proceedings of the Fourth
International Conference on Computer and Information Technology,
2004 IEEE.
[15] R. Bhatti, E. Bertino and A. Ghafoor, A Trust Based Conext Aware
Access Control Model for Web Services, Proceedings of the IEEE
International Conference on Web Services, California, USA, June6-9.
2004.
[16] Christian Emig, Frank Brandt, Sebastian Abeck, An Access Control
Metamodel for Web Service-Oriented Architecture, International
Conference on Software Engineering Advances (ICSEA 2007).
[17] X. Feng, L Jain., Research and development of trust anagement in web
security, Journal of Software 2002, Vol. 13, No. 11, pp. 2057-2064.
[18] W. H. Winsborough, K. E. Seamons, V. E. Jones, Automated trust
negotiation. In DARPA Information Survivability Conference and
Exposition, IEEE Press, 2000, pp. 88-102.
Vol. 2, No. 6, 2011
43 | P a g e
Design and Implementation on a Sub-band based
Acoustic Echo Cancellation Approach
Zhixin Chen
ILX Lightwave Corporation
Bozeman, Montana, USA
[email protected]

Abstract This paper describes the design and implementation of
a sub-band based acoustic echo cancellation approach, which
incorporates the normalized least mean square algorithm and the
double talk detection algorithm. According to the simulation, the
proposed approach works well in the modest linear noisy
environment. Since the proposed approach is implemented in fix-
pointed C, it can be easily ported into fixed-point DSPs to cancel
acoustic echo in real systems.
Keywords - Digital Signal Processing; Adaptive Filter; Acoustic
Echo Cancellation.
I. INTRODUCTION
People have been using telephones as a way of distant voice
communication for more than one century now. In a phone
conversation, echo is the sound of ones own voice being
played back to him after a delay. There are two types of echo
presenting in typical communication networks: hybrid echo and
acoustic echo. Hybrid echo is the outcome of impedance
mismatches along a telephone line, and is relatively easy to
identify and cancel. Acoustic echo is due to the leakage from
loudspeaker to microphone in a hands-free telephone, and is
much more difficult to tackle [1].
Strong and long delayed acoustic echo signals can be very
annoying, and in some cases, make conversation impossible.
The effective removal of acoustic echo is thus the key to
maintaining and improving voice quality on a phone call. This
has led to intensive research into the area of echo cancellation,
with the aim of providing solutions that can significantly
reduce or even remove acoustic echo [1-3].
Good acoustic echo cancellation system can greatly
enhance the audio quality of a communication system by
allowing conversation to progress more smoothly and
naturally, keeping the participants more comfortable, and
preventing listener fatigue. However, a poorly designed
acoustic echo cancellation will not provide these benefits and
can even degrade audio quality significantly. This paper
describes the design and simulation of an acoustic echo
cancellation (AEC) approach, which is based on the sub-band
method and the normalized least mean square (NLMS)
algorithm, and includes the double talk detection. According to
the simulation, the proposed acoustic echo cancellation
approach works well in the modest linear noisy environment.
Since the approach is implemented in fixed-point C language, it
can be easily ported into fixed-point DSPs and used in real
systems to cancel acoustic echo.
The remaining part of this paper is organized as follows.
First, the principle of acoustic echo cancellation is reviewed.
Next, the implementation of the proposed acoustic echo
cancellation approach is described. Then, the simulation results
of the proposed approach are presented. Finally, the paper is
concluded with a summary of results.
II. PRINCIPLE OF ACOUSTIC ECHO CANCELLATION
The way how acoustic echo canceller works is shown in
Figure 1. In general, the near end microphone signal d(n)
consists of signals from near end microphone, near end
background noise and echo of the far-end audio signal
originating from the loudspeaker [1]. The last signal is
undesired. The objective of the echo canceller is thus to form a
replica of the acoustic echo signal picked up by the terminal
microphone by inserting an adaptive filter h(n) parallel to the
signal path through loudspeaker, room, and microphone [4].
The objective of the adaptive filter is to provide a response as
equal as possible to that of the acoustic signal path. The far end
speech signal x(n) is then fed through the adaptive filter h(n) to
generate the signal y(n) to resemble the echo part of the
microphone signal. The echo y(n) is then subtracted from the
microphone signal d(n) to get the desire speech signal e(n).

Figure 1: Principle of the acoustic echo cancellation system.
III. IMPLEMENTATION OF ACOUSTIC ECHO CANCELLATION
APPROACH
From the previous description, the acoustic echo
cancellation system appears as a straight-forward realization of
an adaptive filtering algorithm. Some algorithms, such as LMS
or NLMS, can be used in the full speech band to implement it
directly [5]. However, one issue for the full-band
Far End Signal
e(n) = d(n) y(n)
y(n) = x(n) * h(n)
x(n)
-
Adaptive
Filter

Near End Signal
d(n)
Vol. 2, No. 6, 2011
44 | P a g e
implementation is the high computational complexity. For
example, in a typical conference room, the environment echo
delay varies from 100 to 500 milliseconds. For a sampling rate
of 8 kHz, this means about 800 to 4000 samples and thus
requires a very long adaptive filter length. Moreover, if the full
band method is used, the spectrum may vary dramatically
within the full band of 4 KHz [6]. Consequently, the sub-band
method was used in the proposed acoustic echo cancellation
approach because the spectrum will not vary a lot within each
sub-band.
The LMS algorithm is the most popular algorithm in
adaptive signal processing due to its simplicity and reliability.
Its main drawback is that the speed of convergence gets very
slow if there is a big spread among the eigenvalues [6]. Thus,
the normalized NLMS algorithm is used in the proposed
approach because it is insensitive to the amplitude variations of
the inputs signal so that it can achieve a robust performance.
Double talk is another important issue related to acoustic
echo cancellation. If the state of double talk cannot be detected
accuarately, it can cause divergence of the adaptive algorithm.
Consequently, a robust double-talk and single-talk detection
algorithm was designed in the proposed AEC approach.
In the following sections, the sub-band approach, the LMS
and NLMS algorithm, and the double talk detection algorithm
will be presented.
A. Sub-band Approach
As mentioned previously, the sub-band implementation has
two advantages. First, the complexity is reduced by dividing
the signal into sub-bands and applying adaptive filters to a
decimated signal in each sub-band. Second, the spectral
variability within a sub-band is reduced compared with the full
band signal [1]. To maintain the high quality of speech, the
analysis and synthesis filterbanks in the sub-band realization
must be designed to provide perfect reconstruction, which
means that the signal fed through the analysis and synthesis
filterbank system shall be an exact but delayed copy of the
input signal. In the proposed AEC approach, the Quadrature
Mirror Filters (QMF) was used to decompose the full band into
four sub-bands with an identical bandwidth of 1 kHz. The
reason that the full band is divided into four sub-bands is that
the energy for speech focuses on the lowest 1 kHz range and
the spectrum will be almost constant within the lowest 1 kHz
sub-band. The basic architecture for a two-channel QMF
system is shown in Figure 2.

Figure 2: A block diagram of the QMF filter architecture
In Figure 2, the down arrows represent decimation by 2 and
the up arrows denote interpolation by 2. The two-channel QMF
system consists of two input-output paths, each of which has a
bandwidth requirement that is half the original bandwidth
requirements. In Figure 2, the top path contains a low-pass
filter while the bottom path has a high-pass filter. Since the FIR
filter is used here, a very steep filter cannot be achieved
without a very large filter tap. The frequency response of the
top path will have some overlap with that of the bottom. The
overlap that exists beyond one-fourth the sampling frequency
in the top channel, when re-sampled at half the sampling
frequency, introduces aliasing errors. The bottom path also
introduces aliasing errors. According to [1], the aliasing errors
in the top and bottom path are self-cancelling if the following
condition is met.
0 ) ( )) ( ) ( ) ( ) ( (
1 1 0 0
= + z X z G z H z G z H , (1)
which is satisfied if ) ( ) (
1 0
z H z G = and ) ( ) (
0 1
z H z G = . A
special case assumes that ) (
0
z H and ) (
1
z H are sub-band filters
satisfying the mirror filter relationship ) ( ) (
0 1
z H z H = [1, 6].
Based on the above-mentioned theory, a 45
th
order linear
phase filter that is shown in Figure 3 was used in this paper. It
is shown that the low-pass and high-pass filter are mirror
symmetry. This QMF was used twice to get four sub-bands.

Figure 3: The impulse and frequency response of the QMF used in the
proposed AEC approach

B. NLMS Algorithm
It was described previously that the goal of the AEC system
is to identify an unknown system, the loudspeaker-to-
microphone transfer function, with an adaptive filter. That is,
the input signal needs to be processed so that it matches the
desired signal ) (n d . To measure how well the cancellation is
done, a performance criterion ) (e J to the error is used here.
According to [4], the performance criterion is chosen to be a
squared error measure and it can be defined as follows.
] )) ( ) ( [(
2
1
)) ( (
2
1
) (
2 2
n y n d E n e E e J = = . (2)
Here the room transfer function is assumed to be linear and
wide-sense stationary and the desired and input signals are
assumed to be jointly wide sense stationary and zero mean. The
Vol. 2, No. 6, 2011
45 | P a g e
filter is composed by a certain number of parameters ) (k h
n
that
are set to minimize the error ) (e J . For FIR filters, the output of
the filter can be written in the form ) ( ) ( ) (
1
0
k n x k h n y
N
k
n
=
=
,
(3)
where N is the filter order and ) (n x is the input signal. The
optimal ) (k h
n
parameters are found by solving
0 ))) ( ( ( ) (
2
= V = V n e E J
h h
, (4)
which leads to the Yule-Walker equation [4, 5, 7].
There are many algorithms for this equation, among which
the most popular ones are LMS and NLMS. In the LMS
algorithm, the adaptation rules are shown in the following
equation:
) ( ) ( ) (
) (
1
k n x n e k h k h
n n
+ =
+
u ,
(5)
where N k ,...., 2 , 1 = and ) ( ) ( ) ( n y n d n e = . Here ) (
1
k h
n+
are
the filter parameters at time 1 + n and u is a constant
adaptation step. The algorithm is proven to converge if the
error-performance surface is quadratic and if the adaptation
step satisfies
max
1
u < , where
max
is the biggest eigenvalue of
input correlation matrix R . Usually, the following condition is
taken
power input
2
< u . (6)
The main drawback of the LMS algorithm is the speed of
convergence gets very slow if there is a big spread among the
eigenvalues of R . Different from LMS, the NLMS algorithm is
insensitive to the amplitude variations of the input signal. The
NLMS algorithm is given by
) (
| | ) ( | |
) (
2
*
1
n e
n x
n x
h h
n n
| + =
+
. (7)
According to [5], the NLMS algorithm converges in the mean-
square if 2 0 < < | . In the LMS algorithm, the correction that
is applied to ) 1 ( + k h
n
is proportional to the input vector ) (n x .
Therefore, when ) (n x is large, the LMS algorithm experiences
a problem with gradient noise amplification. With the
normalization of the LMS step size by
2
| | ) ( | | n x in the NLMS
algorithm, however, this noise amplification problem is
diminished. However, a similar problem occurs when ) (n x
becomes too small. An alternative way is therefore to use the
following modification to the NLMS algorithm:
) (
| | ) ( | |
) (
2
*
1
n e
n x
n x
h h
n n
+
+ =
+
c
| , (8)
where c is some small positive number [4, 5].
In the proposed AEC approach, the echo delay that this
system can handle is set to 360 ms, which means that the taps
of the adaptive FIR filters will be up to 720. Since the speech
energy is different for different sub-bands, the taps of the filter
are different for different sub-bands. The taps are 720, 540, 360,
and 240 for the sub-bands from low to high respectively. The
step size was also chosen to be larger for higher sub-bands,
which mean that fast convergence is expected when the echo
signal has shorter delay and has higher frequency components.
This is because the speech energy focuses on the lower
frequency and so does the echo.
For the updating of the filter coefficient, Eq. (8) was used.
Here c was chosen to be
8
10
and the step size was chosen

from
5
10
to
4
10
, which is much larger than the traditional

value of
7
10
but still gets a good result.

C. Double Talk Detection
If the coefficients of the adaptive filters are updated during
double talk, the filter may be divergent. A double talk detection
algorithm was thus included in the proposed AEC approach to
detect the double talk and is described as follows. For both far
end and near end signal, a long-term energy and a short-term
energy were used to judge whether it is silent or not. The
coefficients of the adaptive filters will only be updated when
the far end talks and the near end does not talk because the
transfer function of the room acoustics is expected in this case.
Three convergence states, low, medium, and deep convergence
states are defined here and shown in Figure 4. Initially, the
system is in low convergence state. When neither far end nor
near end talks, the system will keep the convergence state.
When the far end talks and the power of the echo reduced
signal is less than that of the near end signal for at least two
seconds, the system will transit into medium convergence state.
The system will transit into deep convergence state if this
condition continues for two more seconds. If double talk stays
for one second, the system will transit into the low convergence
state no matter what state it is in.
If the system is in deep convergence state, the filter
coefficients will not be updated since in this case the room
transfer function will not change. If the system is in low or
medium convergence state, the coefficients will be updated
when the far end talks. The step size for low convergence state
is larger than that for medium convergence state because the
system is expected to converge faster initially but converge
slower when it is already in the modest convergence.

Figure 4: Transition among the three convergence states
Double talk for 1 second

Far end talks for 2
seconds
Low
Convergenc
e State
System
Initialization
Far end talks for 2 seconds

Double talk for 1 second

Deep
Convergence
State
Medium
Convergenc
e State
Vol. 2, No. 6, 2011
46 | P a g e
With the single talk and double talk detection and the
flexible transition of the convergence states presented above,
the system performs very well in the modest noise
environment.
IV. SIMULATION OF THE ACOUSTIC ECHO CANCELLATION
APPROACH
The proposed acoustic echo cancellation approach was
implemented using fix-point C language. The simulation
results for several different scenarios are presented in the
following subsections.
A. Only Near End Talks
In the first scenario, the far end is set to silent. This
simulates the case that the microphone is turned off in the far
end. The simulation is to test that the quality of the near end
signal does not degrade. The system performs as expected in
this case, which can be seen in Figure 5. In the figure, R
in

means the far end signal, S
in
denotes the near end signal, and
S
out
represents the desired near end output signal.

Figure 5 : Far end does not talk

Figure 6: Microphone is turned off in the near end
B. Only Far End Talks
In the second scenario, the near end was set to silent, which
simulates the case that the microphone is turned off in the near
end. The goal of this simulation is to verify that there is no
output signal. Figure 6 shows that the system performs as
expected.
In the third scenario, the microphone in the near end is set
to open. This means that an echo might present in the near end.
The goal of this simulation is to test how well the system
deletes the echo. The simulated echo signal is a 20%
attenuation and 300 ms delay version of the original signal. As
shown in Figure 7, the system completely removes the acoustic
echo.

Figure 7: Near end does not talk but the microphone is open
In the following scenario, the effect of the number the sub-
bands was tested. One to four sub-bands were used for the
signal and it is shown that the simulation result is almost the
same. It takes about 20000 samples (2.5 seconds) for the
system to converge and remove the acoustic echo. The similar
results for different number of used subbands are because the
energy of the speech focuses on the low frequency range and so
does that of the echo. Consequently, it does not matter whether
the higher sub-bands were processed or not. More details can
be seen in Figure 8.

Figure 8: Effect of the number of sub-bands on the AEC system performance
In the former scenarios, the delay is constant for the entire
signal. In this scenario, the delay is not constant, which means
that the delay for the first sentence can be 150 ms while that for
Vol. 2, No. 6, 2011
47 | P a g e
the second sentence is 200 ms. The system also works well in
this case, which is shown Figure 9.

Figure 9: Simulation of irregular delay
Till now, only one echo component is considered in the
simulations. In this scenario, the case that the echo includes
many copies of the original signal with different delays and
different gain is investigated. This means that the speech from
the speaker can be reflected by the wall and then fed into the
microphone in 10ms, 150ms or 200ms. It is shown in Figure 10
that the system also works well in this case.

Figure 10: Simulation of many delays
C. Double Talk
In the following scenarios, both the far end and the near end
are assumed to be active. A simulation was run to investigate
the case that the far end and the near end talk in different time,
which happens in most of the telephone communication
systems. It is expected that if the far end and the near end do
not talk at the same time, the performance should be as good as
that of the single talk. The simulation result shown in Figure 11
verifies this.
In another simulation, some white noise was added in the
near end. Since the noise is a part of the near end signal, it
should not affect the performance. Figure 12 verifies the
simulation result.

Figure 11: Far end and near end talk in different time

Figure 12: Noise shows up in the near end
It is assumed that the far end and the near end talk at the
same time in the last scenario. Since the room transfer function
changes from time to time, it is expected that the performance
will degrade. Figure 13 shows the detailed simulation result.

Figure 13: Far end and near end talk in the same time
Vol. 2, No. 6, 2011
48 | P a g e
V. CONCLUSION
In this paper, an acoustic echo cancellation system based on
the sub-band approach and the NLMS algorithm is proposed. A
double-talk detection algorithm is also included in the system.
From the simulations, the system works well for the echo with
300 ms delay in modest noise environment. It works very well
when only one end talks or when both end talk in the different
time. But it does not work very well when both end talk in the
same time. One thing deserving the attention is that this system
is designed for the linear echo delay, so it is not appropriate for
the environment that leads to echoes with new frequency
components. An acoustic echo system which can eliminate the
nonlinear echoes will be designed in the future.
REFERENCES
[1] A. Gilloire and M. Vetterli, Adaptive filtering in subbands with critical
sampling: analysis, experiments, and application to acoustic echo
cancellation, IEEE Trans. Signal Process., vol. 40, p.1862-1875, 1992
[2] J. Benesty, T. Gnsler, D. R. Morgan, M. M. Sondhi, and S. L. Gay,
Advances in network and acoustic echo cancellation, 2001, Springer.
[3] C. Antweiler, J. Grunwald and H. Quack, Approximation for optimal
stepsize control for acoustic echo cancellation, Proc. ICASSP-97, IEEE
Conference on Acoustics Speech and Signal Processing, Mnchen,
Germany, vol. 1, pp. 295-298, 1997
[4] Paolo Prandoni, Acoustic echo cancellation
http://lcavwww.epfl.ch/~prandoni/dsp/echo/echo.html
[5] K. A. Lee and W. S. Gan "Improving convergence of the NLMS
algorithm using constrained subband updates", IEEE Signal Process.
Letter, vol. 11, p.736-739, 2004.
[6] G. Zelniker and F. Taylor, Advanced digital signal processing, Marcel
Dekker, 1994
[7] M. G. Siqueira, "Adaptive filtering algorithms in acoustic echo
cancellation and feedback reduction", Ph.D. thesis, University of
California, Los Angeles, 1998

AUTHORS PROFILE
Zhixin Chen is a Firmware Engineer in ILX Lightwave Corporation. He
holds a BS degree from Xiamen University in China, a MS degree from
Xiamen University in China, and a Ph.D. degree from Montana State
University in USA, all in Electrical and Computer Engineering. His research
interest and working experience are in the area of acoustics, audio, and speech
processing, multimedia communication, and embedded system design for high
power current source, temperature controller, and optical power meter.

Vol. 2, No. 6, 2011
49 | P a g e
An Efficient Density based Improved K- Medoids
Clustering algorithm

Raghuvira Pratap A
M.Tech,CSE Department
P.V.P Siddhartha Institute of Technology
Vijayawada-07, AP
J Rama Devi
Sr.Assistant Professor,CSE Department
Vijayawada-07, AP
K Suvarna Vani
Assistant Professor,CSE Department
V.R. Siddhartha Engineering College
Vijayawada -07, AP
Dr.K Nageswara Rao
Professor & Head,CSE Department
Vijayawada-07, AP

Abstract Clustering is the process of classifying objects into
different groups by partitioning sets of data into a series of
subsets called clusters. Clustering has taken its roots from
algorithms like k-medoids and k-medoids. However conventional
k-medoids clustering algorithm suffers from many limitations.
Firstly, it needs to have prior knowledge about the number of
cluster parameter k. Secondly, it also initially needs to make
random selection of k representative objects and if these initial k
medoids are not selected properly then natural cluster may not
be obtained. Thirdly, it is also sensitive to the order of input
dataset.
Mining knowledge from large amounts of spatial data is known
as spatial data mining. It becomes a highly demanding field
because huge amounts of spatial data have been collected in
various applications ranging from geo-spatial data to bio-medical
knowledge. The database can be clustered in many ways
depending on the clustering algorithm employed, parameter
settings used, and other factors. Multiple clustering can be
combined so that the final partitioning of data provides better
clustering. In this paper, an efficient density based k-medoids
clustering algorithm has been proposed to overcome the
drawbacks of DBSCAN and kmedoids clustering algorithms. The
result will be an improved version of kmedoids clustering
algorithm. This algorithm will perform better than DBSCAN
while handling clusters of circularly distributed data points and
slightly overlapped clusters.
Keywords- Clustering; DBSCAN; Centroid; Medoid; k-medoids.
I. INTRODUCTION
Numerous applications require the management of spatial
data, i.e. data related to space. Spatial Database Systems
(SDBS) (Gueting 1994) are database systems for the
management of spatial data. Increasingly large amounts of data
are obtained from satellite images, X-ray crystallography or
other automatic equipment. Therefore, automated knowledge
discovery becomes more and more important in spatial
databases.
Clustering algorithms are attractive for the task of class
identification. However, the application to large spatial
databases raises the following requirements for clustering
algorithms:
(1) Minimal requirements of domain knowledge to
determine the input parameters, because appropriate values are
often not known in advance when dealing with large databases.
(2) Discovery of clusters with arbitrary shape, because the
shape of clusters in spatial databases may be spherical, drawn-
out, linear, elongated etc.
(3) Good efficiency on large databases, i.e. on databases of
significantly more than just a few thousand objects.
Clustering is considered as one of the important techniques
in data mining and is an active research topic for the
researchers. The objective of clustering is to partition a set of
objects into clusters such that objects within a group are more
similar to one another than patterns in different clusters. So far,
numerous useful clustering algorithms have been developed for
large databases, such as K-MEDOIDS [1], CLARANS [2],
BIRCH [3], CURE [4], DBSCAN [5], OPTICS [6], STING [7]
and CLIQUE [8]. These algorithms can be divided into several
categories. Three prominent categories are partitioning,
hierarchical and density-based. All these algorithms try to
challenge the clustering problems treating huge amount of data
in large databases. However, none of them are the most
effective. In density-based clustering algorithms, which are
designed to discover clusters of arbitrary shape in databases
with noise, a cluster is defined as a high-density region
partitioned by low-density regions in data space. DBSCAN
(Density Based Spatial Clustering of Applications with Noise)
[5] is a typical Density-based clustering algorithm. In this
paper, we present a new algorithm which overcomes the
drawbacks of DBSCAN and k-medoids clustering algorithms.
II. LITERATURE SURVEY
A. DBSCAN: Density Based Spatial Clustering of
Applications with Noise.
In this section, we present the algorithm DBSCAN (Density
Based Spatial Clustering of Applications with Noise) which is
designed to discover the clusters and the noise in a spatial
Vol. 2, No. 6, 2011
50 | P a g e
database. Ideally, we would have to know the appropriate
parameters Eps and MinPts of each cluster and at least one
point from the respective cluster. Then, we could retrieve all
points that are density-reachable from the given point using the
correct parameters. But there is no easy way to get this
information in advance for all clusters of the database.
However, there is a simple and effective heuristic to determine
the parameters Eps and MinPts of the "thinnest", i.e. least
dense, cluster in the database. Therefore, DBSCAN uses global
values for Eps and MinPts, i.e. the same values for all clusters.
The density parameters of the thinnest cluster are good
candidates for these global parameter values specifying the
lowest density which is not considered to be noise.
The idea of it was:
1. -neighbor: the neighbors in semi diameter of
an object
2. Kernel object: certain number (MinP) of
neighbors in semi diameter.
3. To a object set D, if object p is the -neighbor of
q, and q is kernel object, then p can get direct
density reachable from q.
4. To a , p can get direct density reachable from
q; D contains Minp objects; if a series object
p
1,
p
2,
.. p
n,
p
1=
q. p
n =
q. then p
i+1
can get direct
density reachable from p
i.
,
p
i
D ,1 i n.
5. To and MinP, if there exist an object o(oD) p
and q can get direct density reachable from o,p and q are
density connected.
Density Reachability and Density Connectivity:
Density reachability is the first building block in dbscan. It
defines whether two distance close points belong to the same
cluster. Points p1 is density reachable from p2 if two conditions
are satisfied: (i) the points are close enough to each other:
distance (p1, p2) <e, (ii) there are enough of points in is
neighborhood: |{r: distance(r, p2)}|>m, where r is a database
point.
Density connectivity is the last building step of dbscan.
Points p0 and pn are density connected, if there is a sequence
of density reachable points p1,i2,...,i(n-1) from p
0
to p
n
such
that p(i+1) is density reachable from pi. A dbscan cluster is a
set of all density connected points.
Explanation of DBSCAN Steps
DBScan requires two parameters: epsilon (eps) and
minimum points (minPts). It starts with an arbitrary
starting point that has not been visited. It then finds
all the neighbor points within distance eps of the
starting point.
If the number of neighbors is greater than or equal to
minPts, a cluster is formed. The starting point and its
neighbors are added to this cluster and the starting
point is marked as visited. The algorithm then repeats
the evaluation process for all the neighbors
recursively.
If the number of neighbors is less than minPts, the
point is marked as noise.
If a cluster is fully expanded (all points within reach
are visited) then the algorithm proceeds to iterate
through the remaining unvisited points in the dataset.
Advantages of DBSCAN
starting point.
If the number of neighbors is greater than or equal to
minPts, a cluster is formed. The starting point and its
neighbors are added to this cluster and the starting
point is marked as visited. The algorithm then repeats
the evaluation process for all the neighbors
recursively.
If the number of neighbors is less than minPts, the
point is marked as noise.
If a cluster is fully expanded (all points within reach
are visited) then the algorithm proceeds to iterate
through the remaining unvisited points in the dataset.
Disadvantages of DBSCAN
starting point.
DBSCAN cannot cluster data sets well with large
differences in densities, since the minPts-epsilon
combination cannot be chosen appropriately for all
clusters then
B. K-Medoids
K-medoid is a classical partitioning technique of clustering
that clusters the data set of n objects into k number of clusters
[1, 3]. This k: the number of clusters required is to be given by
user. This algorithm operates on the principle of minimizing
the sum of dissimilarities between each object and its
corresponding reference point. The algorithm randomly
chooses the k objects in dataset D as initial representative
objects called medoids. A medoid can be defined as the object
of a cluster, whose average dissimilarity to all the objects in the
cluster is minimal i.e. it is a most centrally located point in the
given data set. Then for all objects in the dataset, it assigns
each object to the nearest cluster depending upon the objects
distance to the cluster medoid. After every assignment of a data
object to particular cluster the new medoid is decided.
1) Input
k: the number of clusters.
D: a data set containing n objects.
2) Output
A set of k clusters.
3) Algorithm
1. Randomly choose k objects in D as the initial
representative objects;
2. for all objects in the data set D
Vol. 2, No. 6, 2011
51 | P a g e
a. Find the cluster C which is nearest to
object i by using the dissimilarity measure;
b. assign object i to cluster C;
c. set the member object in cluster C having
minimum intra cluster variance as new
centroid of C
3. Display statistics of clusters obtained.

The problem is K-Medoids does not generate the same
result with each run, because the resulting clusters depend on
the initial random assignments. It is more robust than kmedoids
in the presence of noise and outliers; however its processing is
more costly than the k-medoid method. Lastly, the optimal
number of clusters k is hard to be predicted, so it is difficult for
a user without prior knowledge to specify the value of k.
Problems with kmedoids clustering algorithm
The algorithm is simple and has nice convergence but there
are number of problems with this. Some of the weaknesses of
k-mediods are
When the numbers of data are not so many, initial
grouping will determine the cluster significantly.
The result is circular cluster shape because based on
distance.
The number of cluster, K, must be determined
beforehand. Selection of value of K is itself an issue
and sometimes its hard to predict beforehand the
number of clusters that would be there in data.
We never know the real cluster, using the same data,
if it is inputted in a different order may produce
different cluster if the number of data is few.
Experiments have shown that outliers can be a
problem and can force algorithm to identify false
clusters.
We never know which attribute contributes more to
the grouping process since we assume that each
attribute has the same weight.
III. PROPOSED ALGORITHM
Let D is the Dataset with k points
k be the number of clusters to be found
l be the number of clusters initially found by density
based clustering algorithm
be the Euclidean neighborhood radius
Minimum number of neighbors required in
neighborhood to form a cluster
p can be any point in D
N is a set of points in neighborhood of p
c=0
for each unvisited point p in dataset D

{
N = getNeighbors (p, )
if (sizeof(N) < )
mark p as NOISE
else
++ c
mark p as visited
add p to cluster c
recurse (N)
}
Now will have m clusters
for each detected clusters {
find the cluster centers Cmby taking the representative object.
find the total number of points in each cluster
}
If m>k {
# Join two or more as follows
select two cluster based on density and number of points
satisfying the application criteria and joint them and find the
new cluster center and repeat it until achieving k clusters.
Finally we will have Ck centers
} else {
l =k-m
# split one or more as follows
if ( m >=l ) {
Select a cluster based on density and number of points
satisfying the application criteria and split it using kmedoids
clustering algorithm and repeat it until achieving k clusters.
Finally we will have Ck centers
}
Apply one iteration of k-medoid clustering with k and new
Ck centers as the initial parameters and label all the clusters
with k labels.
Note: in our simulation of the algorithm, we only assumed
overlapped clusters of circular or spheroid in nature. So the
criteria for splitting or joining a cluster can be decided based on
the number of expected points in a cluster or the expected
density of the cluster (derived by using the number of points in
a cluster and the area of the cluster)
IV. EVALUATION AND RESULTS
Metrics Used For Evaluation
In order to measure the performance of a clustering and
classification system, a suitable metric will be needed. For
evaluating the algorithms under consideration, we used Rand
Index and Run Time as two measures
A. Performance in terms of time
We evaluated the three algorithms DBSCAN, k-medoid and
DBkmedoids in terms of time required for clustering. The
Attributes of Multidimensional Data:
The Number of Classes: 5
The Number of Dimensions: 2
The Number of Points per Class: 50, 100, 150, 200,250
The Standard Deviation: 7.000000e-001

Vol. 2, No. 6, 2011
52 | P a g e
S
.
N
o
.

T
o
t
a
l

N
o
.

o
f

R
e
c
o
r
d
s

i
n

P
r
o
t
e
i
n
s

D
a
t
a

B
a
s
e

Time taken for Classification
(in Seconds)
D
B
S
C
A
N

K
-
M
e
d
o
i
d
s

D
e
n
s
i
t
y

b
a
s
e
d

I
m
p
r
o
v
e
d

K
-
M
e
d
o
i
d
s

1

250

0.03

0.02

0.03

2

500

0.09

0.02

0.09

3

750

0.2

0.02

0.2

4

1000

0.31

0.03

0.3

5

1250

0.42

0.01

0.44

B. Performance in terms of accuracy
The Rand index or Rand measure is a commonly used
technique for measure of such similarity between two data
clusters. This measure was found by W. M. Rand and
explained in his paper "Objective criteria for the evaluation of
clustering methods" in Journal of the American Statistical
Association (1971).
Given a set of n objects S = {O1, ..., On} and two data
clusters of S which we want to compare: X = {x1, ..., xR} and
Y = {y1, ..., yS} where the different partitions of X and Y are
disjoint and their union is equal to S; we can compute the
following values:
a is the number of elements in S that are in the same
partition in X and in the same partition in Y,
b is the number of elements in S that are not in the
same partition in X and not in the same partition in Y,
c is the number of elements in S that are in the same
partition in X and not in the same partition in Y,
d is the number of elements in S that are not in the
same partition in X but are in the same partition in Y.

Intuitively, one can think of a + b as the number of
agreements between X and Y and c + d the number of
disagreements between X and Y. The Rand index, R, then
becomes, The Rand index has a value between 0 and 1 with 0
indicating that the two data clusters do not agree on any pair of
points and 1 indicating that the data clusters are exactly the
same.
The Attributes of Multidimensional Data:
The Number Of Classes: 5
The Number Of Dimensions: 2
The Number Of Points Per Class: 50 , 100, 150, 200,250
The standard Deviation: 7.000000e-001
S
.
N
o
.

T
o
t
a
l

N
o
.

o
f

R
e
c
o
r
d
s

i
n

P
r
o
t
e
i
n
s

D
a
t
a

B
a
s
e
Classification Accuracy
Measured by Rand Index
(Found using Original Class Labels and
Calculated Class Labels)
D
B
S
C
A
N

K
-
M
e
d
o
i
d
s

D
e
n
s
i
t
y

b
a
s
e
d

I
m
p
r
o
v
e
d

K
-
M
e
d
o
i
d
s

1

250

0.03

0.02

0.03

2

500

0.09

0.02

0.09

3

750

0.2

0.02

0.2

4

1000

0.31

0.03

0.3

5

1250

0.42

0.01

0.44

The Clustering and Outlier Detection Results
The following show the clusters and outliers marked with
DBSCAN and DBkmedoids
Time taken for Clustering

DBSCAN K-Medoids Improved K-
Medoids
Accuracy of Classification

DBSCAN K-Medoids Improved
K-Medoids
Vol. 2, No. 6, 2011
53 | P a g e

Original Custers

Clusters found by DBKmedoids

Clusters found by DBSCAN

Clusters and outliers marked with DBKMedoids
Clusters and Outliers marked with DBSCAN
From the plotted results, it is noted that DBkmedoids
perform better than DBSCAN.
V. CONCLUSION
This Clustering is an efficient way of reaching information
from raw data and Kmeans, Kmedoids are basic methods for it.
Although it is easy to implement and understand, Kmeans and
Kmedoids have serious drawbacks. The proposed clustering
and outlier detection system has been implemented using Weka
and tested with the proteins data base created by Gaussian
distribution function. The data will form circular or spherical
clusters in space. As shown in the tables and graphs, the
proposed Density based Kmedoids algorithm performed very
well than DBSCAN and k-medoids clustering in term of
quality of classification measured by Rand index. One of the
major challenges in medical domain is the extraction of
comprehensible knowledge from medical diagnosis data.

Vol. 2, No. 6, 2011
54 | P a g e
There is lot of scope for the proposed Density based
KMedoidss clustering algorithm in different application areas
such as medical image segmentation and medical data mining.
Future works may address the issues involved in applying the
algorithm in a particular application area.
REFERENCES
[1] Shu-Chuan Chu;Roddick, J.F.;Tsong-Yi Chen;Jeng-Shyang Pan
Efficient search approaches for k-medoids-based algorithms,
TENCON '02. Proceedings. 2002 IEEE Region 10 Conference on
Computers, Communications, Control and Power Engineering,Feb 2003.
[2] Shu-Chuan Chu,John F.Roddick,J.S. Pan An Efficient K -Medoids-
Based Algorithm Using Previous Medoid Index, Triangular Inequality
Elimination Criteria, and Partial Distance Search 4th International
Conference on Data Warehousing and Knowledge Discovery,2002.
[3] Raymond T. Ng and Jiawei Han, CLARANS: A Method for Clustering
Objects for Spatial Data Mining, IEEE TRANSACTIONS ON
KNOWLEDGE and DATA ENGINEERING, Vol. 14, No. 5,
SEPTEMBER 2002.
[4] K. Mumtaz1 and Dr. K. Duraiswamy, A Novel Density based improved
k-means Clustering Algorithm Dbkmeans International Journal on
Computer Science and Engineering, Vol. 02, No. 02, 2010, 213-218.
[5] Zhang T, Ramakrishnan R., Livny M., and BIRCH: An efficient data
clustering method for very large databases, In: SIGMOD Conference,
pp.103~114, 1996.
[6] K. Alsabti, S. Ranka, and V. Singh, An Efficient k-means Clustering
Algorithm, Proc. First Workshop High Performance Data Mining, Mar.
1998.
[7] M. Ester, H. P. Kriegel, J. Sander, and X. Xu, A density-based
algorithm for discovering clusters in large spatial databases, Proc. 1996
Int. Conf. Knowledge Discovery and Data Mining (KDD96), Portland,
OR, August 1996.
[8] Matheus C.J.; Chan P.K.; and Piatetsky-Shapiro G. 1993. Systems for
Knowledge Discovery in Databases, IEEE Transactions on Knowledge
and Data Engineering 5(6): 903-913.
[9] Ng R.T. and Han J. 1994. Efficient and Effective Clustering Methods for
Spatial Data Mining, Proc. 20th Int. Conf. on Very Large Data Bases,
144-155. Santiago, Chile.
[10] Indukuri, R. K. R. (2011). Dominating Sets and Spanning Tree based
Clustering Algorithms for Mobile Ad hoc Networks. IJACSA -
International Journal of Advanced Computer Science and Applications,
2(2). Retrieved from http://ijacsa.thesai.org.
[11] Kumar, V. (2011). Knowledge discovery from database Using an
integration of clustering and classification. IJACSA - International
Journal of Advanced Computer Science and Applications, 2(3), 29-33.
Retrieved from http://ijacsa.thesai.org.
[12] Jayech, K. (2011). Clustering and Bayesian network for image of faces
classification. IJACSA - International Journal of Advanced Computer
Science and Applications, (Special Issue). Retrieved from
www.ijacsa.thesai.org.
Vol. 2, No. 6, 2011
55 | P a g e
Context Switching Semaphore with Data Security
Issues using Self-healing Approach

M. Anand
Research Scholar, Dept of ECE
St. Peters University
Chennai - 54 India
Kuldeep Chouhan
Research Scholar, Dept of ECE
Dr M. G. R. University
Chennai 95 India
Dr. S. Ravi
Professor and Head, Dept of ECE
Dr M. G. R. University
Chennai 95 India
Syed Musthak Ahmed
Head, Dept of ECE
S. R. Engineering College
Warangal India

Abstract The main objective of a self healing scheme is to
share and secure the information of any system at the same
time. Self-healing techniques ultimately are dependable
computing techniques. Specifically self-healing systems have to
think for itself without human input, able to boot up backup
systems. However, sharing and protection are two
contradictory goals. Protection programs may be completely
isolated from each other by executing them on separate non-
networked computer, however, this precludes sharing.
Keywords- Self-healing; Semaphore; Data Security.
I. INTRODUCTION
Self-healing mechanisms complement approaches that
stop attacks from succeeding by preventing the injection of
code, transfer of control to injected code, or misuse of
existing code. Approaches to automatically defending
software systems have typically focused on ways to
proactively or at runtime protect an application from attack.
These proactive approaches include writing the system in a
safe language, linking the system with safe libraries,
transforming the program with artificial diversity, or
compiling the program with stack integrity checking. The
technique of program shepherding is validates branch
instructions to prevent transfer of control to injected code
and to make sure that calls into native libraries originate
from valid sources. Control Flow Integrity (CFI), observing
that high-level programming often assumes properties of
control flow that is not enforced at the machine level [3,4].
The use of CFI enables the efficient implementation of a
software shadow call stack with strong protection guarantees.
However, such techniques generally focus on integrity
protection at the expense of availability. Control flow is
often corrupted because input is eventually incorporated into
part of an instructions opcode, set as a jump target, or forms
part of an argument to a sensitive system call.
II. SELF HEALING SYSTEMS
A. Self Healing Approach
Self-healing is an approach to detect improper operations
of software applications, transactions and business processes,
and then to initiate corrective action without disrupting users
[8]. Healing systems that require human intervention or
intervention of an agent external to the system can be
categorized as assisted-healing systems. The key focus or
contrasting idea as compared to dependable systems is that a
self-healing system should recover from the abnormal (or
unhealthy) state and return to the normative (healthy) state
and function as it was prior to disruption. Some scholars treat
self-healing systems as an independent one while others view
as a subclass of traditional fault tolerant computing systems.
The system monitors itself for indications of anomalous
behavior. When such behavior is detected, the system enters
a self-diagnosis mode that aims to identify the fault and
extract as much information as possible with respect to its
cause, symptoms, and impact on the system [9]. The system
tries to adapt itself by generating candidate fixes, which are
tested to find the best target state. Self-healing systems can
support decision making in a large way for managerial and
organizational situations [11]. Many of the decision support
systems (DSS) offer passive forms of decision support,
where the decision-making process depends upon the user's
initiative. Such active involvement is especially needed in
complex decision-making environments.
B. Architecture of Self-Healing (S-H)
The term self in self-healing architecture is referred to
the action or response initiated automatically within the
system. A general architecture of a self-healing system is
shown in Fig. 1.
Vol. 2, No. 6, 2011
56 | P a g e

Fig. 1 General architecture of a self-healing system
C. Self Healing technique
The effective remediation strategies include failure-
oblivious computing, error virtualization, rollback of
memory updates, and data-structure repair [5]. These
approaches may cause a semantically incorrect continuation
of execution attempts to address this difficulty by exploring
semantically safe alterations of the programs environment.
The technique is subsequently introduced in a modified form
as failure-oblivious computing, because the program code is
extensively rewritten to include the necessary checks for
every memory access, the system incurs overheads for a
variety of different applications. Data-structure Repair is the
most critical concerns with recovering from software faults
and vulnerability exploits is ensuring the consistency and
correctness of program data and state.
D. Why Self Healing Systems
The software notoriously buggy and crash-prone is
despite considerable work in the fault tolerance and
reliability [1]. The current approach to ensuring the security
and availability of software consists of a mix of different
techniques:
a) Proactive techniques: seek to make the code as
dependable as possible, through a combination of safe
languages, libraries and compilers, code analysis tools,
formal methods and development methodologies.
b) Debugging techniques: aim to make post-fault
analysis and recovery as easy as possible for the programmer
that is responsible for producing a fix.
c) Runtime protection techniques: try to detect the
fault using some type of fault isolation, which address
specific types of faults or security vulnerabilities.
d) Containment techniques: seek to minimize the
scope of a successful exploit by isolating the process from
the rest of the system, e.g., through use of virtual machine.
e) Byzantine fault-tolerance and quorum techniques:
rely on redundancy and diversity to create reliable systems
out of unreliable components.
E. Elements of Self-Healing model
In the Self-healing process model, there are different
categories of aspects to the self-healing system,
a) Fault model: Self-healing systems have the tenets
of
dependable computing is that called a fault model must be
specified for any fault tolerant system. The fault model
answers the question of what faults the system is to tolerate.
Self-healing systems have a fault model in terms of what
injuries (faults), which are expected to be able to self-heal.
b) Fault duration: Faults can be permanent,
intermittent or transient due to an environmental condition. It
is important to state the fault duration assumption of a self-
healing approach to understand what situations it addresses.
c) Fault manifestation: The severity of the fault
manifestation, it affects the system in the absence of a self-
healing response. The faults cause immediate system crashes,
but, many faults cause less catastrophic consequences, such
as system slow-down due to excessive CPU loads, thrashing
due to memory hierarchy overloads, resource leakage, file
system overflow, and so on.
d) Fault source: Thee source of faults can affect self-
healing strategies due to implementation defects,
requirements defects, operational mistakes, and etc. Self-
healing software is designed only to withstand hardware
failures such as loss of memory [6] or CPU capacity and not
software failures.
e) Granularity: The granularity of a failure is the size
of the component that is compromised by that fault. Different
self-healing mechanisms are probably appropriate depending
on the granularity of the failures [7] and hence, the
granularity of recovery actions.
f) Fault profile expectations: The source of the fault
is the profile of fault occurrences that is expected. It
considered for self-healing might be only expected faults that
is based on design analysis or faults that are unexpected [12].
Additionally, faults might be random and independent, might
be correlated in space or time, or might even be intentional
due to malicious intent.
F. Conventional Methods of Security
The conventional methods can overcome only the effects
of passive threats and not the active threats for the
authenticate users. They reduce user-friendliness and also,
the amount of OS resources required to provide security is
high. Different protocol architectures are used for providing
security for each layer of the OSI model and it may not be
generic. Alternately, in this work, the information is allowed
to flow freely through the fetch and decode cycles while an
access or authentication is made only between the decode
and execute cycle before the data is permanently written into
the memory by the user (if authenticated). This is shown in
Fig. 2.

Fig. 2 Process of Execution
G. Proposed features of Security issues
The proposed hardware is shown in Figure 3. The
features of the proposed hardware are that it is (i) PCI
compliant and (ii) Mounted in a single chip
Vol. 2, No. 6, 2011
57 | P a g e

Fig. 3 Hardware Process System
a) Robustness: It provides defence against
vulnerabilities with few false positives or false negatives.
b) Flexible: It adapts easily to cover the continuously
evolving threats.
c) End-to-End: The security policy flows throughout
all the seven layers of the OSI model.
d) Scalable: It can co-exist with the existing circuitry
without any modification [2].
III. METHODOLOGY
A. Semaphores and Resource Sharing
A semaphore is a protected variable or an abstract data
type which restricts the access to shared resources such as
shared memory in a multiprogramming environment. It is a
primitive synchronization mechanism for sharing CPU time
and resources. It is a classic solution to prevent race
conditions.
B. Operation of Semaphore
The value of a semaphore is the number of units of
resources which are free. To avoid busy-waiting, a
semaphore has an associated queue of processes (usually
First in First Out). If a process performs a P operation on a
semaphore which has the value zero, the process is added to
the semaphores queue. When another process increments
the semaphore by performing a V operation, and there are
processes on the queue, one of them is removed from the
queue and resumes operation.
C. Binary Semaphore
In binary semaphore, if there is only one resource, the
semaphore takes value 0 or 1.This is explained in Fig. 4.

Fig. 4 Binary Semaphore
Suppose a P operation busy-waits (uses its turn to do
nothing) or maybe sleeps (tells the system not to give it a
turn) until a resource is available, where upon it immediately
claims one. Now, let V be the operation that simply makes
a resource available again after the process has finished using
it. The P and V operations must be atomic, i.e., no
process may be preempted in the middle of one of those
operations to run another operation on the same semaphore.
When a semaphore is being used, it takes value 0 and when
it takes the value 1, the process directly starts execution
without waiting.
D. Counting Semaphore
The counting semaphore concept can be extended with
the ability of claiming or returning more than one unit from
the semaphore. When multiple resources are to be shared by
many operations, such a semaphore is used. All the resources
must be of the same type. This is shown in Fig. 5.

Fig. 5 Counting Semaphore
In this, the initial value of semaphore is set equal to
number of resources available. As the semaphores are being
used, the value keeps decrementing. As the semaphores are
being released, after use, its value keeps incrementing. Zero
value refers to empty semaphore.
E. Mutex Semaphore
A mutex is a binary semaphore with extra features like
ownership or priority inversion protection. Mutexes are
meant to be used for mutual exclusion only. This is shown in
Figure 6.

Fig. 6 Mutex Semaphore
Initially the semaphore value is set to zero. Once a task
attains ownership, it can access the resources as many times
as it wants and each time, it accesses, the semaphore value
increases.
F. Characteristics of Semaphore
The characteristics of the three semaphores are shown in
TABLE I.
Vol. 2, No. 6, 2011
58 | P a g e
TABLE I. CHARACTERISTICS OF THREE SEMAPHORES
Binary Semaphore Counting
Semaphore
Mutex Semaphore
Anyone can release the
semaphore. Used for
mutual exclusion and
event notification.
Only after a task has
attained access, it
can release the
semaphore.
Only the owner can
release the semaphore.
Used only for mutual
exclusion.
IV. RESULTS AND DISCUSSION
A. Implementation
In the existing architecture of computers, the bottleneck
of connecting a high speed low memory device to a low
speed high memory device is solved by using an
intermediate memory module called the cache. The cache
exists between the high speed CPU and the lower speed
memories. However, there is no security between the data
link layer of the CPU and the cache. The proposed card is
placed on the PCI bus at the maximum possible speed and a
direct connection is established to the CPU via snooping. To
understand how the security layer is to be implemented, the
knowledge of the three basic terms is required: subject,
object and capability. Subject refers to the user or entity
which acts on behalf of the user on the system. Objects may
be defined as resources within the system. The main term
however is capability which is basically a token. The
possession of a capability by a subject confers access right
for an object. They cannot be easily modified, but they can
be reproduced.
The capabilities of the objects are to be stored in the
non-readable section of the HDD. For a subject to access a
particular object, it must possess the capability for doing so.
Hence, before a subject accesses a resource via the CPU, it
will first go through a screening check from the hardware on
whether or not its capabilities allow it to access such
resources. Hence, a security layer is now added to the data
link layer between the CPU and the cache. Additionally,
user also must be prevented from creating arbitrary
capabilities. This can be accomplished by placing the
capabilities in special Capability Segments which users
cannot access. Another approach is to add a tag bit to each
primary storage location. This bit, inaccessible to the user is
ON if the location contains a capability. It should be noted
that the hardware restricts the manipulation of the location
contents to appropriate system routines. If the last remaining
capability is destroyed, then that object cannot be used in
any manner. In this work, special provisions are made for
controlling the copying and movement of capabilities (as
well as
interpretation) depending on the hardware involved.
B. Evolved Function is a Semaphore Selector
The schematic of the control block performing the
evolved function is shown in Fig. 7.

Fig. 7 Schematic representation of Semaphore Selector
The power consumed by the block under read and write
mode is shown in Fig. 8 and Fig. 9 respectively.

Fig. 8 Current and Voltage Variations of semaphore selector during read
cycle

Vol. 2, No. 6, 2011
59 | P a g e

Fig. 9 Current and Voltage Variations of Semaphore selector during write
cycle
C. Evolved Function is a Resource Sharing Selector
The control block performing the evolved function of
resource sharing selector along with its power consumtion
is shown in Fig. 10. Similar graph corresponding to read and
write cycle is shown in Fig. 11 and Fig. 12 respectively.

Fig. 10 Schematic of Evolved resource sharing selector

Fig. 11 Current and Voltage Variations of evolved resource sharing selector
during Read cycle

.
Fig. 12 Current and Voltage Variations of evolved resource sharing selector
during write cycle
D. Evolved Function is a Snoop Selector Function
The schematic of the control block performing the
evolved function of a snoop selector circuit is shown in Fig.
13. The power consumed by the PE under read and write
cycles is shown in Fig. 14 and Fig. 15 respectively.
Vol. 2, No. 6, 2011
60 | P a g e

Fig. 13 Schematic of Evolved Snoop selector circuit

Fig. 14 Current and Voltage Variations of evolved Snoop selector during
read
Cycle
E. Evolved Function is a Context Switching Semaphore
The schematic of Context switching semaphore block of
the evolved function is shown in Fig. 16. The power
consumed
by the PE during read and write cycle is shown in Fig. 17
and Fig. 18 respectively

Fig. 15 Current and Voltage Variations of evolved snoop selector during
write
cycle

Fig. 16 Schematic of Evolved Context switching semaphore

Vol. 2, No. 6, 2011
61 | P a g e

Fig. 17 Current and Voltage Variations of evolved Context switching
semaphore during read cycle

Fig. 18 Current and Voltage Variations of evolved Context switching
semaphore during write cycle

The results of the discussion are tabulated in TABLE II.
It can be inferred from the table that a power level variation
exists between the read and write cycles.
TABLE II. POWER CONSUMED BY EVOLVED PE
Evolved
Function of PE
Average Power Consumed by Evolved
blocks
read write
Semaphore
selector
.3mW 2.325mW
Resource sharing
selector
.54mW 3.5mW
Snoop selector .12mW 3.665mW
Context
switching
semaphore
.456mW 3.5mW
V. CONCLUSION
In this work, self-healing systems prove increasingly
important in countering system software based attacks,
which recover and secure to the data from interrupted
services. Self-healing systems offer an active form of
decision support, without human intervention that can detect
the fault and recover from the fault. Also, with intelligent
architectural models, a self-healing system can select the
proper repair plan to deploy the broken component, if there is
more than one component that needs to be healed, can
prioritize a fault component over the others, etc.
ACKNOWLEDGMENT
This work was supported in part by Micro Logic System,
Chennai-600017.
REFERENCES
[1] Bouricius, W.G., Carter, W.C. & Schneider, P.R, Reliability
modeling techniques for self-repairing computer systems, in
proceedings of 24th National Conference, ACM, 1969, pp. 395-309.
[2] Shelton, C., Koopman, P.&Nace, W., A framework for scalable
analysis and design of system-wide graceful degradation in
distributed embedded systems, WORDS03, January 2003.
[3] G. Edward Suh, J. W. Lee, D. Zhang, and S. Devadas, Secure
Program Execution via Dynamic Information Flow Tracking, in
proceedings of the 11
th
International Conference on Architectural
Support for Programming Languages and Operating Systems
(ASPLOS-XI), October 2004.
[4] Martin Abadi, Mihai Budiu, Ulfar Erlingsson, and Jay Ligatti,
Control-Flow Integrity: Principles, Implementations, and
Applications, in proceedings of the ACM Conference on Computer
and Communications Security (CCS), 2005.
[5] B. Demsky and M. C. Rinard, Automatic Detection and Repair of
Errors in Data Structures, in proceedings of ACM OOPSLA,
October 2003.
[6] M. Locasto, S. Sidiroglou, and A.D Keromytis, Software Self-
Healing Using Collaborative Application Communities, in
proceedings of the Internet Society (ISOC) Symposium on Network
and Distributed Systems Security (SNDSS), February 2006.
[7] J. Kong, X. Hong, J.-S. Park, Y. Yi, and M. Gerla, LHospital: Self-
healing Secure Routing for Mobile Ad-hoc Networks, in Technical
Report CSD-TR040055, Dept. of Computer Science, UCLA, January
2005.
[8] Michael E. Shin and Jung Hoon An, Self-Reconfiguration in Self-
Healing Systems, in proceedings of the Third IEEE International
Workshop on EASE'06, pp 89-98 (2006).
[9] E. M. Dashofy, A. V. D. Hoek, and R. N. Taylor, Towards
architecture-based self-healing systems, in proceedings of the first
workshop on Self-healing systems, Charleston, South Carolina, pp.
21-26, 2002.
[10] H.You, V.Vittal, Z.Yang, Self-healing in power systems: an
approach using islanding and rate of frequency decline based load
shedding, IEEE Transaction on Power System, Vol.18, No.1, 2003,
pp.174 -181.
[11] T.A. Ramesh Kumaar and Dr.I.A.Chidambaram, Self-Healing
Strategy for Dynamic Security Assessment and Power System
Vol. 2, No. 6, 2011
62 | P a g e
Restoration, in International Journal of Computer Science &
Emerging Technologies, (E-ISSN: 2044-6004) Vol. 2, Issue 2, April
2011.
[12] J. Newsome and D. X. Song, Dynamic Taint Analysis, and Signature
Generation of Exploits on Commodity Software, in the Proceedings
of the Network and Distributed System Security Symposium, San
Diego,
CA, Feb.2005

AUTHORS PROFILE
1. Mr. M. Anand has completed his B.Tech. in
the faculty of ECE and M.Tech. in the area of
Applied Electronics. He is presently a research
scholar in the faculty of ECE at St. Peters
University, Avadi, Chennai. Mr. M.Anand shows
interest in the areas of Embedded system design,
Co-design of embedded systems, Scheduling etc.

2. Dr. S. Ravi was born on July 24
th
, 1971 at
Chennai, India and got A.M.I.E. (Electronics
Engineering) from Institution of Engineers,
Calcutta, M.Tech. (Communication systems) from
Anna University, Chennai in the Year 1994 and
Ph.D. from Anna University in the year 2003. He
has got 20 Years of Teaching Experience and 3
Years of Industrial Experience. Presently, he is working as Professor and
Head of the Department of Electronics Engineering in Dr. M.G.R.
University, Chennai. He has so far published nearly thirty four papers in
referred International Journals and successfully guided four Ph.D. scholars.
Dr. S. Ravi is a Life Member IETE and Life Member of Indian Society for
Technical Education

3. Mr. Kuldeep Chouhan has completed his
M.Tech. in the department of Computer Science
and Engineering (CSE). He is presently a research
scholar in the faculty of ECE at Dr. MGR
University, Chennai. His area of interest is
Wireless Sensor Network, Embedded system,
MultiAgent System in Sensor Security

4. Dr. Syed Musthak Ahmed, completed his
BE(Electronics) from Bangalore University,
ME(Electronics), UVCE Bangalore, Bangalore
University, Ph.D.(ECE), Vinayaka Missions
University, Salem. He has 25 years of teaching
experience working in reputed engineering
colleges and is presently working as Professor and
HOD (ECE), SR Engineering College, Warangal.
He is a member of Various Professional Societies like SMIEEE, FIETE,
MISSS, MISTE, MIAENG, MIAMT. He has various publications in
National and International Journals / Conferences
Vol. 2, No. 6, 2011
63 | P a g e
Mining Educational Data to Analyze Students
Performance

Brijesh Kumar Baradwaj
Research Scholor,
Singhaniya University,
Rajasthan, India
Saurabh Pal
Sr. Lecturer, Dept. of MCA,
VBS Purvanchal University,
Jaunpur-222001, India

Abstract The main objective of higher education institutions is
to provide quality education to its students. One way to achieve
highest level of quality in higher education system is by
discovering knowledge for prediction regarding enrolment of
students in a particular course, alienation of traditional
classroom teaching model, detection of unfair means used in
online examination, detection of abnormal values in the result
sheets of the students, prediction about students performance
and so on. The knowledge is hidden among the educational data
set and it is extractable through data mining techniques. Present
paper is designed to justify the capabilities of data mining
techniques in context of higher education by offering a data
mining model for higher education system in the university. In
this research, the classification task is used to evaluate students
performance and as there are many approaches that are used for
data classification, the decision tree method is used here.
By this task we extract knowledge that describes students
performance in end semester examination. It helps earlier in
identifying the dropouts and students who need special attention
and allow the teacher to provide appropriate advising/counseling.
Keywords-Educational Data Mining (EDM); Classification;
Knowledge Discovery in Database (KDD); ID3 Algorithm.
I. INTRODUCTION
The advent of information technology in various fields has
lead the large volumes of data storage in various formats like
records, files, documents, images, sound, videos, scientific data
and many new data formats. The data collected from different
applications require proper method of extracting knowledge
from large repositories for better decision making. Knowledge
discovery in databases (KDD), often called data mining, aims
at the discovery of useful information from large collections of
data [1]. The main functions of data mining are applying
various methods and algorithms in order to discover and extract
patterns of stored data [2]. Data mining and knowledge
discovery applications have got a rich focus due to its
significance in decision making and it has become an essential
component in various organizations. Data mining techniques
have been introduced into new fields of Statistics, Databases,
Machine Learning, Pattern Reorganization, Artificial
Intelligence and Computation capabilities etc.
There are increasing research interests in using data mining
in education. This new emerging field, called Educational Data
Mining, concerns with developing methods that discover
knowledge from data originating from educational
environments [3]. Educational Data Mining uses many
techniques such as Decision Trees, Neural Networks, Nave
Bayes, K- Nearest neighbor, and many others.
Using these techniques many kinds of knowledge can be
discovered such as association rules, classifications and
clustering. The discovered knowledge can be used for
prediction regarding enrolment of students in a particular
course, alienation of traditional classroom teaching model,
detection of unfair means used in online examination,
detection of abnormal values in the result sheets of the
students, prediction about students performance and so on.
The main objective of this paper is to use data mining
methodologies to study students performance in the courses.
Data mining provides many tasks that could be used to study
the student performance. In this research, the classification task
is used to evaluate students performance and as there are many
approaches that are used for data classification, the decision
tree method is used here. Informations like Attendance, Class
test, Seminar and Assignment marks were collected from the
students management system, to predict the performance at the
end of the semester. This paper investigates the accuracy of
Decision tree techniques for predicting student performance.
II. DATA MINING DEFINITION AND TECHNIQUES
Data mining, also popularly known as Knowledge
Discovery in Database, refers to extracting or mining"
knowledge from large amounts of data. Data mining techniques
are used to operate on large volumes of data to discover hidden
patterns and relationships helpful in decision making. While
data mining and knowledge discovery in database are
frequently treated as synonyms, data mining is actually part of
the knowledge discovery process. The sequences of steps
identified in extracting knowledge from data are shown in
Figure 1.
Vol. 2, No. 6, 2011
64 | P a g e

Figure 1: The steps of extracting knowledge from data
Various algorithms and techniques like Classification,
Clustering, Regression, Artificial Intelligence, Neural
Networks, Association Rules, Decision Trees, Genetic
Algorithm, Nearest Neighbor method etc., are used for
knowledge discovery from databases. These techniques and
methods in data mining need brief mention to have better
understanding.
A. Classification
Classification is the most commonly applied data mining
technique, which employs a set of pre-classified examples to
develop a model that can classify the population of records at
large. This approach frequently employs decision tree or neural
network-based classification algorithms. The data classification
process involves learning and classification. In Learning the
training data are analyzed by classification algorithm. In
classification test data are used to estimate the accuracy of the
classification rules. If the accuracy is acceptable the rules can
be applied to the new data tuples. The classifier-training
algorithm uses these pre-classified examples to determine the
set of parameters required for proper discrimination. The
algorithm then encodes these parameters into a model called a
classifier.
B. Clustering
Clustering can be said as identification of similar classes of
objects. By using clustering techniques we can further identify
dense and sparse regions in object space and can discover
overall distribution pattern and correlations among data
attributes. Classification approach can also be used for
effective means of distinguishing groups or classes of object
but it becomes costly so clustering can be used as
preprocessing approach for attribute subset selection and
classification.
C. Predication
Regression technique can be adapted for predication.
Regression analysis can be used to model the relationship
between one or more independent variables and dependent
variables. In data mining independent variables are attributes
already known and response variables are what we want to
predict. Unfortunately, many real-world problems are not
simply prediction. Therefore, more complex techniques (e.g.,
logistic regression, decision trees, or neural nets) may be
necessary to forecast future values. The same model types can
often be used for both regression and classification. For
example, the CART (Classification and Regression Trees)
decision tree algorithm can be used to build both classification
trees (to classify categorical response variables) and regression
trees (to forecast continuous response variables). Neural
networks too can create both classification and regression
models.
D. Association rule
Association and correlation is usually to find frequent item
set findings among large data sets. This type of finding helps
businesses to make certain decisions, such as catalogue design,
cross marketing and customer shopping behavior analysis.
Association Rule algorithms need to be able to generate rules
with confidence values less than one. However the number of
possible Association Rules for a given dataset is generally very
large and a high proportion of the rules are usually of little (if
any) value.
E. Neural networks
Neural network is a set of connected input/output units and
each connection has a weight present with it. During the
learning phase, network learns by adjusting weights so as to be
able to predict the correct class labels of the input tuples.
Neural networks have the remarkable ability to derive meaning
from complicated or imprecise data and can be used to extract
patterns and detect trends that are too complex to be noticed by
either humans or other computer techniques. These are well
suited for continuous valued inputs and outputs. Neural
networks are best at identifying patterns or trends in data and
well suited for prediction or forecasting needs.
F. Decision Trees
Decision tree is tree-shaped structures that represent sets of
decisions. These decisions generate rules for the classification
of a dataset. Specific decision tree methods include
Classification and Regression Trees (CART) and Chi Square
Automatic Interaction Detection (CHAID).
G. Nearest Neighbor Method
A technique that classifies each record in a dataset based on
a combination of the classes of the k record(s) most similar to it
in a historical dataset (where k is greater than or equal to 1).
Sometimes called the k-nearest neighbor technique.
Knowledge
Vol. 2, No. 6, 2011
65 | P a g e
III. RELATED WORK
Data mining in higher education is a recent research field
and this area of research is gaining popularity because of its
potentials to educational institutes.
Data Mining can be used in educational field to enhance
our understanding of learning process to focus on identifying,
extracting and evaluating variables related to the learning
process of students as described by Alaa el-Halees [4]. Mining
in educational environment is called Educational Data Mining.
Han and Kamber [3] describes data mining software that
allow the users to analyze data from different dimensions,
categorize it and summarize the relationships which are
identified during the mining process.
Pandey and Pal [5] conducted study on the student
performance based by selecting 600 students from different
colleges of Dr. R. M. L. Awadh University, Faizabad, India. By
means of Bayes Classification on category, language and
background qualification, it was found that whether new comer
students will performer or not.
Hijazi and Naqvi [6] conducted as study on the student
performance by selecting a sample of 300 students (225 males,
75 females) from a group of colleges affiliated to Punjab
university of Pakistan. The hypothesis that was stated as
"Student's attitude towards attendance in class, hours spent in
study on daily basis after college, students' family income,
students' mother's age and mother's education are significantly
related with student performance" was framed. By means of
simple linear regression analysis, it was found that the factors
like mothers education and students family income were
highly correlated with the student academic performance.
Khan [7] conducted a performance study on 400 students
comprising 200 boys and 200 girls selected from the senior
secondary school of Aligarh Muslim University, Aligarh,
India with a main objective to establish the prognostic value of
different measures of cognition, personality and demographic
variables for success at higher secondary level in science
stream. The selection was based on cluster sampling technique
in which the entire population of interest was divided into
groups, or clusters, and a random sample of these clusters was
selected for further analyses. It was found that girls with high
socio-economic status had relatively higher academic
achievement in science stream and boys with low socio-
economic status had relatively higher academic achievement in
general.
Galit [8] gave a case study that use students data to analyze
their learning behavior to predict the results and to warn
students at risk before their final exams.
Al-Radaideh, et al [9] applied a decision tree model to
predict the final grade of students who studied the C++ course
in Yarmouk University, Jordan in the year 2005. Three
different classification methods namely ID3, C4.5, and the
NaveBayes were used. The outcome of their results indicated
that Decision Tree model had better prediction than other
models.
Pandey and Pal [10] conducted study on the student
performance based by selecting 60 students from a degree
college of Dr. R. M. L. Awadh University, Faizabad, India. By
means of association rule they find the interestingness of
student in opting class teaching language.
Ayesha, Mustafa, Sattar and Khan [11] describes the use of
k-means clustering algorithm to predict students learning
activities. The information generated after the implementation
of data mining technique may be helpful for instructor as well
as for students.
Bray [12], in his study on private tutoring and its
implications, observed that the percentage of students receiving
private tutoring in India was relatively higher than in Malaysia,
Singapore, Japan, China and Sri Lanka. It was also observed
that there was an enhancement of academic performance with
the intensity of private tutoring and this variation of intensity of
private tutoring depends on the collective factor namely socio-
economic conditions.
Bhardwaj and Pal [13] conducted study on the student
performance based by selecting 300 students from 5 different
degree college conducting BCA (Bachelor of Computer
Application) course of Dr. R. M. L. Awadh University,
Faizabad, India. By means of Bayesian classification method
on 17 attribute, it was found that the factors like students grade
in senior secondary exam, living location, medium of teaching,
mothers qualification, students other habit, family annual
income and students family status were highly correlated with
the student academic performance.
IV. DATA MINING PROCESS
In present days educational system, a students
performance is determined by the internal assessment and end
semester examination. The internal assessment is carried out by
the teacher based upon students performance in educational
activities such as class test, seminar, assignments, general
proficiency, attendance and lab work. The end semester
examination is one that is scored by the student in semester
examination. Each student has to get minimum marks to pass a
semester in internal as well as end semester examination.
A. Data Preparations
The data set used in this study was obtained from VBS
Purvanchal University, Jaunpur (Uttar Pradesh) on the
sampling method of computer Applications department of
course MCA (Master of Computer Applications) from session
2007 to 2010. Initially size of the data is 50. In this step data
stored in different tables was joined in a single table after
joining process errors were removed.
B. Data selection and transformation
In this step only those fields were selected which were
required for data mining. A few derived variables were
selected. While some of the information for the variables was
extracted from the database. All the predictor and response
variables which were derived from the database are given in
Table I for reference.

Vol. 2, No. 6, 2011
66 | P a g e
TABLE I. STUDENT RELATED VARIABLES
Variable Description Possible Values
PSM Previous Semester Marks
{First > 60%
Second >45 & <60%
Third >36 & <45%
Fail < 36%}
CTG Class Test Grade
{Poor , Average, Good}
SEM Seminar Performance {Poor , Average, Good}
ASS Assignment {Yes, No}
GP General Proficiency {Yes, No}
ATT Attendance {Poor , Average, Good}
LW Lab Work {Yes, No}
ESM End Semester Marks
{First > 60%
Second >45 & <60%
Third >36 & <45%
Fail < 36%}

The domain values for some of the variables were defined
for the present investigation as follows:
- PSM Previous Semester Marks/Grade obtained in MCA
course. It is split into five class values: First >60%,
Second >45% and <60%, Third >36% and < 45%,
Fail < 40%.
- CTG Class test grade obtained. Here in each semester
two class tests are conducted and average of two class test
are used to calculate sessional marks. CTG is split into
three classes: Poor < 40%, Average > 40% and <
60%, Good >60%.
- SEM Seminar Performance obtained. In each semester
seminar are organized to check the performance of
students. Seminar performance is evaluated into three
classes: Poor Presentation and communication skill is
low, Average Either presentation is fine or
Communication skill is fine, Good Both presentation
and Communication skill is fine.
- ASS Assignment performance. In each semester two
assignments are given to students by each teacher.
Assignment performance is divided into two classes: Yes
student submitted assignment, No Student not
submitted assignment.
- GP - General Proficiency performance. Like seminar, in
each semester general proficiency tests are organized.
General Proficiency test is divided into two classes: Yes
student participated in general proficiency, No Student
not participated in general proficiency.
- ATT Attendance of Student. Minimum 70% attendance
is compulsory to participate in End Semester
Examination. But even through in special cases low
attendance students also participate in End Semester
Examination on genuine reason. Attendance is divided
into three classes: Poor - <60%, Average - > 60% and
<80%, Good - >80%.
- LW Lab Work. Lab work is divided into two classes:
Yes student completed lab work, No student not
completed lab work.
- ESM - End semester Marks obtained in MCA semester
and it is declared as response variable. It is split into five
class values: First >60% , Second >45% and <60%,
Third >36% and < 45%, Fail < 40%.
C. Decision Tree
A decision tree is a tree in which each branch node
represents a choice between a number of alternatives, and each
leaf node represents a decision.
Decision tree are commonly used for gaining information
for the purpose of decision -making. Decision tree starts with a
root node on which it is for users to take actions. From this
node, users split each node recursively according to decision
tree learning algorithm. The final result is a decision tree in
which each branch represents a possible scenario of decision
and its outcome.
The three widely used decision tree learning algorithms are:
ID3, ASSISTANT and C4.5.
D. The ID3 Decision Tree
ID3 is a simple decision tree learning algorithm developed
by Ross Quinlan [14]. The basic idea of ID3 algorithm is to
construct the decision tree by employing a top-down, greedy
search through the given sets to test each attribute at every tree
node. In order to select the attribute that is most useful for
classifying a given sets, we introduce a metric - information
gain.
To find an optimal way to classify a learning set, what we
need to do is to minimize the questions asked (i.e. minimizing
the depth of the tree). Thus, we need some function which can
measure which questions provide the most balanced splitting.
The information gain metric is such a function.
E. Measuring Impurity
Given a data table that contains attributes and class of the
attributes, we can measure homogeneity (or heterogeneity) of
the table based on the classes. We say a table is pure or
homogenous if it contains only a single class. If a data table
contains several classes, then we say that the table is impure or
heterogeneous. There are several indices to measure degree of
impurity quantitatively. Most well known indices to measure
degree of impurity are entropy, gini index, and classification
error.
Entropy =
j
2 j
log p -
j
p
Entropy of a pure table (consist of single class) is zero
because the probability is 1 and log (1) = 0. Entropy reaches
maximum value when all classes in the table have equal
probability.
Vol. 2, No. 6, 2011
67 | P a g e
Gini Index =
j
j
p
2
1
Gini index of a pure table consist of single class is zero
because the probability is 1 and 1-1
2
= 0. Similar to Entropy,
Gini index also reaches maximum value when all classes in the
table have equal probability.
Classification Error = { }
j
p max 1
Similar to Entropy and Gini Index, Classification error
index of a pure table (consist of single class) is zero because
the probability is 1 and 1-max (1) = 0. The value of
classification error index is always between 0 and 1. In fact the
maximum Gini index for a given number of classes is always
equal to the maximum of classification error index because for
a number of classes n, we set probability is equal to
n
p
1
=
and maximum Gini index happens at
2
1
1
n
n =
n
1
1 , while
maximum classification error index also happens at
n n
1
1
1
max 1 =
)
`
.
F. Splitting Criteria
To determine the best attribute for a particular node in the
tree we use the measure called Information Gain. The
information gain, Gain (S, A) of an attribute A, relative to a
collection of examples S, is defined as
) (
| |
| |
) ( ) , (
) (
v
A Values v
v
S Entropy
S
S
S Entropy A S Gain

e
=

Where Values (A) is the set of all possible values for
attribute A, and S
v
is the subset of S for which attribute A has
value v (i.e., S
v
= {s e S | A(s) = v}). The first term in the
equation for Gain is just the entropy of the original collection S
and the second term is the expected value of the entropy after S
is partitioned using attribute A. The expected entropy described
by this second term is simply the sum of the entropies of each
subset , weighted by the fraction of examples
| |
| |
S
S
v
that
belong to Gain (S, A) is therefore the expected reduction in
entropy caused by knowing the value of attribute A.
Split Information (S, A)=
| |
| |
log
| |
| |
1
2
S
S
S
S
i
n
i
i
=

and
Gain Ratio(S, A) =
) , (
) , (
A S n Informatio Split
A S Gain

The process of selecting a new attribute and partitioning the
training examples is now repeated for each non terminal
descendant node. Attributes that have been incorporated higher
in the tree are excluded, so that any given attribute can appear
at most once along any path through the tree. This process
continues for each new leaf node until either of two conditions
is met:
1. Every attribute has already been included along this
path through the tree, or
2. The training examples associated with this leaf node
all have the same target attribute value (i.e., their
entropy is zero).
G. The ID3Algoritm
ID3 (Examples, Target_Attribute, Attributes)
- Create a root node for the tree
- If all examples are positive, Return the single-node
tree Root, with label = +.
- If all examples are negative, Return the single-node
tree Root, with label = -.
- If number of predicting attributes is empty, then
Return the single node tree Root, with label = most
common value of the target attribute in the examples.
- Otherwise Begin
o A = The Attribute that best classifies
examples.
o Decision Tree attribute for Root = A.
o For each possible value, v
i
, of A,
Add a new tree branch below Root,
corresponding to the test A = v
i
.
Let Examples(v
i
) be the subset of
examples that have the value v
i
for
A
If Examples(v
i
) is empty
Then below this new
branch add a leaf node
with label = most common
target value in the
examples
Else below this new branch add the
subtree ID3 (Examples(v
i
),
Target_Attribute, Attributes {A})
- End
- Return Root
V. RESULTS AND DISCUSSION
The data set of 50 students used in this study was obtained
from VBS Purvanchal University, Jaunpur (Uttar Pradesh)
Computer Applications department of course MCA (Master of
Computer Applications) from session 2007 to 2010.
TABLE II. DATA SET
S. No. PSM CTG SEM ASS GP ATT LW ESM
1. First Good Good Yes Yes Good Yes First
2. First Good Average Yes No Good Yes First
3. First Good Average No No Average No First
4. First Average Good No No Good Yes First
5. First Average Average No Yes Good Yes First
6. First Poor Average No No Average Yes First
Vol. 2, No. 6, 2011
68 | P a g e
7. First Poor Average No No Poor Yes Second
8. First Average Poor Yes Yes Average No First
9. First Poor Poor No No Poor No Third
10. First Average Average Yes Yes Good No First
11. Second Good Good Yes Yes Good Yes First
12. Second Good Average Yes Yes Good Yes First
13. Second Good Average Yes No Good No First
14. Second Average Good Yes Yes Good No First
15. Second Good Average Yes Yes Average Yes First
16. Second Good Average Yes Yes Poor Yes Second
17. Second Average Average Yes Yes Good Yes Second
18. Second Average Average Yes Yes Poor Yes Second
19. Second Poor Average No Yes Good Yes Second
20. Second Average Poor Yes No Average Yes Second
21. Second Poor Average No Yes Poor No Third
22. Second Poor Poor Yes Yes Average Yes Third
23. Second Poor Poor No No Average Yes Third
24. Second Poor Poor Yes Yes Good Yes Second
25. Second Poor Poor Yes Yes Poor Yes Third
26. Second Poor Poor No No Poor Yes Fail
27. Third Good Good Yes Yes Good Yes First
28. Third Average Good Yes Yes Good Yes Second
29. Third Good Average Yes Yes Good Yes Second
30. Third Good Good Yes Yes Average Yes Second
31. Third Good Good No No Good Yes Second
32. Third Average Average Yes Yes Good Yes Second
33. Third Average Average No Yes Average Yes Third
34. Third Average Good No No Good Yes Third
35. Third Good Average No Yes Average Yes Third
36. Third Average Poor No No Average Yes Third
37. Third Poor Average Yes No Average Yes Third
38. Third Poor Average No Yes Poor Yes Fail
39. Third Average Average No Yes Poor Yes Third
40. Third Poor Poor No No Good No Third
41. Third Poor Poor No Yes Poor Yes Fail
42. Third Poor Poor No No Poor No Fail
43. Fail Good Good Yes Yes Good Yes Second
44. Fail Good Good Yes Yes Average Yes Second
45. Fail Average Good Yes Yes Average Yes Third
46. Fail Poor Poor Yes Yes Average No Fail
47. Fail Good Poor No Yes Poor Yes Fail
48. Fail Poor Poor No No Poor Yes Fail
49. Fail Average Average Yes Yes Good Yes Second
50. Fail Poor Good No No Poor No Fail
To work out the information gain for A relative to S, we
first need to calculate the entropy of S. Here S is a set of 50
examples are 14 First, 15 Second, 13 Third and 8
Fail..
Entropy (S) =
) ( log ) ( log
) ( log ) ( log
2 2
2 2
Fail Fail third third
Second Second First First
p p p p
p p p p

=
|
.
|
\
|
|
.
|
\
|
|
.
|
\
|
|
.
|
\
|
|
.
|
\
|
|
.
|
\
|
|
.
|
\
|
|
.
|
\
|
50
8
log
50
8
50
13
log
50
13
50
15
log
50
15
50
14
log
50
14
2 2
2 2

= 1.964
To determine the best attribute for a particular node in the
tree we use the measure called Information Gain. The
information gain, Gain (S, A) of an attribute A, relative to a
collection of examples S,
) (
| |
| |
) (
| |
| |
) (
| |
| |
) (
| |
| |
) ( ) , (
Fail
Fail
Third
Third
Second
Second
First
First
S Entropy
S
S
S Entropy
S
S
S Entropy
S
S
S Entropy
S
S
S Entropy PSM S Gain

=

TABLE III. GAIN VALUES
Gain Value
Gain(S, PSM) 0.577036
Gain(S, CTG) 0.515173
Gain(S, SEM) 0.365881
Gain(S, ASS) 0.218628
Gain (S, GP) 0.043936
Gain(S, ATT) 0.451942
Gain(S, LW) 0.453513

PSM has the highest gain, therefore it is used as the root
node as shown in figure 2.

Figure 2. PSM as root node
Gain Ratio can be used for attribute selection, before
calculating Gain ratio Split Information is shown in table IV.
TABLE IV. SPLIT INFORMATION
Split Information Value
Split(S, PSM) 1.386579
Split (S, CTG) 1.448442
Split (S, SEM) 1.597734
Split (S, ASS) 1.744987
Split (S, GP) 1.91968
Split (S, ATT) 1.511673
Split (S, LW) 1.510102
Gain Ratio is shown in table V.
TABLE V. GAIN RATIO
Gain Ratio Value
Gain Ratio (S, PSM) 0.416158
Gain Ratio (S, CTG) 0.355674
Gain Ratio (S, SEM) 0.229
Gain Ratio (S, ASS) 0.125289
Gain Ratio (S, GP) 0.022887
Gain Ratio (S, ATT) 0.298968
Gain Ratio (S, LW) 0.30032
This process goes on until all data classified perfectly or run
out of attributes. The knowledge represented by decision tree
can be extracted and represented in the form of IF-THEN rules.
PSM
Second First Third Fail
Vol. 2, No. 6, 2011
69 | P a g e
IF PSM = First AND ATT = Good AND CTG = Good or
Average THEN ESM = First
IF PSM = First AND CTG = Good AND ATT = Good
OR Average THEN ESM = First
IF PSM = Second AND ATT = Good AND ASS = Yes
THEN ESM = First
IF PSM = Second AND CTG = Average AND LW = Yes
THEN ESM = Second
IF PSM = Third AND CTG = Good OR Average AND
ATT = Good OR Average THEN PSM = Second
IF PSM = Third AND ASS = No AND ATT = Average
THEN PSM = Third
IF PSM = Fail AND CTG = Poor AND ATT = Poor
THEN PSM = Fail
Figure 3. Rule Set generated by Decision Tree
One classification rules can be generated for each path from
each terminal node to root node. Pruning technique was
executed by removing nodes with less than desired number of
objects. IF- THEN rules may be easier to understand is shown
in figure 3.
CONCLUSION
In this paper, the classification task is used on student
database to predict the students division on the basis of
previous database. As there are many approaches that are used
for data classification, the decision tree method is used here.
Informations like Attendance, Class test, Seminar and
Assignment marks were collected from the students previous
database, to predict the performance at the end of the semester.
This study will help to the students and the teachers to
improve the division of the student. This study will also work
to identify those students which needed special attention to
reduce fail ration and taking appropriate action for the next
semester examination.
REFERENCES
[1] Heikki, Mannila, Data mining: machine learning, statistics, and
databases, IEEE, 1996.
[2] U. Fayadd, Piatesky, G. Shapiro, and P. Smyth, From data mining to
knowledge discovery in databases, AAAI Press / The MIT Press,
Massachusetts Institute Of Technology. ISBN 0262 560976, 1996.
[3] J. Han and M. Kamber, Data Mining: Concepts and Techniques,
Morgan Kaufmann, 2000.
[4] Alaa el-Halees, Mining students data to analyze e-Learning behavior: A
Case Study, 2009..
[5] U . K. Pandey, and S. Pal, Data Mining: A prediction of performer or
underperformer using classification, (IJCSIT) International Journal of
Computer Science and Information Technology, Vol. 2(2), pp.686-690,
ISSN:0975-9646, 2011.
[6] S. T. Hijazi, and R. S. M. M. Naqvi, Factors affecting students
performance: A Case of Private Colleges, Bangladesh e-Journal of
Sociology, Vol. 3, No. 1, 2006.
[7] Z. N. Khan, Scholastic achievement of higher secondary students in
science stream, Journal of Social Sciences, Vol. 1, No. 2, pp. 84-87,
2005..
[8] Galit.et.al, Examining online learning processes based on log files
analysis: a case study. Research, Reflection and Innovations in
Integrating ICT in Education 2007.
[9] Q. A. AI-Radaideh, E. W. AI-Shawakfa, and M. I. AI-Najjar, Mining
student data using decision trees, International Arab Conference on
Information Technology(ACIT'2006), Yarmouk University, Jordan,
2006.
[10] U. K. Pandey, and S. Pal, A Data mining view on class room teaching
language, (IJCSI) International Journal of Computer Science Issue,
Vol. 8, Issue 2, pp. 277-282, ISSN:1694-0814, 2011.
[11] Shaeela Ayesha, Tasleem Mustafa, Ahsan Raza Sattar, M. Inayat Khan,
Data mining model for higher education system, Europen Journal of
Scientific Research, Vol.43, No.1, pp.24-29, 2010.
[12] M. Bray, The shadow education system: private tutoring and its
implications for planners, (2nd ed.), UNESCO, PARIS, France, 2007.
[13] B.K. Bharadwaj and S. Pal. Data Mining: A prediction for performance
improvement using classification, International Journal of Computer
Science and Information Security (IJCSIS), Vol. 9, No. 4, pp. 136-140,
2011.
[14] J. R. Quinlan, Introduction of decision tree: Machine learn, 1: pp. 86-
106, 1986.
[15] Vashishta, S. (2011). Efficient Retrieval of Text for Biomedical Domain
using Data Mining Algorithm. IJACSA - International Journal of
Advanced Computer Science and Applications, 2(4), 77-80.
[16] Kumar, V. (2011). An Empirical Study of the Applications of Data
Mining Techniques in Higher Education. IJACSA - International
Journal of Advanced Computer Science and Applications, 2(3), 80-84.
Retrieved from http://ijacsa.thesai.org.

AUTHORS PROFILE
Brijesh Kumar Bhardwaj is Assistant Professor in the
Department of Computer Applications, Dr. R. M. L. Avadh
University Faizabad India. He obtained his M.C.A degree
from Dr. R. M. L. Avadh University Faizabad (2003) and
M.Phil. in Computer Applications from Vinayaka mission
University, Tamilnadu. He is currently doing research in
Data Mining and Knowledge Discovery. He has published
one international paper.

Saurabh Pal received his M.Sc. (Computer Science)
from Allahabad University, UP, India (1996) and obtained
his Ph.D. degree from the Dr. R. M. L. Awadh University,
Faizabad (2002). He then joined the Dept. of Computer
Applications, VBS Purvanchal University, Jaunpur as
Lecturer. At present, he is working as Head and Sr. Lecturer
at Department of Computer Applications.
Saurabh Pal has authored a commendable number of research papers in
international/national Conference/journals and also guides research scholars in
Computer Science/Applications. He is an active member of IACSIT, CSI,
Society of Statistics and Computer Applications and working as
Reviewer/Editorial Board Member for more than 15 international journals. His
research interests include Image Processing, Data Mining, Grid Computing and
Artificial Intelligence.

Vol. 2, No. 6, 2011
70 | P a g e
Comparison between Traditional Approach and
Object-Oriented Approach in Software Engineering
Development
Nabil Mohammed Ali Munassar 1
PhD Student 3
rd
year of Computer Science & Engineering
Jawaharlal Nehru Technological University
Kuktapally, Hyderabad- 500 085, Andhra Pradesh, India

Dr. A. Govardhan 2
Professor of Computer Science & Engineering
Principal JNTUH of Engineering College, Jagityal,
Karimnagar (Dt), A.P., India
Abstract This paper discusses the comparison between
Traditional approaches and Object-Oriented approach.
Traditional approach has a lot of models that deal with
different types of projects such as waterfall, spiral, iterative and
v-shaped, but all of them and other lack flexibility to deal with
other kinds of projects like Object-Oriented.
Objectoriented Software Engineering (OOSE) is an object
modeling language and methodology. The approach of using
object oriented techniques for designing a system is referred to
as objectoriented design. Objectoriented development
approaches are best suited to projects that will imply systems
using emerging object technologies to construct, manage, and
assemble those objects into useful computer applications. Object
oriented design is the continuation of object-oriented analysis,
continuing to center the development focus on object modeling
techniques.
Keywordss- Software Engineering; Traditional Approach;
Object-Oriented Approach; Analysis; Design; Deployment; Test;
methodology; Comparison between Traditional Approach and
Object-Oriented Approach.
I. INTRODUCTION
All software, especially large pieces of software produced
by many people, should be produced using some kind of
methodology. Even small pieces of software developed by one
person can be improved by keeping a methodology in mind. A
methodology is a systematic way of doing things. It is a
repeatable process that we can follow from the earliest stages
of software development through to the maintenance of an
installed system. As well as the process, a methodology should
specify what were expected to produce as we follow the
process. A methodology will also include recommendation or
techniques for resource management, planning, scheduling and
other management tasks. Good, widely available
methodologies are essential for a mature software industry.
A good methodology will address at least the following
issues: Planning, Scheduling, Resourcing, Workflows,
Activities, Roles, Artifacts, Education. There are a number of
phases common to every development, regardless of
methodology, starting with requirements capture and ending
with maintenance. During the last few decades a number of
software development models have been proposed and
discussed within the Software Engineering community. With
the traditional approach, youre expected to move forward
gracefully from one phase to the other. With the modern
approach, on the other hand, youre allowed to perform each
phase more than once and in any order [1, 10].
II. TRADITIONAL APPROACH
There are a number of phases common to every
development, regardless of methodology, starting with
requirements capture and ending with maintenance. With the
traditional approach, will be expected to move forward
gracefully from one phase to the other. The list below describes
the common phases in software development [1, 6].
A. Requirements
Requirements capture is about discovering what is going to
achieve with new piece of software and has two aspects.
Business modeling involves understanding the context in which
software will operate. A system requirement modeling (or
functional specification) means deciding what capabilities the
new software will have and writing down those capabilities [1].
B. Analysis
Analysis means understanding what are dealing with.
Before designing a solution, it needs to be clear about the
relevant entities, their properties and their inter-relationships.
Also needs to be able to verify understanding. This can involve
customers and end users, since theyre likely to be subject-
matter experts [1].
C. Design
In the design phase, will work out, how to solve the
problem. In other words, make decisions based on experience,
estimation and intuition, about what software which will write
and how will deploy it. System design breaks the system down
into logical subsystems (processes) and physical subsystems
(computers and networks), decides how machines will
communicate, and chooses the right technologies for the job,
and so on [1].
D. Specification
Specification is an often-ignored, or at least often-
neglected, phase. The term specification is used in different
ways by different developers. For example, the output of the
requirements phase is a specification of what the system must
be able to do; the output of analysis is a specification of what
are dealing with; and so on [3].
Vol. 2, No. 6, 2011
71 | P a g e
Code Test
System/information
engineering
Analysis Design
E. Implementation
In this phase is writing pieces of code that work together to
form subsystems, which in turn collaborate to form the whole
system. The sort of the task which is carried out during the
implementation phase is Write the method bodies for the
Inventory class, in such a way that they conform to their
specification [5].
F. Testing
When the software is complete, it must be tested against the
system requirements to see if it fits the original goals. It is a
good idea for programmers to perform small tests as they go
along, to improve the quality of the code that they deliver [5].
G. Deployment
In the deployment phase, are concerned with getting the
hardware and software to the end users, along with manuals
and training materials. This may be a complex process,
involving a gradual, planned transition from the old way of
working to the new one [1].
H. Maintenance
When the system is deployed, it has only just been born. A
long life stretches before it, during which it has to stand up to
everyday use this is where the real testing happens. The sort
of the problem which is discovered discover during the
maintenance phase is When the log-on window opens, it still
contains the last password entered.' As the software developers,
we normally interested in maintenance because of the faults
(bugs) that are found in software. Must find the faults and
remove them as quickly as possible, rolling out fixed versions
of the software to keep the end users happy. As well as faults,
users may discover deficiencies (things that the system should
do but doesnt) and extra requirements (things that would
improve the system) [3, 6].

Figure 1: The linear Sequential Model [6].

III. OBJECT-ORIENTED APPROACH
In object-oriented approach, a system is viewed as a set
of objects. All object-orientation experts agree that a good
methodology is essential for software development, especially
when working in teams. Thus, quite a few methodologies have
been invented over the last decade. Broadly speaking, all
object-oriented methodologies are alike they have similar
phases and similar artifacts but there are many small
differences. Object-oriented methodologies tend not to be too
prescriptive: the developers are given some choice about
whether they use a particular type of diagram, for example.
Therefore, the development team must select a methodology
and agree which artifacts are to be produced, before they do
any detailed planning or scheduling. In general, each
methodology addresses:
The philosophy behind each of the phases.
The workflows and the individual activities within
each phase.
The artifacts that should be produced (diagrams,
textual descriptions and code).
Dependencies between the artifacts.
Notations for the different kinds of artifacts.
The need to model static structure and dynamic
behavior.
Static modeling involves deciding what the logical or
physical parts of the system should be and how they should be
connected together. Dynamic modeling is about deciding how
the static parts should collaborate. Roughly speaking, static
modeling describes how we construct and initialize the system,
while dynamic modeling describes how the system should
behave when its running. Typically, we produce at least one
static model and one dynamic model during each phase of the
development.
Some methodologies, especially the more comprehensive
ones, have alternative development paths, geared to different
types and sizes of development.[1,4]
The benefits of Object-Oriented Development are reduced
time to market, greater product flexibility, and schedule
predictability and the risks of them are performance and start-
up costs [5].
A. Analysis
The aim of the analysis process is to analyze, specify, and
define the system which is to be built. In this phase, we build
models that will make it easier for us to understand the system.
The models that are developed during analysis are oriented
fully to the application and not the implementation
environment; they are "essential" models that are independent
of such things as operating system, programming language,
DBMS, processor distribution, or hardware configuration.
Two different models are developed in analysis; the
Requirements Model and the Analysis Model. These are based
on requirement specifications and discussions with the
prospective users. The first model, the Requirements Model,
should make it possible to delimit the system and to define
what functionality should take place within it. For
this purpose we develop a conceptual picture of the system
using problem domain objects and also specific interface
descriptions of the system if it is meaningful for this system.
We also describe the system as a number of use cases that are
performed by a number of actors. The Analysis Model is an
architectural model used for analysis of robustness. It gives a
conceptual configuration of the system, consisting of various
object classes: active controllers, domain entities, and interface
objects. The purpose of this model is to find a robust and
extensible structure for the system as a base for construction.
Each of the object types has its own special purpose for this
robustness, and together they will offer the total functionality
that was specified in the Requirements Model. To manage the
development, the Analysis Model may combine objects into
Subsystems [2].
Vol. 2, No. 6, 2011
72 | P a g e
B. Construction
We build our system through construction based on the
Analysis Model and the Requirements Model created by the
analysis process. The construction process lasts until the coding
is completed and the included units have been tested. There are
three main reasons for a construction process:
1) The Analysis Model is not sufficiently formal.
2) Adaptation must be made to the actual
implementation environment.
3) We want to do internal validation of the analysis
results.
The construction activity produces two models, the Design
Model and the Implementation Model. Construction is thus
divided into two phases; design and implementation, each of
which develops a model. The Design Model is a further
refinement and formalization of the Analysis Model where
consequences of the implementation environment have been
taken into account. The Implementation model is the actual
implementation (code) of the system. [2].
C. Testing
Testing is an activity to verify that a correct system is being
built. Testing is traditionally an expensive activity, primarily
because many faults are not detected until late in the
development. To do effective testing we must have as a goal
that every test should detect a fault.
Unit testing is performed to test a specific unit, where a unit
can be of varying size from a class up to an entire subsystem.
The unit is initially tested structurally, that is, "white box
testing." This means that we use our knowledge of the inside of
the unit to test it. We have various coverage criteria for the test,
the minimum being to cover all statements. However, coverage
criteria can be hard to define, due to polymorphism; many
branches are made implicit in an object-oriented system.
However, polymorphism also enhances the independence of
each object, making them easier to test as standalone units. The
use of inheritance also complicates testing, since we may need
to retest operations at different levels in the inheritance
hierarchy. On the other hand, since we typically have less code,
there is less to test. Specification testing of a unit is done
primarily from the object protocol (so-called "black box
testing). Here we use equivalence partitioning to find
appropriate test cases. Test planning must be done early, along
with the identification and specification of tests [2].
D. UML
By the mid-1990s, the best-known methodologies were
those invented by Ivar Jacobson, James Rumbaugh and Grady
Booch. Each had his own consulting company using his own
methodology and his own notation. By 1996, Jacobson and
Rumbaugh had joined Rational Corporation, and they had
developed a set of notations which became known as the
Unified Modeling Language (UML). The three amigos, as
they have become known, donated UML to the Object
Management Group (OMG) for safekeeping and improvement.
OMG is a not-for-profit industry consortium, founded in 1989
to promote open standards for enterprise-level object
technology; their other well-known work is CORBA [1].
1) Use Case Diagram
A use case is a static description of some way in which a
system or a business is used, by its customers, its users or by
other systems. A use case diagram shows how system use cases
are related to each other and how the users can get at them.
Each bubble on a use case diagram represents a use case and
each stick person represents a user. Figure 2 depicts a car rental
store accessible over the Internet. From this picture, we can
extract a lot of information quite easily. For example, an
Assistant can make a reservation; a Customer can look for car
models; Members can log on; users must be logged on before
they can make reservations; and so on [1, 3].

Figure 2: A use Case Diagram
2) Class Diagram (Analysis Level)
A class diagram shows which classes exist in the business
(during analysis) or in the system itself (during subsystem
design). Figure 3 shows an example of an analysis-level class
diagram, with each class represented as a labeled box. As well
as the classes themselves, a class diagram shows how objects
of these classes can be connected together. For example, Figure
3 shows that a CarModel has inside it a CarModelDetails,
referred to as its details.U3: View Car Model Details. (Extends
U2, extended by U7.) Preconditions: None.
a) Customer selects one of the matching Car Models.
b) Customer requests details of the selected Car Model.
c) iCoot displays details of the selected Car Model
(make, engine size, price, description, advert and
poster).
d) If Customer is a logged-on Member, extend with U7.
Postconditions: iCoot has displayed details of selected Car
Models.
Nonfunctional Requirements: r1. Adverts should be
displayed using a streaming protocol rather than requiring a
download [1, 5].
Vol. 2, No. 6, 2011
73 | P a g e
Figure 3: A class Diagram at the Analysis Level.
3) Communication Diagram
A communication diagram, as its name suggests, shows
collaborations between objects. The one shown in Figure 4
describes the process of reserving a car model over the Internet:
A Member tells the MemberUI to reserve a
CarModel; the MemberUI tells the ReservationHome to create
a Reservation for the given CarModel and the current Member;
the MemberUI then asks the new Reservation for its number
and returns this to the Member [1].
1) Deployment Diagram
A deployment diagram shows how the completed system
will be deployed on one or more machines. A deployment
diagram can include all sorts of features such as machines,
processes, files and dependencies. Figure 5 shows that any
number of HTMLClient nodes (each hosting a Web Browser)
and GUIClient nodes communicate with two server machines,
each hosting a WebServer and a CootBusinessServer; each
Web Server communicates with a CootBusinessServer; and
each CootBusinessServer communicates with a DBMS running
on one of two DBServer nodes [1].
Figure 4: A communication Diagram

Figure 5: A deployment Diagram.
2) Class Diagram (Design Level)
The class diagram shown in Figure 6 uses the same notation
as the one introduced in Figure 3. The only difference is that
design-level class diagrams tend to use more of the available
notation, because they are more detailed. This one expands on
part of the analysis class diagram to show methods,
constructors and navigability [1, 3].

Vol. 2, No. 6, 2011
74 | P a g e
Figure 6: A design-level Class Diagram
3) Sequence Diagram
A sequence diagram shows interactions between objects.
Communication diagrams also show interactions between
objects, but in a way that emphasizes links rather than
sequence. Sequence diagrams are used during subsystem
design, but they are equally applicable to dynamic modeling
during analysis, system design and even requirements capture.
The diagram in Figure 7 specifies how a Member can log off
from the system. Messages are shown as arrows flowing
between vertical bars that represent objects (each object is
named at the top of its bar). Time flows down the page on a
sequence diagram. So, Figure 7 specifies, in brief: a Member
asks the AuthenticationServlet to logoff; the
AuthenticationServlet passes the request on to the
AuthenticationServer, reading the id from the browser session;
the AuthenticationServer finds the corresponding Member
object and tells it to set its session id to 0; the Member passes
this request on to its InternetAccount; finally, the Member is
presented with the home page [1, 5].

Figure 7: A sequence Diagram from the Design Phase
IV. COMPARISON BETWEEN TRADITONAL APPROACH AND
OBJECT-ORIENTED APPROACH TO DEVELOPMENT IN
SOFTWARE ENGINEERING
Summarize the comparison between Traditional Approach
and Object-Oriented Approach shows through the table1.
TABLE 1. COMPARISON BETWEEN TRADITIONAL APPROACH AND OBJECT-
ORIENTED APPROACH
TABLE I.

Traditional Approach Object-Oriented Approach
Used to develop the
Traditional Projects that uses
procedural programming.
Used to develop Object-oriented
Projects that depends on Object-
Oriented programming.
Uses common processes
likes: analysis, design,
implementation, and testing.
Uses UML notations likes: use case,
class diagram, communication
diagram, development diagram and
sequence diagram.
Depends on the size of
projects and type of projects.
Depends on the experience of the
team and complexity of projects
through the numbers of objects.
Needs to large duration
sometimes to development
the large projects.
Need to more time than Traditional
approach and leads that to more
cost.
The problem of Traditional
approach using classical life
cycle [7, 8].
The object-oriented software life
cycle identifies the three traditional
activities
of analysis, design, and
implementation.[8].

Vol. 2, No. 6, 2011
75 | P a g e
Waterfall
Model
Iterative
Dev.
V-Shape
Model
Spiral
Model
XP Model
Large Projects
Medium Projects
Small Projects
Figure 8: Illustrate The Different Models of Traditional Approach with
Different Projects. [1, 6, 11]
From the previous figure 8 which illustrates the five
models from traditional approach that deals with three types of
projects, where we notice the waterfall model deals properly
with large and medium projects like spiral model and iterative
model that needs more time more cost and experience for team,
however the V-shape model and XP model use properly with
medium and small projects, because they need little time and
some experience of team to perform projects.

Figure 9: Illustrate The Different Criteria (Complexity, Experience and Cost)
for Traditional Approach and Object-oriented Approach. [3, 5, 10]
From the previous chart illustrates the some criteria such
as (Complexity, Experience, and Cost). In Traditional
Approach this criterion depends on the type of model and size
of project, but in general as shows from figure 9 is little above
from the middle, however the Object-Oriented Approach
depends on the complexity of project that leads to increase the
cost than other approach.
After completing this paper, it is concluded that:
1. As with any technology or tool invented by human
beings, all SE methodologies have limitations [9].
2. The software engineering development has two ways to
develop the projects that: traditional approach and object-
oriented approach.
3. The traditional approach uses traditional projects that
used in development of their procedural programming
like C, this approach leads software developers to focus
on Decomposition of larger algorithms into smaller ones.
4. The object-oriented approach uses to development
the object-oriented projects that use the object-
oriented programming like: C++ and Java.
5. The object-oriented approach to software
development has a decided advantage over the
traditional approach in dealing with complexity and
the fact that most contemporary languages and tools
are object-oriented.
Finally, some topics can be suggested for future works:
1. Design the model that includes the features of
traditional approach and object-oriented approach to
develop and deals with different projects in software
engineering.
2. Updating some traditional approach to be able to use
different types of projects.
3. Simplifying the object-oriented approach through its
steps to use the smallest projects that deal with simple
programming.

REFERENCES
[1] Mike ODocherty, "Object-Oriented Analysis and Design Understanding
System Development with UML 2.0", John Wiley & Sons Ltd,
England, 2005.
[2] Magnus Christerson and Larry L. Constantine, Object-Oriented
Software Engineering- A Use Case Driven Approach , Objective
Systems, Sweden, 2009.
[3] Ian sommerville, Software Engineering, Addison Wesley, 7th edition,
2004.
[4] Pankaj Jalote, An Integrated Approach to Software Engineering,
Springer Science Business Media, Inc, Third Edition, 2005.
[5] Grady Booch, Object-Oriented Analysis and Design with applications,
Addison Wesley Longman, Inc, second Edition, 1998.
[6] Roger S. Pressman, Software Engineering a practitioners approach,
McGraw-Hill, 5th edition, 2001.
[7] M M Lehman,Process Models, Process Programs, Programming
Support, ACM, 1987.
[8] Tim Korson and John D. McGregor, Understanding Object-Oriented: A
Unifying Paradigm, ACM, Vol. 33, No. 9, 1990.
[9] Li Jiang and Armin Eberlein, Towards A Framework for Understanding
the Relationships between Classical Software Engineering and Agile
Methodologies, ACM, 2008.
[10] Luciano Rodrigues Guimares and Dr. Plnio Roberto Souza Vilela,
Comparing Software Development Models Using CDM, ACM, 2005.
[11] Alan M. Davis and Pradip Sitaram, A Concurrent Process Model of
Software Development, ACM, Software Engineering Notes Vol. 19 No.
2, 1994.

AUTHORS PROFILE
Nabil Mohammed Ali Munassar
Was born in Jeddah, Saudi Arabia in 1978. He studied
Computer Science at University of Science and
Technology, Yemen from 1997 to 2001. In 2001 he
received the Bachelor degree. He studied Master of
Information Technology at Arab Academic, Yemen, from
2004 to 2007. Now he Ph.D. Student 3
rd
year of CSE at
Jawaharlal Nehru Technological University (JNTU),
Hyderabad, A. P., India. He is working as Associate
Professor in Computer Science & Engineering College in
University Of Science and Technology, Yemen. His areas of
interest include Software Engineering, System Analysis and
Design, Databases and Object Oriented Technologies.
0
10
20
30
40
50
60
70
80
90
Traditiona App.
Complixity
Experience
Cost
Object-Oriented
App.
Vol. 2, No. 6, 2011
76 | P a g e
Dr.A.Govardhan
Received Ph.D. degree in Computer Science and Engineering
from Jawaharlal Nehru Technological University in 2003,
M.Tech. from Jawaharlal Nehru University in 1994 and B.E.
from Osmania University in 1992. He is working as a Principal
of Jawaharlal Nehru Technological University,
Jagitial. He has published around 108 papers in various national
and international Journals/conferences. His research of interest
includes Databases, Data Warehousing & Mining, Information
Retrieval, Computer Networks, Image Processing, Software
Engineering, Search Engines and Object Oriented Technologies.
Vol. 2, No. 6, 2011
77 | P a g e
Estimation of Dynamic Background and Object
Detection in Noisy Visual Surveillance

M.Sankari
Department of Computer Applications,
Nehru Institute of Engineering and Technology,
Coimbatore, INDIA
C. Meena
Head, Computer Centre,
Avinashilingam University,
Coimbatore, INDIA

AbstractDynamic background subtraction in noisy
environment for detecting object is a challenging process in
computer vision. The proposed algorithm has been used to
identify moving objects from the sequence of video frames which
contains dynamically changing backgrounds in the noisy
atmosphere. There are many challenges in achieving a robust
background subtraction algorithm in the external noisy
environment. In connection with our previous work, in this
paper, we have proposed a methodology to perform background
subtraction from moving vehicles in traffic video sequences that
combines statistical assumptions of moving objects using the
previous frames in the dynamically varying noisy situation.
Background image is frequently updated in order to achieve
reliability of the motion detection. For that, a binary moving
objects hypothesis mask is constructed to classify any group of
lattices as being from a moving object based on the optimal
threshold. Then, the new incoming information is integrated into
the current background image using a Kalman filter. In order to
improve the performance, a post-processing has been done. It
has been accomplished by shadow and noise removal algorithms
operating at the lattice which identifies object-level elements. The
results of post-processing can be used to detect object more
efficiently. Experimental results and analysis show the
prominence of the proposed approach which has achieved an
average of 94% accuracy in real-time acquired images.
Keywords- Background subtraction; Background updation; Binary
segmentation mask; Kalman filter; Noise removal; Shadow
removal; Traffic video sequences.
I. INTRODUCTION
In visual surveillance model, estimating the dynamic
background and detecting the object from the noisy
environment is a computationally challenging problem. Our
main target is to identify the object from the multi model
background using background subtraction, shadow removal
and noise removal techniques. For that we need to detect and
extract the foreground object from the background image.
After detecting the foreground object there may a large number
of possible degradations that an image can suffer. Common
degradations are blurring, motion and noise. Blurring can be
caused when an object in the image is outside the cameras due
to loss of depth information during the exposure. In the
proposed approach after detecting the object image, converting
it into its spatial frequencies, developing a point spread
function (PSF) to filter the image with, and then converting the
filtered result back into the spatial domain to see if blur was
removed.
This can be done in several steps. At the end, an
algorithm was developed for removing blur from an already
blurry image with no information regarding the blurring PSF.
In-class variability, occlusion, and lighting conditions also
change the overall appearance of vehicles. Region along the
road changes continuously while the lighting conditions depend
on the time of the day and the weather. The entire process is
automatic and uses computation time that scales according to
the size of the input video sequence.
The remainder of the paper is organized as follows: Section
II gives the overview of the related work. Section III describes
the architecture and modeling of proposed methodology for
background elimination and object detection. Implementation
and performance are analyzed in section IV. Section V contains
the concluding remarks and future work.
II. OVERVIEW OF THE RELATED WORK
Scores of research have been done in the literature in order
to attain a solution to an efficient and reliable background
subtraction. To detect moving objects in a dynamic scene,
adaptive background subtraction techniques have been
developed [1] [2] [3]. Adaptive Gaussian mixtures are
commonly chosen for their analytical representation and
theoretical foundations. For these reasons, they have been
employed in real-time surveillance systems for background
subtraction [4] [5] and object tracking [6]. For foreground
analysis [7] [8], a method for foreground analysis was proposed
for moving object, shadow, and ghost by combining the motion
information. The computation cost is relatively expensive for
real-time video surveillance systems because of the
computation of optical flow. In [9], a work has presented on a
novel background subtraction algorithm that is capable of
detecting objects of interest while all pixels are in motion.
Background subtraction technique is mostly used for motion
pictures to segment the foreground object by most of the
researchers [10] [11]. Liyuan Li, et al. [12] proposed
foreground object detection through foreground and
background classifications under bayesian framework. In
addition, moving object segmentation with background
suppression is affected by the problem of shadows [6] [13].
Indeed, the moving object detection do not classify shadows as
belonging to foreground objects since the appearance and
Vol. 2, No. 6, 2011
78 | P a g e
geometrical properties of the object can be distorted which, in
turn, affects many subsequent tasks such as object
classification and the assessment of moving object position. In
this paper, we propose a novel simple method that exploits all
these features, combining them so as to efficiently provide
detection of moving objects, ghosts, and shadows. The main
contribution of this proposal is the integration of knowledge of
detected objects, shadows, and ghosts in the segmentation
process to enhance both object segmentation and background
keep posted. The resulting method proves to be accurate and
reactive and, at the same time, fast and flexible in the
applications.
III. PROPOSED WORK
The proposed system extracts foreground objects such as
people, objects, or events of interest in variety of noisy
environment. The schematic flow of the proposed algorithm is
shown in Fig.1. This is an extension work of our previous
method [14]. Typically, these systems consist of stationary
cameras placed in the highways. These cameras are integrated
with, intelligent computer systems that perform preprocessing
operation from the captured video images and notify human
operators or trigger control process. The objective of this real-
time motion detection and tracking algorithm is to provide low-
level functionality for building higher-level recognition
capabilities.

Figure 1. Schematic flow of the proposed algorithm to detect the objects in
the noisy environment.
A. Preprocessing
Preprocessing is the key step and the starting point for
image analysis, due to the wide diversity of resolution, image
format, sampling models, and illumination techniques that are
used during acquisition. In our method, preprocessing step was
done by statistical method using adaptive median filter. The
resultant frames are then utilized as an input for the background
subtraction module. Image I(x,y) at time t is shown in Fig.2.
The background image B(x,y) at time t is shown Fig.3.

Figure 2. Image at time t: I (x; y; t).

Figure 3. Image at time t: B (x; y; t).
In order to get the estimated background, we have used an
adaptive median filter. Basically, impulse noise is a major
artifact that affects the sequence of frame in the surveillance
system. For this reason to estimate the background in the noisy
environment we have proposed an adaptive median filter
(AMF). The AMF can be used to enhance the quality of noisy
signals, in order to achieve better forcefulness in pattern
recognition and adaptive control systems. It executes on spatial
processing to determine which pixels in an image have been
exaggerated by impulse noise. The AMF categorizes pixels as
noise by contrasting each pixel in the image to its close
proximity of neighbor pixels. The size of the neighborhood and
its threshold are adaptable for the robust assessment. A pixel
that is dissimilar from a mainstream of its neighbors, as well as
being not logically aligned with those pixels to which it is
similar, is labeled as impulse noise. These noise pixels are then
substituted by the median pixel value of the pixels in the
neighborhood that have passed the noise labeling test [15]. The
following steps were used for back ground estimation.
Step 1: Estimate the background at time t using adaptive
median filter method.
Step 2: Subtract the estimated background from the input
frame.
Step 3: Apply a threshold m to the absolute difference to
get the binary moving objects hypothesis mask.
Assuming that the background is more likely to appear in a
scene, we can use the median of the previous n frames as the
background model
)), , , ( (( ) , , ( i t y x I t y x B O = (1)
, | )) , , ( ( ) , , ( | m > O = i t y x I t y x I (2)
where } 1 ,..., 2 , 1 , 0 { e n i and computation of

(.) O is
based on the AMF. The following are the algorithm for median
filter computation. Algorithm consists of two steps. Step 1 is
described for deciding whether the median of the gray values in
the size of the neighborhood or not.
Step 1:
) , ( )) , ( (
) , ( )) , ( (
max 2
min 1
y x y x
y x y x
o o =
o o =
O =
O =
,
If 0
1
> = && 0
2
< = then do the Step 2 otherwise
enlarge the neighborhood window size based on the maximum
allowed size of the input image. Step 1 is executed until
max
S S
w
> otherwise step 2 is computed.

Vol. 2, No. 6, 2011
79 | P a g e
Step 2:
) , ( ) , (
) , ( ) , (
max 2
min 1
y x y x G
y x y x G
o =
o =
=
=
,
If 0
1
> = && 0
2
< = then we have taken ) , ( y x G as an
output else )) , ( ( y x o O .
where ) , ( y x o is the AMF neighborhood size. ) , (
min
y x o is
the minimum gray level value in neighborhood ) , ( y x o ,
) , (
max
y x o represents the maximum gray level value in
) , ( y x o , )) , ( ( y x o O denotes the median of gray levels in
) , ( y x o , ) , ( y x G is the gray level at coordinates (x,y),
w
S
represents window size of the current process and
max
S is the
maximum allowed size of ) , ( y x o .
B. Foreground Detection
In this module estimated background and foreground mask
images are used as an input for further processing. Thus, we
use grayscale image sequences as input. Elements of the scene
and the sizes of the traffic objects (vehicles and pedestrians) are
unknown. The Foreground detection is done by using
accumulative difference method, which is change-detection
based on subtraction of a background image. It is necessary to
update the background image frequently in order to guarantee
reliable object detection. The basic idea in background
adaptation is to integrate the new incoming information into the
current background image using a Kalman filter:
t t t t t
D M a M a B B ] * 1 * [
2 1 ) 1 (
+ + =
+
, (3)
where
t
B represents the background model at time
t
D t, is
the difference between the present frame and the background
model, and
t
M is the binary moving objects hypothesis mask.
The gain are based on an estimate of the rate of
change of the background. The larger it is, the faster new
changes in the scene are updated to the background frame. In
our approach, , they are kept small
and the update process based on Eq.(3) is only intended for
adapting to slow changes in overall lighting.
>
=
Otherwise
T x D if
x M
t t
t
, 0
| ) ( | , 1
) ( . (4)
Foreground detection is started by computing a pixel based
absolute difference between each incoming frame ) , ( y x I and
an adaptive background frame ) , ( y x B
t
. The pixels are
assumed to contain motion if the absolute differences exceed a
predefined threshold level.
T
y x B y x I
y x F
t
>

=
o
u | ) , ( ) , ( |
) , ( (5)

As a result, a binary image is formed where active pixels
are labeled with a "1" and non-active ones with a "0". With the
updated background image strategy using Kalman filter, we get
the better foreground detection result. This is a simple, but
efficient method to monitor the changes in active during a few
consecutive frames. Those pixels which tend to change their
activity frequently are masked out from the binary image
representing the foreground detection result.
C. Shadow Removal
Shadows appear as surface features, when they are caused
by the interaction between light and objects. This may lead to
problems in scene understanding, object segmentation,
tracking, recognition, etc. Because of the undesirable effects of
shadows on image analysis, much attention was paid to the area
of shadow detection and removal over the past decades and
covered many specific applications such as traffic surveillance.
In this paper, 8-neighborhood gray clustering method is used to
define the precise shadow and remove it. The mean clustering
threshold and the initial cluster seed of the gray are calculated
by the following equations.
2
) ) , ( max( ) 3 / 1 (
i i
u y x G T = , (6)
where ) , ( y x G is a gray value of the pixel in
) , ( y x I
,
i
u
is the mean of ) , ( y x G . The initial seed
i
C
locates in the centre
of ) , ( y x G .The clustering starts from the seed
i
C
, and the
point
i
P
is examined in turn. If at least one point in the 8-
neighborhood of
i
P
has been marked as a shadow region,
standard deviation of
i
P
is calculated by the following
equation.
2
) ) , ( (
i i i
u y x P P = o , (7)
where
)) , ( ( y x I P
i
is the gray value of I(x,y). If
T P
i
< o
,
i
P

must be shadow point; otherwise, the point need not be
marked. The point
i
P
is checked constantly until no new point
is marked. At last all the marked shadow points are removed.
D. Noise removal
In regular practice due to the camera noise and irregular
object motion, there are some noise regions existed in both the
object and background regions. In our method we have
incorporated Gaussian noise with the acquired image and
propose a solution to see how the background subtraction
module would behave while the traditional background
algorithms are not providing the significant results. The focus
is on the background subtraction module because image noise
mostly impacts the foreground extraction process. If the
foreground objects are not detected well, the rest of the
modules will possibly fail at their tasks.
In the proposed method, after finding the foreground object,
noise is estimated and modeled using the following algorithm.
Step 1: Convert RGB image of ) , ( y x F into gray scale
image.
Step 2: Motion blurring can be estimated in a spatially
linear invariant system under certain conditions. If we assume
the object translates at a constant velocity V during the
1 2
a and a
1 2
0.1 0.01 a and a = =
Vol. 2, No. 6, 2011
80 | P a g e
exposure time T with angle , then the distortion d = VT and
define the point spread function (PSF) as

= s s
=
otherwise
d y d x
d
y x h
0
sin , cos | | 0 '
1
) , (
o o
, (8)

The motion blur can be described mathematically as the
result of a linear filter

) , ( ) , ( * ) , ( ) , ( y x y x F y x h y x g n + = ,
(9)

where ) , ( y x g denotes the blurred image, ) , ( y x f
denotes the original image, ) , ( y x n denotes additive noise
and * represents 2-D convolution.
Step 3: Dividing the Fourier transform of the PSF into the
transform of the blurred image, and performing an inverse FFT,
reconstruct the image without noise.

)) , ( ( ) ( * )) , ( ( y x g PSF y x f = , (10)
) (
)) , ( (
) , (
1
PSF
y x g
y x f , (11)

where ) , ( y x f is the original image and ) , ( y x g acquired
blurred image. (.) and (.)
1
represent Fourier and inverse
Fourier transform.
In order to estimate the point-spread-function (PSF) we
make use of autocorrelation functions ) (n K of an M pixel
image line l , which is defined as,
=
+ =
M
M i
i l n i l n K ) ( ) ( ) ( ] , [ M M n e , (12)
where 0 ) ( = i l outside the image line range. The above
Equation describes how pairs of pixels at particular
displacements from each other are correlated. It is high where
they are well correlated and low where poorly correlated. For a
normal image, the ) (n K will be some function of distance
from the origin plus random noises. But for a motion blurred
image, the ) (n K will decline much more slowly in the
direction of the blur than in other directions.
Step 4: The Noise to Signal Power Ratio was computed
using the following equations:
PSNR is measured in decibels. It's defined as

| |
=
=
1
0
1
0
2
) , ( ) , (
1
m
i
n
j
j i K j i I
mn
MSE
, (13)
Where i, j are the width and height of the frame,
respectively, in pixels. The PSNR is defined as
|
|
.
|
\
|
=
MSE
PSNR
MAXI
2
10
log 20
, (14)
Step 5: Apply Wiener filter with and d to deblur the
image.
Step 6: The Wiener filtering is employed on the resultant
Autocorrelation matrices. Wiener filtering minimizes the
expected squared error between the restored and perfect
images. A simplified Wiener filter is as follows:
) , (
) , ( / ) , ( ) , (
) , (
) , (
1
) , (
2
2
y x G
y x S y x S y x H
y x H
y x H
y x F
f
+
=
n
, (15)

where ) , ( y x H is the degradation function
) , ( ) , ( * ) , (
2
y x H y x H y x H = , (16)
) , ( * y x H is the complex conjugate of ) , ( y x H ,
2
) , ( ) , ( y x N y x S =
n
represents PSF of the noise,
2
) , ( ) , ( y x F y x S
f
= denotes the PSF of the original image
) , ( / ) , ( y x S y x S
f n
is called the noise to signal power ratio.
If noise power spectrum is zero, the Wiener filter reduces to
a simple inverse filter. If K is large compared with ) , ( y x H ,
then the large value of the inverse filtering term ) , ( / 1 y x H is
balanced out with the small value of the second term inside the
brackets.
Step 7: The Lucy-Richardson filtering on the resultant
Autocorrelation matrices. Using Lucy-Richardson method the
image is modeled by maximizing the likelihood function gives
an equation that is satisfied when the following iteration
converges
=
+
) , (
* ) , (
) , (
* ) , ( ) , (
) , (
1
y x f y x h
y x g
y x h y x f y x f
k
k k
, (17)
where * indicates the convolution. f
is the estimate of the

original image f at step k of the iteration. The iteration starts
with ) , ( * ) , (
0
y x h y x g f = and ) , ( y x g denotes the blurred
image, ) , ( y x h denotes the PSF.
E. Detection
After post-processing, the image is compared with the one
of the original frames (usually, the first frame). If the pixels are
less than certain threshold, then they are ignored. Otherwise,
they are replaced by the pixels of original image. This resulting
Vol. 2, No. 6, 2011
81 | P a g e
image will be consisting of the moving object ignoring the
background and hence satisfying our requirement.
IV. IMPLEMENTATION AND PERFORMANCE ANALYSIS
This system was implemented on an Intel Pentium IV 280
GHz PC. We have tested the system on image sequences on
different scenarios like traffic junction intersection, highways
etc. Real life traffic video sequence is used to demonstrate the
vehicle tracking from traffic video sequences using the
proposed framework. All the videos chosen for vehicle tracking
have same light intensity and have been taken during day time.
We convert the colour video frames to gray scale images.
Automatic monitoring visual surveillance system
implementation needs to detect vehicles using automatic
background extraction. Background subtraction is the main
step for vehicle detection. Fig. 4 shows number of successive
frames that are used to extract the background. Digital camera
used to take shots. The camera placed over the highway
directly. It shots eight frames per second.

Figure 4. Number of successive frames that are used for preprocessing.

Estimated background using Median Filter for n = 12. This
is shown in Fig.5
Figure 5. Estimated background image.

After applying the threshold, to the absolute difference
we got the binary moving objects hypothesis mask which is
shown in Fig. 6. With no noise information, the Wiener and
other filters do a poor job at realizing the original, non-
degraded, image. However, the Lucy-Richardson filter works
really well, despite having no information about the noise in
the image. With noise information, the Wiener filter gives
better result than other filters. Fig 7. shows the comparison of
PSNR value with varies filters.

Figure 6. Binary moving objects hypothesis mask.

Figure 7. Comparison of PSNR value with a mixture of filters.
Table I shows the result of a mixture of filters used in the Noise
removal module. Fig 8 shows the Comparative Analysis of
mixture of filters using chart.
Table I. Result of Nose removal module in decibel.
O
b
j
e
c
t

u
s
e
d

f
o
r

a
n
a
l
y
s
i
s

G
a
u
s
s
i
a
n

N
P
R

W
i
e
n
e
r

N
P
R

+

A
u
t
o
c
o
r
r
e
l
a
t
i
o
n

L
-
R

car 13 13 25 13 24
bus 12 13 24 13 22
motorcycle 14 15 30 16 28
Lorry 10 25 24 23 25
Truck 14 20 26 20 24
Vol. 2, No. 6, 2011
82 | P a g e

Figure 8. Comparative Analysis of a mixture of filters.
Fig.9 shows the foreground detected objects obtained after
background subtraction, shadow and noise. The automatic
background extraction results are very good and promising.
The most effective parameters that are playing a main role for
automatic background extraction are the threshold level. This
threshold is used to extract the moving vehicles from the
background. Matlab built-in function has been employed for
the evaluation of the threshold.
Table II gives vehicle detection results. The background is
subtracted from the current image then the resulted image is
filtered to get moving vehicles only. By using this technique
most of vehicles are detected. Moving vehicles are detected
easily after background is subtracted. Performance analysis is
shown in Fig.10.

Figure 9. Sequence of steps in the foreground object detection process.

TABLE II. The results for vehicle detection after noise removal
Type of
Vehicle

Actual
Number of
vehicles
Detected
Vehicles

Rate
%

Car 30 28 93
Motor cycle 30 29 97
Bus 30 27 90
Lorry 30 27 90
Truck 30 28 93

Figure.10 Performance analysis of the proposed method in detection different
vehicles in the noisy traffic environment.
V. CONCLUSION
The experimental results of using this approach lead to
detect moving vehicles efficiently. This algorithm has been
implemented and evaluated experimentally using natural traffic
images. It gives promising and effective results where an
average vehicle detection rate was around 94%. In this
approach the background subtraction and edge detection are
used. It is mainly focused on the autocorrelation method. And
then an adaptive algorithm is applied to autocorrelation. PSNR
is used for the evaluation to measure the performance of the
filters. It could be improved and used as a basis for automatic
traffic monitoring. Failure detection resulted from occluding
large vehicles with small ones and the far moving vehicles that
appear as a point in the image. These difficulties could be
solved in the future work.
ACKNOWLEDGMENT
Authors thank their family members and children for their
continuous support and consent encouragement to do this
research work successfully.
REFERENCES
[1] M. Cristani, M. Bicegi, and V. Murino, Integrated Region- and Pixel-
based Approach to Background Modeling, Proceedings of the
MOTION, 2002.
[2] H. Eng, J. Wang, A. Kam, and W. Yau, Novel Regionbased Modeling
for Human Detection within High Dynamic Aquatic Environment,
Proceedings on CVPR, 2004.
[3] P. KaewTraKulPong and R. Bowden, An Improved Adaptive
Background Mixture Model for Real-time Tracking with Shadow
Detection, In Proc. 2nd European Workshop on Advanced Video Based
Surveillance Systems, 2001.
0
5
10
15
20
25
30
35
G
a
u
s
N
P
R
W
i
e
n
e
r
N
P
R
+
L
-
R
vechicle type
d
b
car
bus
motorcycle
Lorry
Truck
0
5
10
15
20
25
30
35
93% 97% 90% 90% 93%
Car Motor
cycle
Bus Lorry Truck
Actual
Number of
Vehicles
Detected
Vehicles
Vol. 2, No. 6, 2011
83 | P a g e
[4] M. Harville, G. Gordon, and J. Woodfill, Foreground Segmentation
Using Adaptive Mixture Models in Color and Depth, Proc. ICCV
Workshop Detection and Recognition of Events in Video, July 2001.
[5] C. Stauffer and W.E.L. Grimson, Adaptive Background Mixture Models
for Real-Time Tracking, Proc. Conf. Computer Vision and Pattern
Recognition, vol. 2, pp. 246-252, June 1999.
[6] S.J. McKenna, Y. Raja, and S. Gong, Object Tracking Using Adaptive
Color Mixture Models, Proc. Asian Conf. Computer Vision, vol. 1, pp.
615-622, Jan. 1998.
[7] R. Cucchiara, C. Grana, M. Piccardi, and A. Prati, Detecting Moving
Objects, Ghosts, and Shadows in Video Streams, IEEE Trans. on
PAMI, 25: (10), October 2003.
[8] Connell, J., Detection and Tracking in the IBM PeopleVision System,
IEEE ICME, June 2004.
[9] Dongxiang Zhou, Hong Zhang and Nilanjan Ray, Texture Based
Background Subtraction, Proc. IEEE Int. Conf. on Information and
Automation, Zhangjiajie, China, pp. 601-605, 2008.
[10] A. McIvor, Background subtraction techniques, in Proc. Image Video
Computing, pp. 147-153, 2000.
[11] Y. Ivanov, A. Bobick, and J. Liu, Fast lighting independent background
subtraction, Int. Journal on Comp. Vision, vol. 37, no. 2, pp. 199207,
Jun. 2000.
[12] Liyuan Li, Weimin Huang, Irene Yu-Hua Gu and Qi Tian, Statistical
Modeling of Complex Backgrounds for Foreground Object Detection,
IEEE Transactions on Image Proc., vol. 13, no. 11, pp. 1459-1472, Nov
2004.
[13] A. Elgammal, D. Harwood, and L.S. Davis, Non-Parametric Model for
Background Subtraction, Proc. IEEE Intl Conf. Computer Vision 99
FRAME-RATE Workshop, 1999.
[14] M. Sankari, Dr. C. Meena "Adaptive Background Estimation and object
detection applying in Automated Visual surveillance" International
Journal of Computer science and Information Security July 2010, Vol. 8
No. 4.
[15] Peng Lei, Adaptive median filtering, Machine vision, 140.429 Digital
image processing.

AUTHORS PROFILE
Mrs. M. Sankari received her B.Sc. and M.Sc.
degrees in Computer science from Bharathidasan
University in 1988 and 1990, respectively. She has
completed her Master of Philosophy degree in Computer
science from Regional Engineering College, Trichy in
2000. Presently, she is a Assistant Professor & Head in the department of MCA
at NIET. She is pursuing her Ph.D. degree in Computer science at
Avinashilingam University for women, Coimbatore, India. She has published
several technical papers in IEEE/IET conferences and Journal. Her field of
research includes Computer vision, Pattern recognition, Analysis of algorithms,
Data structure, Computer graphics and multimedia.

Dr. C. Meena received her B.Sc (Physics), and
Master of Computer Applications degrees from Madurai
Kamaraj University in 1987 and 1990, respectively. She
has completed her Ph.D. degree in Computer science from
Bharathiyar University in 2006. Presently, she is Head,
Computer Centre at Avinashilingam University,
Coimbatore, India. She is guiding for funded Research Project in UGC and
Naval Research Board. She has presented international/national journals. She
has published various technical papers in IEEE/IET conferences and national
level conferences. Her field of research includes Image processing, Computer
vision and Pattern recognition.
Vol. 2, No. 6, 2011
84 | P a g e
Improvement of Brain Tissue Segmentation Using
Information Fusion Approach

Lamiche Chaabane
Department of Computer Science
University of Msila
Algeria
Moussaoui Abdelouahab
Department of Computer Science
University of Setif
Algeria

Abstract The fusion of information is a domain of research in
full effervescence these last years. Because of increasing of the
diversity techniques of images acquisitions, the applications of
medical images segmentation, in which we are interested,
necessitate most of the time to carry out the fusion of various data
sources to have information with high quality. In this paper we
propose a system of data fusion through the framework of the
possibility theory adapted for the segmentation of MR images.
The fusion process is divided into three steps : fuzzy tissue maps
are first computed on all images using Fuzzy C- Means
algorithm. Fusion is then achieved for all tissues with a fusion
operator. Applications on a brain model show very promising
results on simulated data and a great concordance between the
true segmentation and the proposed system.
Keywords- Information fusion; possibility theory; segmentation;
FCM; MR images.
I. INTRODUCTION
Recent technical advances have led to the multiplication of
imaging systems, which are often used for observing a
phenomenon from different points of view. They provide a
large amount of information that must a whole in order to draw
correct conclusions. This development has made data fusion in
image processing an important step, now well recognized, in
modern multi source image analysis. In medical imaging in
particular the clinician may use images issued from different
sources, each of them highlighting specific properties of tissues
and pathologies. They may be images acquired with a single
imaging technique using different acquisition parameters (for
instance multi echo MRI), or images were obtained from
several imaging techniques (for instance anatomical MR
imaging combined with functional PET imaging). The
association of such images allows the medical expert to
confirm and to complete his diagnosis because it can arrive as
none the images available contains a sufficient information
separately. In most data fusion problems, the images to be
combined are partly redundant as they represent the same
scene, and partly complementary as they may highlight
different characteristics [1].
Typically, none of the images provide a completely
decisive and reliable information. In addition, the information
is often imprecise and uncertain, and these characteristics are
inherent to the images; due to observed phenomenon, sensors,
numerical reconstruction algorithms and resolution etc. The
aim of data fusion techniques is therefore to improve the
decision by increasing the amount of global information, while
decreasing its imprecision and uncertainty by making us of
redundancy and complementarities [1].
Some mathematical models were discussed in the literature
for the modelization of both uncertainty and imprecision.
Traditionally probabilities theory was the primary model used
to deal with uncertainty problems, but they suffer from
drawbacks which are still a matter of discussion. Whereas the
Dempster-Shafer theory also allows to representing these two
natures of information using functions of mass but the set of
operators used by this theory in fusion step is very restricted.
Alternative to this approach is the possibility theory where
uncertainty and imprecision are easily modeled, in this article
we will focus on this last one for two essential reasons : this
theory allows to combining information coming from various
sources by the use a wide range of available combination
operators. In addition, this theory seems to us the most adapted
to the considered problem in the modeling step [1][2].
We present in this article, a fuzzy information fusion
framework for the automatic segmentation of human brain
tissues using T2- weighted (T2) and proton density (PD)
images. This framework consists of the computation of fuzzy
tissue maps in both images by means of Fuzzy C-Means
algorithm, the creation of fuzzy maps by a combination
operator and a segmented image is computed in decision step.
The organization of the paper as follows : In section II,
some previous related works are presented. In section III, we
briefly outline the principals of possibility theory reasoning.
Section IV discusses the architecture of the fusion system.
Steps of fusion in possibility theory are explained in Section
V. In section VI the traditional FCM algorithm is briefly
reviewed. Section VII presents our proposed approach. Some
experiment results using two routing MR sequences T2 and PD
feature images are shown in section VIII and section IX
contains conclusions and addresses future works.
II. RELATED WORKS
Many works have been done in the field of fuzzy
information fusion in the literature. A brief review of some of
them is presented in this section. Waltz [6] presented three
basic levels of image data fusion : pixel level, feature level and
decision level, which correspond to three processing
architectures. I. Bloch [1] have outlined some features of
Dempster-Shafer evidence theory, which can very useful for
Vol. 2, No. 6, 2011
85 | P a g e
medical image fusion for classification, segmentation or
recognition purposes. Examples were provided to show its
ability to take into account a large variety of situations.
Registration-based methods are considered as pixel-level
fusion, such as MRI-PET (position emission tomography) data
fusion [7]. Some techniques of knowledge-based segmentation
can be considered as the feature-level fusion such as the
methods proposed in [11].
Some belief functions, uncertainty theory, Dempster-Shafer
theory are often used for decision-level fusion such as in [9]. In
[12], I. Bloch proposed an unified framework of information
fusion in the medical field based on the fuzzy sets, allow to
represent and to process the numerical data as well as symbolic
systems, the fuzzy sets theory is applied to three levels: at the
low level to treat the basic numerical information contained in
the images, as well as possible ambiguity between the classes;
on the level object, to represent objects or structures in the
images such as a fuzzy objects. at the higher level, to take into
account a structural information and some characteristics as the
distance, adjacency, and the relative position between objects.
V. Barra and J. Y. Boire [4] have described a general
framework of the fusion of anatomical and functional medical
images. The aim of their work is to fuse anatomical and
functional information coming from medical imaging, the
fusion process is performed in possibilistic logic frame, which
allows for the management of uncertainty and imprecision
inherent to the images. They particularly focus on the
aggregation step with the introduction of a new class of
operators based on information theory and the whole process is
finally illustrated in two clinical cases : the study of
Alzheimers disease by MR/SPECT fusion and the study of
epilepsy with MR/PET/SPECT. The obtained results was very
encouraging.
V. Barra and J. Y. Boire [10] proposed a new scheme of
information fusion to segment intern cerebral structures. The
information is provided by MR images and expert knowledge,
and consists of constitution, morphological and topological
characteristics of tissues. The fusion of multimodality images is
used in [8]. In [3], the authors have presented a framework of
fuzzy information fusion to automatically segment tumor areas
of human brain from multispectral magnetic resonance imaging
(MRI) such as T1-weighted, T2-weighted and proton density
(PD) images; in this approach three fuzzy models are
introduced to represent tumor features for different MR image
sequences. They allow to create corresponding fuzzy feature
space of tumor. All the t-norm or fuzzy intersection operators
can be used as fusion operators for this fuzzy features. the
geometric mean is chosen using experiments allowing us to
take correctly into account the three fuzzy spaces in a simple
way. The fuzzy region growing is used to improve the fused
result.
Maria del C and al [5] proposed a new multispectral MRI
data fusion technique for white matter lesion segmentation, in
that a method is described and comparison with thresholding in
FLAIR images is illustrated.

III. THE POSSIBILITY THEORY
Possibilistic logic was introduced by Zadeh (1978)
following its former works in fuzzy logic (Zadeh, 1965) in
order to simultaneously represent imprecise and uncertain
knowledge. In fuzzy set theory, a fuzzy measure is a
representation of the uncertainty, giving for each subset Y of
the universe of discourse X a coefficient in [0,1] assessing the
degree of certitude for the realization of the event Y. In
possibilistic logic, this fuzzy measure is modeled as a measure
of possibility Hsatisfying:
0 ) ( 1 ) ( = H = H o et X

) ( ) ( )) ( (
i
i
i
i
i
Y Sup Y Y H = H

An event Y is completely possible if 1 ) ( = H Y and is
impossible if
0 ) ( = H Y
. Zadeh showed that H could
completely be defined from the assessment of the certitude on
each singleton of X. Such a definition relies on the definition
of a distribution of possibility t satisfying :
] 1 , 0 [ : X t
{ } 1 ) ( / ) ( = x Sup x x
x
t t
Fuzzy sets F can then be represented by distributions of
possibility, from the definition of their characteristic function
F
u
:
) ( ) ( ) ( x x X x
F
t u = e
Distributions of possibility can mathematically be related to
probabilities, and they moreover offer the capability to declare
the ignorance about an event. Considering such an event A
(e.g., voxel v belongs to tissue T, (where v is at the interface
between two tissues), the probabilities would assign
5 . 0 ) ( ) ( = = A P A P , whereas the possibility theory allows fully
possible 1 ) ( ) ( = H = H A A . We chose to model all the
information using distributions of possibility, and equivalently
we represented this information using fuzzy sets [18].
The literature classically distinguishes three modes for
combination of uncertainty and imprecise information in a
possibility theory framework [23] :
The conjunction: gather the operators of t-norms (fuzzy
intersection), this mode of combination must be used if
measurements are coherent, i.e. without conflict.
The compromise: gather the median operator and some
average operators, it must be used when measurements are in
partial conflict.
The Disjunction: gather the operators of t-conorms (fuzzy
union), it must be used when measurements are in disaccord,
i.e. in severe conflict.
In introduction, we underlined the inopportunity to
combining information in a fixed mode: if observations are in

Vol. 2, No. 6, 2011
86 | P a g e
accord, it is legitimate to combine them in a conjunctive mode
or compromise in order to extract a more relevant information.
But if a serious conflict appears, it is better to combining in a
disjunctive mode. For example, if two measurements of the
same parameter prove completely different, it is not judicious
to make an average of it, better is worth to say than one or the
other is true [24].
IV. THE FUSION PROCESS AND TYPE
OF ARCHITECTURES
A general information fusion problem can be stated in the
following terms : given l sources S
1
, S
2
,S
l
representing
heterogeneous data on the observed phenomenon, take a
decision d
i
on an element x, where x is higher level object
extracted from information, and Di belongs to a decision space
D={d
1
, d
2
, d
3
,, d
n
} (or set of hypotheses). In numerical fusion
methods, the information relating x to each possible decision d
i

according to each source S
j
is represented as a number M
ij

having different properties and different meanings depending
on the mathematical fusion framework. In the centralized
scheme , the measures related to each possible decision i and
provided by all sources are combined in a global evaluation of
this decision, taking the form, for each i : Mi = F(M
i1
, M
i2
, M
i3
,
, M
in
), where F is a fusion operator. Then a decision is taken
from the set of M
i
, 1in. in this scheme, no intermediate
decision is taken and the final decision is issued at the end of
the processing chain. In decentralized scheme decisions at
intermediate steps are taken with partial information only,
which usually require a difficult control or arbitration step to
diminish contradictions and conflicts [2][4].
The three-steps fusion can be therefore described as :
- Modeling of information in a common theoretical
frame to manage vague, ambiguous knowledge and
information imperfection. In addition, in this step the
M
ij
values are estimated according to the chosen
mathematical framework.
- Combination : the information is then aggregated with
a fusion operator F. This operator must affirm
redundancy and manage the complementarities and
conflicts.
- Decision : it is the ultimate step of the fusion, which
makes it possible to pass from information provided by
the sources to the choice of a decision d
i
.
V. FUSION IN POSSIBILITY THEORY
A. Modeling Step
In the framework of possibility theory and fuzzy sets
[13][14[15], the M
ij
s represent membership degrees to a fuzzy
set or possibility distribution t , taking the form for each
decision d
i
and source S
i
:. ) (
i j ij
d M t = .
B. Fusion step
For the aggregation step in the fusion process, the
advantages of possibility theory rely in the variety of
combination operators, which must affirm redundancy and
manage the complementarities. And may deal with
heterogeneous information [16][17][18]. It is particular interest
to note that, unlike other data fusion theories like Bayesian or
Dempster-Shafer combination, possibility theory provides a
great flexibility in the choice of the operator, that can be
adapted to any situation at hand [2].
C. Decision step
Is usually taken from maximum of memberships values
after the aggregation step. Many constraints can be added to
this decision, typically for checking for the reliability of the
decision (is he obtained value high enough?) or for the
discrimination power of he fusion (is the difference between
the two highest values high enough ?) [2].
VI. THE FCM ALGORITHM CLUSTERING
Clustering is a process of finding groups in unlabeled
dataset based on a similarity measure between the data patterns
(elements) [12]. A cluster contains similar patterns placed
together. The fuzzy clustering technique generates fuzzy
partitions of the data instead of hard partitions. Therefore, data
patterns may belong to several clusters, having different
membership values with different clusters. The membership
value of data pattern to a cluster denotes similarity between the
given data pattern to the cluster. Given a set of N data patterns
X={x
1
, x
2
, x
3
, , x
n
} the Fuzzy C-Means (FCM) clustering
algorithm minimizes the objective function [26][27]:
) , ( ) , , (
2
1 1
i j
C
i
N
j
m
ij
b x d u X U B J
= =
=
(1)
Where x
j
is the j-th P-dimensional data vector, b
i
is the
center of cluster i, u
ij
is the degree of membership of x
j
in the j-
th cluster, m is the weighting exponent d
2
(x
j
,b
i
) is the Euclidean
distance between data x
j
and cluster center b
i
.
The minimization of objective function J(B,U,X) can be
brought by an iterative process in which updating of
membership u
ij
and the cluster centers are done for each
iteration.

1
1
) 1 /( 2
2
2
) , (
) , (
|
|
.
|
\
|
=

C
k
m
k j
i j
ij
b x d
b x d
u
. (2)

.
1
1
=
=
=
N
k
m
ik
N
k
k
m
ij
i
u
x u
b
(3)
Where :

{ } { }
| |
< <
e
e e
=
N
i
ij
ij
N u
u
N j C i
1
0
1 , 0
.. 1 , .. 1
. (4)
{ } 1 .. 1
1
= e

=
C
i
ij
u N j . (5)
Vol. 2, No. 6, 2011
87 | P a g e
The algorithm of the FCM consists then of the reiterated
application of (2) and (3) until stability of the solutions.
VII. PROPOSED METHOD
In this section we propose a framework of data fusion
based on the possibility theory which allows the segmentation
of MR images. The operation is limited to the fusion of T2 and
PD images. Then information to combine are thus
homogeneous and the scheme of our proposed fusion system as
shown in figure 1 below:

Figure 1. Scheme of the proposed fusion system.
If it is supposed that these images are registered, our
approach of fusion consists of three steps:
A. Modeling of the data
In this phase the choice of the fuzzy framework is retained
to modeling information resulting from the various images.
More precisely, MR images are segmented in C = 4 classes
using the FCM algorithm described in section VI. For each MR
image I, C distributions of possibility
I
T
t , 1 sT sC are then
obtained, and are represented by memberships of the pixels to
the classes.
B. Combination
The aggregation step is most fundamental for a relevant
exploitation of information resulting from the images I
T2
, I
PD
.
The operator must combine for a given tissue T, the
distributions of possibility
2 T
t
and
PD
t
, by underlining the
redundancies and managing ambiguities and
complementarities between the T2-weighted and proton density
images.
1) Choice of an operator : One of the strengths of the
possibility theory is to propose a wide range of operators for
the combination of memberships. I. Bloch [20] classified these
operators not only according to their severe (conjunctive) or
cautious (disjunctive) nature but also with respect to their
context-based behavior. Three classes were thus defined:

- Context independent and constant behavior operators
(CICB);
- Context independent and variable behavior operators
(CIVB);
- Context dependent operators (CD).
For our T2/PD fusion, we chose a (CICB) class of
combination operators because in the medical context, both
images were supposed to be almost everywhere concordant,
except near boundaries between tissues and in pathologic areas
[20]. Three operators (minimum, maximum, and arithmetic
mean) of this class who does not need any parameter were
tested related to the fusion of MR images acquired in
weighting T2, PD. They were carried out on a range of 70
slices of Brain1320 volume of Brainweb
1
If
2 T
T
t ,
PD
T
t are
the possibility distributions of tissue T derived from T2 and PD
maps , then the fused possibility as defined for any gray level v
as :
The minimum operator: )) ( ), ( ( ) (
2
v v Min v
PD
T
T
T T
t t t =
The maximum operator: )) ( ), ( ( ) (
2
v v Max v
PD
T
T
T T
t t t =
The arithmetic mean operator: 2 / )) ( ) ( ( ) (
2
v v v
PD
T
T
T T
t t t + =
These operators are compared with the reference result
using the coefficient DSC
2
. Which measures the overlap
between two segmentations S1 and S2. It is defined as:
) 2 1 ( / ) 2 1 ( . 2 ) 2 , 1 ( S S card S S card S S DSC + =
The results of these tests are shown on figure 2:

Figure 2. Comparison of the operators by the DSC measurement.
The results drawn up in the figure 2 show the predominance
of the minimum operator compared to the maximum operator
and the arithmetic mean operator. Thus we will retaining this
operator for our study.
C. Decision
A segmented image was finally computed using all maps of
different tissues T, 1sT sC. So certain theories make it
possible to consider several types of decision, the theory of the
possibilities proposes only the rule of the maximum of

1
http://www.bic.mni.mcgill.ca/brainweb/
2
Dice Similarity Coefficient
0.75
0.80
0.85
0.90
0.95
CSF WM GM
CSF 0.85 0.82 0.84
WM 0.95 0.92 0.91
GM 0.88 0.83 0.87
Min Max M. Arth.
Membership degrees u
T2
Membership degrees u
PD

FCM Classification

Possibilistic combination rule
FCM Classification
Decision rule
Result of fusion
T2 image PD image
Vol. 2, No. 6, 2011
88 | P a g e
possibility. We thus retain this one and assign each pixel to the
tissue for which it has the greatest membership.
The general algorithm of our system is .

General algorithm
Modeling of the image
For i in {T2,PD} do
FCM (i) { Computation of membership degrees
for both images}
End For
Fusion
Possibilistic fusion {Between each class of T2 image
and the same one of PD image }
Decision
Segmented image

It should be noted that the stability of our system depend to
the stability of the algorithm used in the modeling step[26].
VIII. RESULTS AND DISCCUSION
Since the ground truth of segmentation for real MR images
is not usually available, it is impossible to evaluate the
segmentation performance quantitatively, but only visually.
However, Brainweb provides a simulated brain database (SBD)
including a set of realistic MRI data volumes produced by an
MRI simulator. These data enable us to evaluate the
performance of various image analysis methods in a setting
where the truth is known [30][31][32].
to have tests under realistic conditions, three volumes were
generated with a thickness of 1 mm and a level of noise of 0%,
3% and 5%. We fixed at 20% the parameter of heterogeneity.
The results of each step of fusion are presented on a noisy
90th brain only slice is shown in figure 3. This noisy slice was
segmented into four clusters: background, CSF, white matter,
and gray matter using FCM algorithm, however the
background was neglected from the viewing results.

Simulated MR T2 image Simulated MR PD image
(a)

CSF WM GM
Figure 3. (a) Simulated T2, PD images illustrate the fusion. (b) Maps of
CSF, WM and GM obtained by FCM for T2 image. (c) Maps of CSF, WM
and GM obtained by FCM for PD image . (d) Maps of CSF, WM and GM
obtained by proposed system.
The results of final segmentation are shown in figure 4 below.

(a) (b) (c)
Figure 4. Segmentation results. (a) T2 segmented by FCM (b) PD
segmented by FCM, (c) Image of fusion
The CSF map of PD image is improved significantly by fusion within
the noise levels 0% 5%.
The WM fused map is strongly improved compared to that
obtained by the PD only, but This improvement is small
compared to that obtained by the segmentation of T2 only.
Information in GM fused map reinforced in area of
agreement (mainly in the cortex). and the fusion showed a
significant improvement and reduces the effect of noise in
images.

(b)
(c)
(d)
Vol. 2, No. 6, 2011
89 | P a g e
0.00
0.20
0.40
0.60
0.80
1.00
CSF WM GM
CSF 0.92 0.71 0.85
WM 0.92 0.89 0.95
GM 0.85 0.76 0.88
T2 PD Fusion T2/PD
These remarks demonstrate the superior capabilities of the
proposed approach compared to the taking into account of only
one weighting in MR image segmentation.
The performances of our system led us to reflect on the
validity of the segmentation obtained. It appeared to us to
measure and quantify the performances of our segmentation of
the whole of brain. Measurement used is the DSC coefficient
described in section 7 and the results are reported in figures 5, 6
and 7.
The graphics of figures 5, 6 and 7 underline the advantages
of the fusion of multimodality images within the fuzzy
possibilistic framework to improve the results clearly. DSC
coefficients obtained by the proposed approach augments the
improvement of the segmentation from 2% to 3% for the white
matter and from 1% to 3% for the gray matter in T2 image.
And from 2% to 12% for the white matter and from 3% to 19%
for the grey matter in PD. Image. Moreover one indeed notes
that the CSF is improved only compared to the weighting PD,
in that the improvement increases by 7% to 21%.

Figure 5. Comparison results between different segmentations with 0% noise

IX. CONCLUSION
In this article we presented a system of data fusion to
segment MR images in order to improve the quality of the
segmentation. Since we outlined in here some features of
possibility theory, which can be very useful for medical images
fusion. And which constitute advantages over classical
theories. They include the high flexibility of the modeling
offered by possibility theory, taking into account both
imprecision and uncertainty and prior information not
necessarily expressed as probabilities. The Effectiveness of our
system is affirmed by the choice of the model to representing
data and the selected operator in a combination step. Results
obtained are rather encouraging and underline the potential of
the data fusion in the medical imaging field.
As a perspective of this work and on the level of modeling
we would wish to integrate other information or new
techniques of MR acquisitions and thus to use a more effective
and more robust algorithms to representing a data. on fusion
level an adaptive operators of fusion are desired for the
combination of the data in order to improve the segmentation
of the MR images or to detect anomalies in the pathological
images.
REFERENCES
[1] I. Bloch, Some aspects of Dempster-Shafer evidence theory for
classification of multi-modality medical images taking partial volume
effect into account, Pattern Recognition Letters, vol. 17, pp. 905919,
1996.
[2] I. Bloch, and H. Maitre, Data fusion in 2D and 3D image processing:
An overview, X Brazilian symposium on computer graphics and
image processing, Brazil, pp. 127134, 1997.
[3] W. Dou, S. Ruan, Y. Chen, D. Bloyet, J. M. Constans, A framwork of
fuzzy information fusion for the segmentation of brain tumor tissues on
MR images, Image and vision Computing, vol. 25, pp. 164171, 2007
[4] V. Barra and J. Y. Boire, A General Framework for the Fusion of
Anatomical and Functional Medical Images, NeuroImage, vol. 13,
410424, 2001.
[5] D. C. Maria, H. Valds, J. F. Karen, M. C. Francesca, M. W. Joanna,
New multispectral MRI data fusion technique for white matter lesion
segmentation: method and comparison with thresholding in FLAIR
images, Eur Radiol, vol. 20, 16841691, 2010.
[6] E.D Waltz, the principals and practice of image and spatial data fusion
in : Davis L, Hall, James llinas (Eds), Proceedings of the Eight, national
Data Fusion Conference ,(Dalls, TX. March 15-17, 1995) Handbook of
Multisensor Data Fusion, CRC pess, West Bam Beach, FL,1995, pp.
257278. (p41418).
[7] F. Behloul, M. Janier, P. Croisille, C. Poirier, A. Boudraa, R.
Unterreiner, J. C. Mason, D. Revel, Automatic assessment of
myocardial viability based on PET-MRI data fusion, Eng. Med. Biol.
Mag., Proc. 20th Ann. Int. Conf. IEEE, vol. 1, pp. 492495, 1998.
[8] M. Aguilar, R. Joshua, New fusion of multi-modality volumetric
medical imagery, ISIF, pp. 12061212, 2002.
[9] E. Lefevre, P. Vannoorenberghe, O. Colot, About the use of
Dempster-Shafer theory for color image segmentation, First
international conference on color in graphics and image processing,
Saint-Etienne, France, 2000.
[10] V.Barra, and J-Y Boire, Automatic segmentation of subcortical brain
structures in MR images using information fusion, IEEE Trans. on
Med. Imaging., vol. 20, n7, pp. 549558, 2001.
[11] M. C. Clarck, L.O. Hall, D.B. Goldgof, R Velthuizen, F.R. Murtagh,
M.S. Silbiger, Automatic tumor segmentation using knowledge-based
techniques, IEEE Trans. Med. Imaging, vol. 17, pp. 187201, 1998.
[12] I. Bloch, Fusion of numerical and structural image information in
medical imaging in the framework of fuzzy sets, Fuzzy Systems in
Medicine, ed. par P. Szczepaniak et al., pp. 429447, Springer Verlag,
2000.
0.00
0.20
0.40
0.60
0.80
1.00
CSF WM GM
CSF 0.92 0.62 0.83
WM 0.86 0.76 0.88
GM 0.76 0.59 0.78
T2 PD Fusion T2/PD
0.70
0.75
0.80
0.85
0.90
0.95
1.00
CSF WM GM
CSF 0.93 0.80 0.87
WM 0.94 0.94 0.96
GM 0.89 0.87 0.90
T2 PD Fusion T2/PD
Vol. 2, No. 6, 2011
90 | P a g e
[13] L. A. Zadeh, Fuzzy sets, Information and Control, vol. 8, pp. 338
353, 1965.
[14] L. A. Zadeh, fuzzy sets as a basic for a theory of possibility, Int. Jour.
of Fuzzy Sets and Systems, vol. 1, pp. 328, 1978.
[15] D. Dubois and H. Prade. Fuzzy sets and systems : Theory and
Application, New-York: Academic Press, 1980.
[16] D. Dubois and H. Prade, A review of Fuzzy Set Aggregation
Connectives, Information Sciences, vol. 36, pp. 85121, 1985.
[17] R. R. Yager, Connectives and quantifiers in fuzzy sets, Int. Jour. of
Fuzzy sets and systems, vol. 40, pp. 3975,1991.
[18] D. Dubois and H. Prade, Combination of information in the
framework of possibility theory, In M. AL Abidi et al., editor, Data
Fusion in Robotics and Machine Intelligence: Academic Press, 1992.
[19] V. Barra, Fusion d'images 3D du cerveau : Etude de modles et
Applications. Thse de doctorat, Universit dAuvergne, 2000.
[20] I. Bloch, Information combination operators for data fusion : A
Comparative Review with Classification, IEEE Transactions en
systems, Man. and Cybernitics, vol. 1, pp. 5267, 1996.
[21] W. Dou, Segmentation des images multispectrales base sur la fusion
d'information: application aux images IRM. Thse de doctorat
Universit CAEN/Basse-Normandi , 2006.
[22] A.Bendjebbour, W.Pieczynski, Segmentation d'images multisenseurs
par fusion videntielle dans un contexte markovien, Traitement du
Signal, vol. 14, n15, pp. 453463, 1997.
[23] H.J. Zimmermann. Fuzzy Sets Theory and its Applications. 2nd ed.
Boston: Kluwer Academic Publishers, 1991.
[24] S. Deveughelle, B. Dubuisson, Adaptabilit et combinaison
possibiliste: application la vision multi camras, Traitement du
Signal, vol. 11, n6, pp. 559568, 1994.
[25] C.Barillot, J.C.Gee,L.Le Briquer,G.Le Goualher, Fusion intra et inter
individus en imagerie mdicale applique la modlisation anatomique
du cerveau humain, Traitement du Signal, vol. 11, n6, 1994.
[26] J. Bezdek, A convergence theorem for the fuzzy data clustering
algorithms, IEEE Transactions on Pattern Analysis and Machine
Intelligence TPAMI, vol. 2, pp. 18,1980.
[27] J. Bezdek , L. Hall, L. Clarke, Review of MR image segmentation
using pattern recognition, Medical Physics, vol. 20, pp. 10331048,
1993.
[28] V.Barra, J. H. Boire, Segmentation floue des tissus crbraux en IRM
3D: une approche possibiliste versus autres mthodes, Rencontres
Francophones sur la logique floue et ses applications, Valenciennes,
Editions Cpadus, pp. 193198, 1999.
[29] I.Bloch, H. Maitre, Fusion de donnes en traitement d'images :
modles d'informations et dcisions, Traitement du Signal, vol. 11, n6,
pp. 435446, 1994.
[30] BrainWeb [Online]. Available: www.bic.mni.mcgill.ca/brainweb/
[31] C. A. Cocosco, V. Kollokian, R. K. -S. Kwan, and A. C. Evans,
BrainWeb: Online interface to a 3D MRI simulated brain database,
NeuroImage, pt. 2/4, vol. 5, n4, p. S425, 1997.
[32] D. L. Collins, A. P. Zijdenbos, V. Kollokian, J. G. Sled, N. J. Kabani,
C. J. Holmes, and A. C. Evans, Design and construction of a realistic
digital brain phantom, IEEE Trans. Med. Imaging, vol. 17, n3, pp.
463468, 1998.
[33] I. Bloch, Fusion d'informations en traitement du signal et des images,
unpublished.
AUTHORS PROFILE
Chaabane Lamiche received his BSc in Computer Science in 1997 from the
Department of Computer Science from Ferhat Abbas University, Algeria. He
also received Master's degree in Computer Science in 2006 from University of
M'sila. He has 4 years experience in teaching. His areas of interests include
Data mining and Warehousing, artificial intelligence, Image processing and
operational research. His current research interests include the data mining
techniques and medical image analysis.
Abdelouahab Moussaoui received his BSc in Computer Science in 1990
from the Department of Computer Science from the University of Science and
Technology of Houari Boumedienne (USTHB), Algeria. He also received and
MSc in Space Engineering in 1991 from University of Science and
Technology of Oran (USTO). He received also a MSc degree in Machine
Learning from Reims University (France) since 1992 and Master's degree in
Computer Science in 1995 from University of Sidi Bel-abbes, Algeria and
PhD degree in Computer Science from Ferhat Abbas University, Algeria. He
is IEEE Member and AJIT, IJMMIA & IJSC Referee. His researches are in
the areas of clustering algorithms and multivariate image classification
applications. His current research interests include the fuzzy neuronal network
and non parametric classification using unsupervised knowledge system
applied to biomedical image segmentation. He also works from a long time on
pattern recognitions algorithm, complex data mining and medical image
analysis.

Vol. 2, No. 6, 2011
91 | P a g e
Image Retrieval using DST and DST Wavelet
Sectorization

Dr. H.B.Kekre
Sr. Professor, Computer Engineering Dept.
MPSTME, SVKMs NMIMS (Deemed-to be University)
Vile Parle West, Mumbai, INDIA
Dhirendra Mishra
Associate Professor and PhD. Research scholar
MPSTME, SVKMs NMIMS (Deemed-to be University)
Vile Parle West, Mumbai, INDIA

AbstractThe concept of sectorization of the transformed images
for CBIR is an innovative idea. This paper introduces the concept
of Wavelet generation for Discrete sine transform (DST).The
sectorization of the DST transformed images and the DST
wavelet transforms has been done into various sector sizes
i.e.4,8,12, and 16. The transformation of the images is tried and
tested threefold i.e. row wise transformation, column wise
transformation and Full transformation. We have formed two
planes i.e. plane 1 and plane 2 for sectorization in full
transformation .The performance of the all the approaches has
been tested by means of the three plots namely average precision-
recall cross over point plot, LIRS (Length of initial relevant
string of images) plot, LSRR (Length of string to recover all
relevant images in the database) plot. The algorithms are
analyzed to check the effect of three parameters on the retrieval
First the way of transformation (Row, Column, Full), Second the
size of sector generated, Third the type of similarity measures
used. With the consideration of all these the overall comparison
has been performed.
Keywords-CBIR, Feature extraction; Precision; Recall; LIRS;
LSRR; DST; DST Wavelet.
I. INTRODUCTION
The digital world Innovations has evolved itself to a very
large extent. The result of which has increased the more and
more dependency on the digital data and in turn on computer
system. The information of any form i.e. multimedia,
documents, images etc. everything has got its own place in this
digital world. The computer system has been accepted to be
the very powerful mechanism to use these digital data, for its
secured storage and efficient accessibility whenever required.
Digital Images play a very good role for describing the
detailed information about man, money, machine etc. almost
in every field. The various processes of digitizing the images
to obtain it in the best quality for the more clear and accurate
information leads to the requirement of more storage space
and better storage and accessing mechanism in the form of
hardware or software. As far as the accessing of these images
are concerned one needs to have the good mechanism of not
only for accessing of the images but also for any other image
processing to be done one needs to have the faster, accurate,
efficient retrievals of these images. There are various
approaches of proposing the methodologies of retrieving the
images from the large databases consisting of millions of
images stored.
Content Based Image Retrieval (CBIR) [1-4] is one of the
evolving fields of image processing. CBIR needs to have the
innovative algorithm to extract the perfect features to define
identity of an image. It has been researched upon to use
content of the image itself to draw out its unique identity. This
unique identity can make one to differentiate the images with
each other with better and accurate retrieval of images. There
are mainly three contents i.e. shape, color and textures of the
image as of now is being experimented by many researchers.
These contents leads one to extract the exact feature of the
image which can be well utilized to compare with all images
available in the database by means of some similarity
measures like Euclidean distance, sum of absolute difference
etc. The tremendous use of images in the digital world of
today has proved the CBIR as very useful in several
applications like Finger print recognition, Iris Recognition,
face recognition, palm print recognition, speaker
identification, pattern matching and recognition etc.
There are various approaches which have been
experimented to generate the efficient algorithm for image
feature extraction in CBIR. These approaches advocate
different ways of extracting features of the images to improve
the result in the form of better match of the query image in the
large database. Some papers discuss the variation in the
similarity measures in order to have its lesser complexity and
better match [5-15]. Methods of feature extraction using
Vector Quantization [16], bit truncation coding [17,18],Walsh
Transform[20,21] has also provided the new horizon to the
feature extraction methodology. The method of sectorization
has already been experimented on DCT [22], DST [23], DCT-
DST Plane [24] , Haar Wavelet [25] and Kekres Transform
[26] earlier.
This paper proposes the use of sectorization of DST
Wavelet for feature extraction in CBIR. The outcome of which
has been compared with the DST sectorization performance.
II. DST WAVELET
A. Generation of DST Wavelet[4]
The wavelet analysis procedure is to adopt a wavelet
prototype function, called an analyzing wave or mother wave.
Other wavelets are produced by translation and contraction of
the mother wave. By contraction and translation infinite set of
functions can be generated. This set of functions must be
orthogonal and this condition qualifies a transform to be a
Vol. 2, No. 6, 2011
92 | P a g e
wavelet transform. Thus there are only few functions which
satisfy this condition of orthogonality. Generation of DST
Wavelet transform matrix of size N
2
xN
2
from DST matrix of
size NxN is given in [4]. However in this case we require DST
matrix of size 128x128 where 128 is not a square. To resolve
this situation, this paper proposes an algorithm to generate
discrete sine wavelet transform from discrete sine transform of
8x8 and 16x16.
In this paper the DST Wavelet has been generated using
the contraction and translation. Due to the size of images in
the database is 128x128 we need the wavelet transform to be
of size 128x128. The 128x128 Wavelet transform matrix
generated from 16x16 orthogonal DST matrix and 8x8 DST
matrix. First 16 rows of Wavelet transform matrix are
generated by repeating every column of DST matrix of
dimension 16x16, 8 times. To generate next 17 to 32 rows,
second row of DST (8X8) is translated using groups of 8
columns with horizontal and downward shifts. To generate
next 33 to 48 rows, third row of DST (8X8) matrix is used in
the same manner. Like wise to generate last 113 to 128 rows,
8th row of transform DST (8x8) matrix is used. Note that by
repeating every column of the basic transform 8 times we get
global components. Other wavelets are generated by using
rows of DST matrix of size 8x8 giving local components of
the DST Wavelet.
III. FEATURE VECTOR GENERATION
The figure given below shows the formal steps of feature
vector generation in brief.

Figure 1 Steps to generate the feature vectors
A. Sectorization [7-13][23-28]
The individual components of row and column wise
transformed (DST and DST Wavelet separately) images are
distributed into different co-ordinates of Cartesian coordinate
system according to their sign change to form four sectors.
The even rows/columns components of the transformed image
and the odd rows/columns components of the transformed
images are checked for positive and negative signs. The even
and odd DST values are assigned to each quadrant
The division of each of these 4 sectors into 2 partitions
forms the 8 sectors which distributes the transformed image
components into its appropriate sectors.
Continuing the same division concept further the 4 sectors
already generated has been divided into 3 parts with
consideration of 30 degree angle to generate sector sizes of 12.
Each sector of 8 sectors are individually divided into two
to obtain 16 sectors. For each individual sector sizes the mean
of each sectors are taken as the feature vector component. The
feature vector components of each plane i.e. R, G and B has
been calculated and concatenated together with the average of
first row/column and last row/column to form the final feature
vector for each method separately. The size of the feature
vector and its component varies for each sector sizes the
maximum sector size will have maximum feature vector
components.
The feature database consists of feature vectors of all
images. The features of the query image extracted are
compared with the feature database using similarity measures.
There are two similarity measures used in this experiment i.e.
Eucledian distance (ED) and sum of absolute difference (AD)
as given in the equation (1) and (2) shown below:
( ) ( )
()
( ) | |

()
The match of query image feature with the feature
database with the minimum value of ED/AD gives the perfect
match.
The performance measure of the algorithms proposed are
done with the calculation of the precision-recall and LIRS
(Length of initial relevant string of retrieval) and LSRR
(Length of string to recover all relevant images in the
database) refer equations (3) to (6).

()

()

()

()

The performance of the proposed methods are checked by
means of calculating the class wise average performance and
overall average performance of each approach with respect to
the transformation method applied, the way of applying the
transformation i.e. row wise, column wise, full, sector sizes
used and type of similarity measures used.
IV. EXPERIMENTAL RESULTS
A. Image Database
The sample Images of the augmented Wang database [29]
consists of 1055 images having 12 different classes such as
Cartoons, Flowers, Elephants, Barbie, Mountains, Horses,
Images are
transdormed using
DST and DST Wavelet
seperately
Transformed images
are sectored in
different sizes i.e.
4,8,12 and 16
[ 7-13],[23-26]
Feature vector
generated taking the
mean of each sectors
with augmentation of
extra components
Vol. 2, No. 6, 2011
93 | P a g e
Buses, Sunset, Tribal, Beaches, Monuments and Dinosaur
shown in the Figure 2.

Figure 2. Sample images in the Database
The class wise distribution of all images in its respective
classes such as ther are 46 images of cartoon in the class, there
are 100 images for flower and so on are shown in the Figure 3
below.

Figure3. Class wise distribution of images
B. Sectorization of Row wise Transformed (DST) images.
The class wise precision-recall cross over point plot for
sectorization of row wise DST transformed images has been
shown in the Figure 4. The x axis of the plot denotes the class
of images in the database and the y axis denotes the average
precision-recall cross over point plots of five randomly
selected images per class. These values are taken into
percentage. Comparing all classes of images it is found that
the dinosaur class has the best retrieval result of more than
80% for 12 and 16 sectors with sum of absolute difference as
similarity measure. Other classes like flower, horse, sunset and
elephants have the retrieval up to 68% (12 and 16 sectors with
AD), 55% (4,8,12 sectors with ED), more than 50% (16
sectors with AD) and more than 50% (12 sectors with AD)
respectively. Looking at the LIRS plot for the same (as shown
in the Figure 5) to check the performances it has been found
that the initial length of string containing the relevant images
must be more. In this case the higher value of LIRS is very
clearly visible for the best performer class i.e. dinosaur class.
The LIRS varies for all other classes from 1% to 20% which
indicates that the first image is always relevant image. The
Figure 6 shows the LSRR Plot checks for the length of string
containing all relevant images retrieved. It must be minimum
which is very much seen for dinosaur class.

Figure4. Class wise Average Precision-Recall cross over point plot of DST
row wise sectorization for all sector sizes with respect to similarity measures
i.e. Euclidian distance (ED) and sum of absolute difference (AD)

Figure5. Class wise LIRS plot of DST row wise sectorization for all sector
sizes with respect to similarity measures i.e. Euclidian distance (ED) and sum
of absolute difference (AD)

Figure 6. Class wise LSRR plot of DST row wise sectorization
for all sector sizes with respect to similarity measures i.e. Euclidian distance
(ED) and sum of absolute difference (AD)
C. Sectorization of Row wise Transformed (DST Wavelet)
images
This section of the paper discusses the performance of the
DST Wavelet (row wise) sectorization with respect to all three
performance measuring parameters namely average precision-
recall cross over point plot (see Figure7), LIRS(see Figure8)
and LSRR(see Figure9). Comparing the performance of all the
classes within once again the dinosaur class outperforms all
classes with 90% retrieval for 12 sectors with sum of absolute
difference as similarity measure. This performance is better
than the simple DST (row wise) sectorization as discussed in
section 4.2. The flower class has retrieval of more than 65% in
DST Wavelet whereas it is below 60% in the case of DST.
Horse class has the result of 50%,The cartoon class has
improved a lot goes up to 60% compared to only 45% in
normal DST sectorization. The LIRS and LSRR shows its
relevant plots in the Figure8 and Figure9.
D. Sectorization of Column wise Transformation (DST)
There are 12 classes of the images used in the database.
The performance of the algorithm varies from class to class.
The class wise average precision-recall cross over point
plotted din the Figure10 shows the performance of the
sectorization of DST (column wise) in various sectors.
Vol. 2, No. 6, 2011
94 | P a g e

Wavelet row wise sectorization for all sector sizes with respect to similarity
measures i.e. Euclidian distance (ED) and sum of absolute difference (AD)

Figure 8. Class wise LIRS plot of DST Wavelet row wise sectorization for all
sector sizes with respect to similarity measures i.e. Euclidian distance (ED)
and sum of absolute difference (AD)

Figure9. Class wise LSRR plot of DST Wavelet row wise sectorization for all
sector sizes with respect to similarity measures i.e. Eucledian distance (ED)
Looking at the class wise performance the Dinasaour class
has the best retrieval close to 80%, Flower class reaches up to
70% (16 sectors with AD), sunset and horses class has the
retrieval more than 50%.
The elephant class has the resultant retrieval rate close to
50% whereas the performance for the Barbie class is more
than 40%. The LIRS and LSRR which keeps check on the
performance evaluation of the method shown in the Figure11
and Figure 12.The maximum f LIRS has been achieved for the
dinosaur class for the combination of all sectors and the sum
of absolute difference.

column wise sectorization for all sector sizes with respect to similarity
measures i.e. Euclidian distance (ED) and sum of absolute difference (AD)

Figure 11. Class wise LIRS plot of DST column wise sectorization for all

Figure12. Class wise LSRR plot of DST column wise sectorization for all

Vol. 2, No. 6, 2011
95 | P a g e
E. Sectorization of Column wise Transformation (DST
Wavelet).
The sectorization of DST Wavelet has better performance
of average precision-recall cross over points as it can be easily
seen in the Figure 13.There is increase in the performance of
the Dinosaur, Flowers, Cartoon, Barbie classes i.e. 85%(12
and 16 sector with AD), 75% (16 sectors with ED), 42%(16
sectors with AD), 50%(16 sectors with AD) respectively.
The LIRS performances of the method are very
interestingly depicts the high rises for the better performance
of precision-recall. The minimum value of LSRR for the
dinosaur, flower, sunset classes depicts the good performance.

Wavelet column wise sectorization for all sector sizes with respect to
similarity measures i.e. Euclidian distance (ED) and sum of absolute
difference (AD)

Figure14. Class wise LIRS plot of DST Wavelet column wise sectorization for
all sector sizes with respect to similarity measures i.e. Euclidian distance (ED)

Figure15. Class wise LSRR plot of DST Wavelet column wise sectorization
for all sector sizes with respect to similarity measures i.e. Euclidian distance
(ED) and sum of absolute difference (AD)
F. Overall Comparison of all Approaches.
The overall performance of all approaches gives very clear
idea about the overall average performance of the retrieval
rates as shown in the Figure16 Figure 18. The overall
average precision-recall cross over point plot as shown in the
Figure16 depicts that on average performance of the retrieval
for all methods proposed is 40%. It is observed that In most of
the cases the sectorization with the sum of absolute difference
as similarity measure has better retrieval than Euclidian
distance. As far as sector sizes are concerned all have good
performance except for 8 sectors in DST-WT (Row).

Figure16. Overall Average Precision-Recall cross over point plot comparison
of DST and DST Wavelet Sectorization for all sector sizes with respect to
similarity measures i.e. Euclidian distance (ED) and sum of absolute
difference (AD)
Vol. 2, No. 6, 2011
96 | P a g e

Figure17. Overall Average LIRS plot comparison of DST and DST Wavelet
Sectorization for all sector sizes with respect to similarity measures i.e.
Euclidian distance (ED) and sum of absolute difference (AD)

Figure18. Overall Average LSRR plot comparison of DST and DST Wavelet
Sectorization for all sector sizes with respect to similarity measures i.e.
Euclidian distance (ED) and sum of absolute difference (AD)
V. CONCLUSION
This paper discusses the new idea of generating the DST
wavelet transform and sectoring the transformed image into
4,8,12 and 16 sectors. The results obtained clearly show that
the performance of DST Wavelet transform is better than DST
using sectorization of full-Plane1 and full-plane2
transformations with the sum of absolute difference giving
best results close to 45% (see the Figure 16). The similarity
measure plays vital role in the applications of CBIR.
Similarity measures provide the efficient and faster access and
matching of the images in the database. We have employed
and analyzed two similarity measures i.e Euclidian Distance
(ED) and sum of absolute difference (AD).The sum of
absolute difference as shown in the equation (2) has lesser
complexity of calculation and better retrieval rates in the
approaches applied for all sector sizes (see figure 16). The
performance measurement of all approaches has been checked
by means of LIRS and LSRR which are very powerful tools to
comment on the retrieval rate of any algorithm. The maximum
LIRS and the Minimum LSRR are the best measures for it.

REFERENCES
[1] Kato, T., Database architecture for content based image retrieval in
Image Storage and Retrieval Systems (Jambardino A and Niblack W
eds),Proc SPIE 2185, pp 112-123, 1992.
[2] Ritendra Datta,Dhiraj Joshi,Jia Li and James Z. Wang, Image
retrieval:Idea,influences and trends of the new age,ACM Computing
survey,Vol 40,No.2,Article 5,April 2008.
[3] John Berry and David A. Stoney The history and development of
fingerprinting, in Advances in Fingerprint Technology, Henry C. Lee
and R. E. Gaensslen, Eds., pp. 1-40. CRC Press Florida, 2
nd
edition,
2001.
[4] H.B.Kekre, Archana Athawale, Dipali sadavarti, Algorithm to generate
wavelet transform from an orthogonal transform, International journal
of Image processing, Vol.4(4), pp.444-455
[5] H. B. Kekre, Dhirendra Mishra, Digital Image Search & Retrieval using
FFT Sectors published in proceedings of National/Asia pacific
conference on Information communication and technology(NCICT 10)
5
TH
& 6
TH
March 2010.SVKMS NMIMS MUMBAI
[6] H.B.Kekre, Dhirendra Mishra, Content Based Image Retrieval using
Weighted Hamming Distance Image hash Value published in the
proceedings of international conference on contours of computing
technology pp. 305-309 (Thinkquest2010) 13th & 14
th
March 2010.
[7] H.B.Kekre, Dhirendra Mishra,Digital Image Search & Retrieval using
FFT Sectors of Color Images published in International Journal of
Computer Science and Engineering (IJCSE) Vol.
02,No.02,2010,pp.368-372 ISSN 0975-3397 available online at
http://www.enggjournals.com/ijcse/doc/IJCSE10-02- 02-46.pdf
[8] H.B.Kekre, Dhirendra Mishra, CBIR using upper six FFT Sectors of
Color Images for feature vector generation published in International
Journal of Engineering and Technology(IJET) Vol. 02, No. 02, 2010,
49-54 ISSN 0975-4024 available online at
http://www.enggjournals.com/ijet/doc/IJET10-02- 02-06.pdf
[9] H.B.Kekre, Dhirendra Mishra, Four walsh transform sectors feature
vectors for image retrieval from image databases, published in
international journal of computer science and information technologies
(IJCSIT) Vol. 1 (2) 2010, 33-37 ISSN 0975-9646 available online at
http://www.ijcsit.com/docs/vol1issue2/ijcsit2010010201.pdf
[10] H.B.Kekre, Dhirendra Mishra, Performance comparison of four, eight
and twelve Walsh transform sectors feature vectors for image retrieval
from image databases, published in international journal of
Engineering, science and technology(IJEST) Vol.2(5) 2010, 1370-1374
ISSN 0975-5462 available online at
http://www.ijest.info/docs/IJEST10-02-05-62.pdf
[11] H.B.Kekre, Dhirendra Mishra, density distribution in walsh transfom
sectors ass feature vectors for image retrieval, published in international
journal of compute applications (IJCA) Vol.4(6) 2010, 30-36 ISSN
0975-8887 available online at
http://www.ijcaonline.org/archives/volume4/number6/829-1072
[12] H.B.Kekre, Dhirendra Mishra, Performance comparison of density
distribution and sector mean in Walsh transform sectors as feature
vectors for image retrieval, published in international journal of Image
Processing (IJIP) Vol.4(3) 2010, ISSN 1985-2304 available online at
http://www.cscjournals.org/csc/manuscript/Journals/IJIP/Volume4/Issue
3/IJIP-193.pdf
[13] H.B.Kekre, Dhirendra Mishra, Density distribution and sector mean
with zero-sal and highest-cal components in Walsh transform sectors as
feature vectors for image retrieval, published in international journal of
Computer scienece and information security (IJCSIS) Vol.8(4) 2010,
ISSN 1947-5500 available online http://sites.google.com/site/ijcsis/vol-
8-no-4-jul-2010
[14] Arun Ross, Anil Jain, James Reisman, A hybrid fingerprint matcher,
Intl conference on Pattern Recognition (ICPR), Aug 2002.
[15] A. M. Bazen, G. T. B.Verwaaijen, S. H. Gerez, L. P. J. Veelenturf, and
B. J. van der Zwaag, A correlation-based fingerprint verification
system, Proceedings of the ProRISC2000 Workshop on Circuits,
Systems and Signal Processing, Veldhoven, Netherlands, Nov 2000.
[16] H.B.Kekre, Tanuja K. Sarode, Sudeep D. Thepade, Image Retrieval
using Color-Texture Features from DCT on VQ Codevectors
obtained by Kekres Fast Codebook Generation, ICGST International
Vol. 2, No. 6, 2011
97 | P a g e
Journal on Graphics, Vision and Image Processing (GVIP),
Available online at http://www.icgst.com/gvip
[17] H.B.Kekre, Sudeep D. Thepade, Using YUV Color Space to Hoist the
Performance of Block Truncation Coding for Image Retrieval, IEEE
International Advanced Computing Conference 2009 (IACC09), Thapar
University, Patiala, INDIA, 6-7 March 2009.
[18] H.B.Kekre, Sudeep D. Thepade, Image Retrieval using Augmented
Block Truncation Coding Techniques, ACM International Conference
on Advances in Computing, Communication and Control (ICAC3-2009),
pp.: 384-390, 23-24 Jan 2009, Fr. Conceicao Rodrigous College of
Engg., Mumbai. Available online at ACM portal.
[19] H.B.Kekre, Tanuja K. Sarode, Sudeep D. Thepade, DCT Applied to
Column mean and Row Mean Vectors of Image for Fingerprint
Identification, International Conference on Computer Networks and
Security, ICCNS-2008, 27-28 Sept 2008, Vishwakarma Institute of
Technology, Pune.
[20] H.B.Kekre, Sudeep Thepade, Archana Athawale, Anant Shah,
Prathmesh Velekar, Suraj Shirke, Walsh transform over row mean
column mean using image fragmentation and energy compaction for
image retrieval, International journal of computer science and
engineering (IJCSE),Vol.2.No.1,S2010,47-54.
[21] H.B.Kekre, Vinayak Bharadi, Walsh Coefficients of the Horizontal &
Vertical Pixel Distribution of Signature Template, In Proc. of Int.
Conference ICIP-07, Bangalore University, Bangalore. 10-12 Aug 2007.
[22] H.B.Kekre, Dhirendra Mishra, DCT Sectorization for feature vector
generation in CBIR, International journal of computer Applications
(IJCA),Vol.9,No.1,pp.19-26
[23] H.B.Kekre, Dhirendra Mishra, DST Sectorization for Feature vector
generation, Universal journal of computer science and and Engineering
Technology (UniCSE),Vol.1, No.1, Oct.2010,pp.6-15,Available online
at
http://www.unicse.org/index.php?option=com_content&view=article&id
=54&Itemid=27
[24] H.B.Kekre, Dhirendra Mishra, DCT-DST Plane sectorization of row
wise transformed color images in CBIR, International journal of
engineering science and technology, Vol.2, No.12, Dec.2010, pp.7234-
7244, ISSN No.0975-5462. Available at
http://www.ijest.info/docs/IJEST10-02-12-143.pdf
[25] H.B.Kekre, Dhirendra Mishra, Sectorization of Haar and Kekres
Wavelet for feature extraction of color images in image retrieval,
International journal of computer science and information security
(IJCSIS), USA, Vol.9, No.2, Feb 2011, pp.180-188,
http://sites.google.com/site/ijcsis/volume-9-no-2-feb-2011
[26] H.B.Kekre, Dhirendra Mishra, Sectorization of Kekres transform for
image retrieval in content based image retrieval, Journal of
Telecommunication (JOT),UK, Vol.8, No.1 April 2011, pp. 26-33.
http://sites.google.com/site/journaloftelecommunications/volume-8-
issue-1-april-2011
[27] H.B.Kekre, Dhirendra Mishra, Sectorization of DCT- DST Plane for
column wise transformed color images in CBIR, International
Conference of Technology Systems & Management (ICTSM-2011) held
at SVKMs NMIMS Mumbai India, published in Springer Link CCIS
145, pp. 5560, 2011. Available online at
http://www.springerlink.com/content/m573256n53r07733/
[28] H.B.Kekre, Dhirendra Mishra, Full DCT sectorization for Feature
vector generation in CBIR, Journal of graphics, vision, image
processing, Vol.11, No.2, April 2011, pp. 19 30
http://www.icgst.com/gvip/Volume11/Issue2/P1151041315.html
[29] Jia Li, James Z. Wang, ``Automatic linguistic indexing of pictures by a
statistical modeling approach,'' IEEE Transactions on Pattern Analysis
and Machine Intelligence, vol. 25, no. 9, pp. 1075-1088, 2003
AUTHORS PROFILE

H. B. Kekre has received B.E. (Hons.) in
Telecomm. Engg. from Jabalpur University in 1958,
M.Tech (Industrial Electronics) from IIT Bombay in
1960, M.S.Engg. (Electrical Engg.) from University of
Ottawa in 1965 and Ph.D.(System Identification) from
IIT Bombay in 1970. He has worked Over 35 years as
Faculty and H.O.D. Computer science and Engg. At IIT
Bombay. From last 13 years working as a professor in Dept. of Computer
Engg. at Thadomal Shahani Engg. College, Mumbai. He is currently senior
Professor working with Mukesh Patel School of Technology Management and
Engineering, SVKMs NMIMS University vile parle west Mumbai. He has
guided 17 PhD.s 150 M.E./M.Tech Projects and several B.E./B.Tech Projects.
His areas of interest are Digital signal processing, Image Processing and
computer networking. He has more than 350 papers in National/International
Conferences/Journals to his credit. Recently twelve students working under
his guidance have received the best paper awards. Two research scholars
working under his guidance have been awarded Ph. D. degree by NMIMS
University. Currently he is guiding 10 PhD. Students. He is life member of
ISTE and Fellow of IETE.

Dhirendra Mishra has received his BE (Computer
Engg) degree from University of Mumbai. He completed
his M.E. (Computer Engg) from Thadomal shahani Engg.
College, Mumbai, University of Mumbai. He is PhD
Research Scholar and working as Associate Professor in
Computer Engineering department of, Mukesh Patel
School of Technology Management and Engineering,
SVKMs NMIMS University, Mumbai, INDIA. He is life member of Indian
Society of Technical education (ISTE), Member of International association
of computer science and information technology (IACSIT), Singapore,
Member of International association of Engineers (IAENG). He has more than
30 papers in National/International conferences and journals to his credit. His
areas of interests are Image Processing, Operating system, Information
Storage and Management.
Vol. 2, No. 6, 2011
98 | P a g e
Performance Analysis of UMTS Cellular Network
using Sectorization Based on Capacity and Coverage

A.K.M Fazlul Haque, Mir Mohammad Abu Kyum,
Md. Baitul Al Sadi, Mrinal Kar
Department of Electronics and Telecommunication
Engineering.
Daffodil International University
Md. Fokhray

Hossain
Department of Computer Science and Engineering.
Daffodil International University

AbstractUniversal Mobile Telecommunications System (UMTS)
is one of the standards in 3rd generation partnership project
(3GPP). Different data rates are offered by UMTS for voice,
video conference and other services. This paper presents the
performance of UMTS cellular network using sectorization for
capacity and coverage. The major contribution is to see the
impact of sectorization on capacity and cell coverage in 3G
UMTS cellular network. Coverage and capacity are vitally
important issues in UMTS cellular Network. Capacity depends
on different parameters such as sectorization, energy per bit
noise spectral density ratio, voice activity, inter-cell interference
and intra-cell interference, soft handoff gain factor, etc and
coverage depends on frequency, chip rate, bit rate, mobile
maximum power, MS Antenna Gain, EIRP, interference Margin,
Noise figure etc. Different parameters that influence the capacity
and coverage of UMTS cellular network are simulated using
MATLAB 2009a. In this paper, the outputs of simulation for
increasing amount of sectorization showed that the number of
users gradually increased. The coverage area also gradually
increased.
Keywords-UMTS; Capacit; Coverage and data rates; sectoring.
I. INTRODUCTION
A cellular cell can be divided into number of geographic
areas, called sectors. It may be 3 sectors, 4 sectors, 6 sectors
etc. When sectorization is done in a cell, interference is
significantly reduced resulting in better performance for
cellular network. Capacity in WCDMA standards of UMTS
refers to maximum number of users per cell, where the area
covered by RF signal from Node B or UE (User Equipment) is
called coverage area of UMTS. Capacity and coverage are two
dynamic phenomena in UMTS network. Parameters that define
capacity and coverage of UMTS are dynamic in nature, where
increasing or decreasing values of these parameters affects
capacity and coverage of UMTS cellular network. One of the
parameters is sectorization in UMTS. There are some works on
sectorization scheme [1-4].
Bo Hagerman, Davide Imbeni and Jozsef Barta considered
WCDMA 6-sector deployment case study of a real installed
UMTS-FDD network [1]. Romeo Giuliano, Franco Mazzenga,
Francesco Vatalaro described Adaptive Cell Sectorization for
UMTS Third Generation CDMA Systems [2]. Achim Wacker,
Jaana Laiho-Steffens, Kari Sipila, and Kari Heiska considered
the impact of the base station sectorisation on WCDMA radio
network performance [3]. S. Sharma, A.G. Spilling and A.R.
Nix considered Adaptive Coverage for UMTS Macro cells
based on Situation Awareness [4]. Most of the works analyzed
the performance considering sectors with static parameters but
it is needed to analyze the performance along with all dynamic
parameters.
This paper optimizes the performance of both capacity and
coverage of UMTS not only considering sectors but also with
dynamic parameters as energy per bit noise spectral density
ratio, voice activity, inter-cell interference, soft handoff factor,
and data rates.
II. BACKGROUND
A. Capacity in WCDMA for UMTS:
As the downlink capacity of UMTS is related to transmit
power of Node B and uplink capacity is related to numbers of
users, uplink capacity is considered in this paper.
If the number of users is N
s
then for a single CDMA cell,
the number of users will be [5],
o
n 1
)
/
/
( 1
S N E
R W
N
o b
s
+ = (i)
Where, N
s=
total number of users, W=chip rate,
R= base band information bit rate, E
b
/N
o
=Energy per bit to
noise power spectral density ratio, = background thermal
noise, S=signal power=S1-P (d)-shadow fading, S1=UE power,
P (d) =Propagation loss.
For WCDMA, the chip rate is 3.84 Mcps [8], and the
channel bandwidth is 5 MHz [8]. It is also necessary to
consider the affects of multiple cells or intra-cell interference
()[12], cell sectoring(D)[6], soft handover factor(H)[11],
Array antenna gain (A
g
)[10].Thus the capacity for WCDMA in
UMTS yields:

g
o b
s
A H D
S N E
R W
N
+
+ =
o |
n
) 1 (
1
)
/
/
( 1 (ii)
B. Coverage and data rates in WCDMA for UMTS:
UMTS offered different data rates for multi services. Table
1 shows different standard bit rates offered by UMTS. Higher
class of service makes cell radius small resulting in small
Vol. 2, No. 6, 2011
99 | P a g e
coverage area. If different class of services is classified in terms
of coverage area, it will look like figure 1.
TABLE 1: DIFFERENT CLASSES OF SERVICES
Bit Rate(Kbps) Class
12.2 Class 5
32 Class 4
64 Class 3
144 Class 2
384 Class 1

Figure 1 shows that for service class 1, maximum distance
is observed by UE (User Equipment) from Node B. Similarly
for service class 2 and service class 3, UEs maintain maximum
distance from Node B. From this figure it is clear that different
coverage areas are needed to maintain different data rates. So
coverage area needs to increase for better class of services.
This paper optimizes the coverage area for particular services
with sectors.
Figure 2 shows a UMTS cell where Node B received power
(P
R
) from User Equipment (UE). The Node B sensitivity is the
power level for minimum signal necessary at the input of the
Node B receiver to meet requirements in terms of E
b
/N
o,
processing gain

(G
p
) and Node B interference and noise power
given as [8]
Node B sensitivity=E
b
/N
o
-G
p
+N
Node B interference and noise power

Where G
p
=Processing gain
|
.
|
\
|
= |
.
|
\
|
=
R
Mcps
Bitrate
Chiprate 84 . 3
log 10 log 10

Figure 1: Different Classes of Services vs. Maximum Distance

Figure 2: UMTS cell

Now the maximum allowable path loss for Node B,
L
P
=EIRP-Node b sensitivity + G
p
- fast fading margin-----
(iii)
from radio propagation model, Path loss for dense urban area
[5],
3 log ) log
55 . 6 9 . 44 ( 4.97 )] h [log(11.75
3.2 log 82 . 13 ) log( 9 . 33 3 . 46
2
UE
+
+ +
+ =
d h
h f L
NodeB
b c

---------------------------------------------------------------------- (iv)
From equation (iii) and (iv) a relationship can be expressed
for coverage and data rates in dense urban case,
Margin fading Fast -
rate Chip
log 10
y sensitivit NodeB - EIRP 3 log ) log
55 . 6 9 . 44 ( 4.97 )] h [log(11.75
3.2 log 82 . 13 ) log( 9 . 33 3 . 46
2
UE
|
.
|
\
|
+
= +
+ +
+
R
d h
h f
NodeB
b c

where d is the coverage radius and R is the data rates.
After calculating the cell range d, the coverage area can be
calculated. The coverage area for one cell in hexagonal
configuration can be estimated with [9]
Coverage area, S =K.d
2

where S is the coverage area, d is the maximum cell range,
and K is a constant. In Table 2, some of the K values are listed.
TABLE 2: K VALUES FOR THE SITE AREA CALCULATION [9]:
Site
configuration

Omni
or no
sector
Two
sectors

Three
sectors

Four
sectors

Value of K

2.6 1.3 1.95 2.6

Vol. 2, No. 6, 2011
100 | P a g e
III. SIMULATIONS AND RESULTS
The analysis has been done for capacity and coverage with
sectoring cell for dense urban using MATLAB R2009a. The
simulated values for sectorization are shown in Table 3, Table
4, Table 5, Table 6 and Table 7. The performances are also
described in Figure 3, Figure 4, Figure 5, Figure 6, and Figure
7. The algorithms of the evaluation process have also been
introduced in appendix.
Figure 3 shows that Energy per bit to noise spectral density
ratio (E
b
/N
o
) needs to maintain small value for increasing
number of simultaneous 384 Kbps users. From this figure it is
observed that for dynamic value of E
b
/N
o
with changing the
sectors, the number of simultaneous 384 Kbps data users
increased or decreased. For example, if E
b
/N
o
value is 4 db,
then for 6 sectors, the number of simultaneous users will be 88
but for 3 sectors, the number of simultaneous users will be 45.
Thus the dynamic values of E
b
/N
o
can be the increasing or
decreasing factors in UMTS and sectorization scheme can be
effective in this case.

Figure 3: Number of simultaneous 384 Kbps users vs. Eb/No in sectors cell
TABLE 3: SIMULATED VALUES FOR NUMBER OF SIMULTANEOUS
384 KBPS USERS VS. EB/NO IN SECTORS CELL

Energy per
bit to Noise
spectral
density
ratio(E
b
/N
o
)
Users
without
sector
Users
with 2
sectors
Users
with 3
sectors
Users
with 4
sectors
Users
with 6
sectors
1 59.065 117.13 175.19 233.26 349.39
4 15.516 30.032 44.548 59.065 88.097
8 8.2581 15.516 22.774 30.032 44.548
10 6.8065 12.613 18.419 24.226 35.839
14 5.1475 9.2949 13.442 17.59 25.885
16 4.629 8.2581 11.887 15.516 22.774
18 4.4156 7.4516 10.677 13.903 20.355
20 3.9032 6.8065 9.7097 12.613 18.419

The interference from other cell is known as inter-cell
interference (). For multi-cell configuration, the number of
outer cells can reduce cell capacity in UMTS. Figure 4 shows,
for increasing demand of users the value of in UMTS needs
to be small. Figure 4 also represents dynamic inter-cell
interference with changing of sectors, where number of
simultaneous 384 Kbps data users increases or decreases. From
Figure 4 it has been observed that for increasing value of , it is
needed to increase sectors.

Figure 4: Number of simultaneous 384 Kbps users vs.inter-cell interference in
sectors cell

The overlapped cell can lead an extra power thus
introducing soft handover factor (H) in a UMTS cell. The value
of H in UMTS can be a factor to increase the number of users.
Figure 5 shows that for increasing H and changing value of
sectorization the number of simultaneous 384 Kbps data users
increases.
For example, if H value is 2.5 db, then for 2 sectors the
number of simultaneous users will be 195 but for 4 sectors the
number of simultaneous users will be 388.
Figure 6 shows that the number of voice users depends on
the value voice activity factors ().This is true only for 12.2
Kbps voice users, not for data users, as for data services it will
always be 1.
Figure 3 also shows that, for increasing amount voice users
the value of in UMTS needs to as small as possible. Varying
and changing the sectors the number of simultaneous voice
users from figure 6 is observed.

Vol. 2, No. 6, 2011
101 | P a g e
USERS VS. INTER-CELL INTERFERENCE IN SECTORS CELL

USERS VS. SOFT HANDOVER FACTOR IN SECTORS CELL

Figure 5: Number of simultaneous 384 Kbps users vs. soft handover factor in
sectors cell

Figure 6: Number of simultaneous voice users vs. voice activity factor in
sectors cell.

USERS VS. VOICE ACTIVITY FACTOR IN SECTORS CELL

Inter-
cell
interfer
ence
Users
without
sector
Users
with
2
sectors
Users
with
3
sectors
Users
with
4
sectors
Users
with
6 sectors
0.1 135.33 419.6 730.73 1201 5401
0.5 27.866 84.721 146.95 241 1081
1 14.433 42.86 73.973 121 541
1.5 9.9552 28.907 49.649 81 361
1.7 8.9017 25.624 43.925 71.588 53.941
2.0 7.7164 21.93 37.486 61 271
Soft
Hand
-over
Factor
Users
without
sector
Users
with
2 sectors
Users
with
3 sectors
Users
with
4 sectors
Users
with
6 sectors
0.1 4.871 8.7419 12.613 16.484 24.226
0.4 16.484 31.968 47.452 62.935 93.903
1 39.71 78.419 117.13 155.84 233.26
1.5 59.065 117.13 175.19 233.26 349.39
2.5 97.774 194.55 291.32 388.1 581.65
3 117.13 233.26 349.39 465.52 697.77
Voice
activity
factor
Users
without
sector
Users
with
2 sectors
Users
with
3 sectors
Users
with
4 sectors
Users
with
6 sectors
0.2 228.31 455.63 682.94 910.26 1364.9
0.4 114.66 228.31 341.97 455.63 682.94
0.6 76.771 152.54 228.31 304.09 455.63
0.8 57.828 114.66 171.49 228.31 341.97
1 46.463 91.926 137.39 182.85 273.78
Vol. 2, No. 6, 2011
102 | P a g e
TABLE 7: SIMULATED VALUES FOR COVERAGE VS. DATA RATES
IN DENSE URBAN USING COST 231 MODEL IN SECTORS CELL

Figure 7: Coverage vs. bit rates for dense urban using COST 231 model in
sectors cell

Finally, consider for coverage vs. data rates in dense urban
area, where operating frequency is considered 2000 MHz with
COST 231 Model as a radio propagation model. In Figure 7 the
x axis represents data rate in Kbps and y axis represents
coverage area in meter
2
with cell radius in meter. Parameters
that are related to coverage setting first, then for increasing data
rates in x axis, the coverage area is observed in y axis. From
figure 1, it is known that for higher data rates, the coverage will
be smaller. It is true only when cell area is considered without
sectors. This phenomenon is revealed by figure 7. Figure 7 also
shows, for higher data rates comprehensive coverage area is
found with increasing sectors.
IV. CONCLUSION
In this paper, the performance analyses in coverage and
capacity of UMTS cellular network using sectorization have
been simulated and evaluated for dynamic parameters. The
number of simultaneous users increases or decreases for
increasing or decreasing sectors with dynamic parameters.
Coverage has been estimated for dense urban using COST 231
model where higher data rates need higher processing gain
resulting in smaller coverage area. But increasing sectors with
same parameters makes extensive coverage for higher data
rates.
REFERENCES
[1] Bo Hagerman, Davide Imbeni and Jozsef Barta WCDMA 6 - sector
Deployment-Case Study of a Real Installed UMTS-FDD Network
IEEE Vehicular Technology Conference, spring 2006, page(s): 703 -
707.
[2] S. Sharma, A.G. Spilling and A.R. Nix Adaptive Coverage for UMTS
Macro cellsbased on Situation Awareness. IEEE Vehicular Technology
Conference, spring 2001, page(s):2786 - 2790
[3] A. Wacker, J. Laiho-Steffens, K. Sipila, K. Heiska, "The impact of
the base station sectorisation on WCDMA radio network performance",
IEEE Vehicular Technology Conference ,September 1999,page(s): 2611
- 2615 vol.5.
[4] Romeo Giuliano, Franco Mazzenga, Francesco Vatalaro, Adaptive cell
sectorization for UMTS third generation CDMA systems IEEE
Vehicular Technology Conference, May 2001, page(s): 219 - 223 vol.1.
[5] T.S.Rappaport, Wireless Communications Principles and
Practice-Second Edition, Prentice Hall.
[6] Bernard Sklar, Digital Communications - Fundamentals and
Applications- Second Edition Prentice Hall.
[7] INGO FO, MARC SCHINNENBURG, BIANCA WOUTERS PERFORMANCE
EVALUATION OF SOFT HANDOVER IN A REALISTIC UMTS NETWORK
IEEE VEHICULAR TECHNOLOGY CONFERENCE, SPRING 2003, PAGE(S):
1979 - 1983 VOL.3.
[8] HARRI HOLMA AND ANTTI TOSKALA, WCDMA FOR UMTS-RADIO
ACCESS THIRD GENERATION MOBILE COMMUNICATIONS- JOHN WILEY
& SONS.
[9] Jaana Laiho, Achim Wacker, Tomas Novosad, Radio Network Planning
and Optimisation for UMTS-Second Edition John Wiley & Sons.
[10] A. Skopljak, Multiantenna CDMA systems and their usage in 3G
network, University of Sarajevo, 2007.
[11] Faruque, Saleh, Cellular Mobile System Engineering, Artech
House Publishers, 1996.
[12] Jhong Sam Lee, Leonard E. Miller, CDMA Systems Engineering
Handbook-Artech House Publishers.

Data
rate
(Kbps)
Cell
range
in
(meter)
Cell
Area
without
sector
(meter
2
)
Cell
Area
with 2
sectors
(meter
2
)
Cell
Area
with 3
sectors
(meter
2
)
Cell
Area
with 4
sectors
(meter
2
)
200 773.67 598.57 778.14 1167.2 1556.3
400 635.43 403.77 524.9 78.7.34 1049.8
600 566.31 320.71 416.92 625.38 833.84
800 521.88 272.36 354.07 531.1 708.14
1000 489.83 239.94 311.92 467.88 623.84
1200 465.12 216.33 281.23 421.85 562.47
1400 445.19 198.2 257.66 386.49 515.31
1600 428.63 183.72 238.84 358.26 477.67
1800 414.53 171.83 223.38 335.07 446.76
2000 402.31 161.85 210.41 315.61 420.81
Vol. 2, No. 6, 2011
103 | P a g e
Appendix

Algorithm for Capacity Analysis Using Sectorization:
Begin
Set energy per bit to noise spectral density ratio (Eb/No) = [1 2 3 4 5 6 7 8 9 10 16 20]
Set soft handover gain (H) factor = [0.1 1 1.5 2 3]
Set inter-cell interference () = [.1 1.2 1.55 2]
Set channel activity for data () = [1]
Set channel activity for voice () = [0.1 .3 .38 0.7 0.9]
Set thermal noise () in (20 Kelvin) dbm/Hz= [-173.93]
Set user signal power (S1) in dbm= [21]
Set shadow fading (sh_fd) in db = [8]
Set cell range in Km (R
cell
) = [2]
Set chip rate (W) = [3840000]
Set base band information rate in Kbps (R) = [12.2 64 144 384 2000]
Set base antenna height in meter (h
b
) = [20]
Set user antenna height in meter (h
UE
) = [2]
Set sector (D) = [1 2 3 4 6]
Set frequency range in MHz (f
c
) = [2000]
Set data rate in Kbps (R) = [12.2 64 144 384 2000]
Set array antenna gain (A
g
) in db= [1 2 3.5 5]
//Processing
Processing gain (PG) = 10log (W/R)
Propagation loss in dense urban (Pro_loss) =
3 log ) log 55 . 6 9 . 44 ( 4.97 )] .75 3.2[log(11 log 82 . 13 ) log( 9 . 33 3 . 46
2
UE
+ + + + d h h f
b b c

Signal Power (S) = S1-Pro_loss-sh_hd
//Output
Number of Users
g
o b
s
A H D
S N E
R W
N
+
+ =
o |
n
) 1 (
1
)
/
/
( 1
End

Algorithm for Coverage and Data rates Analysis Using Sectorization:
Begin
Set Transmitter=User Equipment
Set Receiver=Node B
Set mobile max power in dbm (mo_mx) = [21]
Set mobile gain in db (M_G) = [0]
Set cable and connector losses in db (ca_cn_loss) = [3]
Set thermal noise in dbm/Hz () = [-173.93]
Set node B noise figure in db (nodeB_NF) = [5]
Set target load (tar_ld) = [.4]
Set chip rate (W) = [3840000]
Set base antenna height in meter (h
b
) = [20]
Set user antenna height in meter (h
UE
) = [2]
Set energy per bit to noise spectral density ratio (Eb/No) = [5]
Set Power Control Margin or Fading Margin (MPC) = [4]
Set Value for sectors (Sec) = [1 2 3 4]
Set constant value for sectors (K) = [2.6 1.6 1.95 2.6]
Set data rate in Kbps (R) = [100 200 300 400 500 600 2000]
//Processing
Chip rate in (W) = [3840000]
Processing gain (PG) = (W/R)
Effective isotropic radiated power (EIRP) = mo_mx-ca_cn_loss+M_G
Node B noise density (nodeB_ND) = + nodeB_NF
Node B noise power (nodeB_NPW) = nodeB_ND +W_db
Interference margin (IM) =-10log (1-tar_ld)
Vol. 2, No. 6, 2011
104 | P a g e
Node B Interference Power (nodeB_IP) = ( )
10 / 10 / ) arg (
10 10 log 10
Noisepower in eM Interfrenc noisepower
+

Node B Noise and interference (nodeB_NIFPW) = ( )
10 / ) ( 10 / ) (
10 10 log 10
epower Interfrenc noisepower

Node B antenna gain (NodeB_AG) in db = [18]
Receiver Sensitivity (S
rx
) = Eb/No-PG+ nodeB_NIFPW
Total Allowable Path loss=EIRP-S
rx
+ nodeB_AG-MPC=EIRP-(E
b
/N
o
-PG+ nodeB_NIFPW) +
NodeB_AG-MPC
Path loss in dense urban (Durban_Ploss) =

36.37logd 142.17
3 log ) log 55 . 6 9 . 44 ( 4.97 )] .75 3.2[log(11 log 82 . 13 ) log( 9 . 33 3 . 46
2
UE
+ =
+ + + + d h h f
b b c

//Output
Cell radius (d) = 10^ ((1/36.37)* (EIRP-(Eb/No-PG+ nodeB_NIFPW) + nodeB_AG-MPC-142.17))
Cell Area (A) = K*d2
End
Vol. 2, No. 6, 2011
105 | P a g e
Architecture Aware Programming on
Multi-Core Systems

M.R. Pimple
Department of Computer Science & Engg.
Visvesvaraya National Institute of Technology,
Nagpur 440010 (India)
S.R. Sathe
Department of Computer Science & Engg.
Visvesvaraya National Institute of Technology,
Nagpur 440010 (India)

Abstract In order to improve the processor performance, the
response of the industry has been to increase the number of cores
on the die. One salient feature of multi-core architectures is that
they have a varying degree of sharing of caches at different levels.
With the advent of multi-core architectures, we are facing the
problem that is new to parallel computing, namely, the
management of hierarchical caches. Data locality features need
to be considered in order to reduce the variance in the
performance for different data sizes. In this paper, we propose a
programming approach for the algorithms running on shared
memory multi-core systems by using blocking, which is a well-
known optimization technique coupled with parallel
programming paradigm, OpenMP. We have chosen the sizes of
various problems based on the architectural parameters of the
system like cache level, cache size, cache line size. We studied the
cache optimization scheme on commonly used linear algebra
applications matrix multiplication (MM), Gauss-Elimination
(GE) and LU Decomposition (LUD) algorithm.
Keywords- multi-core architecture; parallel programming; cache
miss; blocking; OpenMP; linear algebra.
I. INTRODUCTION
While microprocessor technology has delivered significant
improvements in clock speed over the past decade, it has also
exposed a variety of other performance bottlenecks. To
alleviate these bottlenecks, microprocessor designers have
explored alternate routes to cost effective performance gains.
This has led to use of multiple cores on a die. The design of
contemporary multi-core architecture has progressively
diversified from more conventional architectures. An
important feature of these new architectures is the integration
of large number of simple cores with software managed cache
hierarchy with local storage. Offering these new architectures
as general-purpose computation platforms creates number of
problems, the most obvious one being programmability. Cache
based architectures have been studied thoroughly for years
leading to development of well-known programming
methodologies for these systems, allowing a programmer to
easily optimize code for them. However, multi-core
architectures are relatively new and such general directions for
application development do not exist yet.
Multi-core processors have several levels of memory
hierarchy. An important factor for software developers is how
to achieve the best performance when the data is spread across
local and global storage. Emergence of cache based multi-
core systems has created a cache aware programming
consensus. Algorithms and applications implicitly assume the
existence of a cache. The typical example is linear algebra
algorithms. To achieve good performance, it is essential that
algorithms be designed to maximize data locality so as to best
exploit the hierarchical cache structures. The algorithms must
be transformed to exploit the fact that a cache miss will move
a whole cache-line from main memory. It is also necessary to
design algorithms that minimize I/O traffic to slower
memories and maximize data locality. As the memory
hierarchy gets deeper, it is critical to efficiently manage the
data. A significant challenge in programming these
architectures is to exploit the parallelism available in the
architecture and manage the fast memories to maximize the
performance. In order to avoid the high cost of accessing off-
chip memory, algorithms and scheduling policies must be
designed to make good use of the shared cache[12]. To
improve data access performance, one of the well-known
optimization technique is tiling[3][10]. If this technique is
used along with parallel programming paradigm like OpenMP,
considerable performance improvement is achieved.
However, there is no direct support for cache aware
programming using OpenMP for shared memory environment.
Hence, it is suggested to couple OpenMP with tiling
technique for required performance gain.
The rest of the paper is organized as follows. Section II
describes the computing problem which we have considered.
The work done in the related area is described in section III.
Implementation of the problems is discussed in section IV.
Experimental setup and results are shown in section V. The
performance analysis is carried out in section VI.
II. COMPUTING PROBLEM
As multi-core systems are becoming popular and easily
available choice, for not only high performance computing
world but also as desktop machines, the developers are forced
to tailor the algorithms to take the advantage of this new
platform. As the gap between CPU and memory performance
continues to grow, so does the importance of effective
utilization of the memory hierarchy. This is especially evident
in compute intensive algorithms that use very large data sets,
such as most linear algebra problems. In the context of high
performance computing world, linear algebra algorithms have
to be reformulated or new algorithms have to be developed in
order to take advantage of the new architectural features of
Vol. 2, No. 6, 2011
106 | P a g e
these new processors. Matrix factorization plays an important
role in a large number of applications. In its most general
form, matrix factorization involves expressing a given matrix
as a product of two or more matrices with certain properties.
A large number of matrix factorization techniques have been
proposed and researched in the matrix computation literature
to meet the requirements and needs arising from different
application domains. Some of the factorization techniques are
categorized into separate classes depending on whether the
original matrix is dense or sparse. The most commonly used
matrix factorization techniques are LU, Cholesky, QR and
singular value decomposition (SVD).
The problem of dense matrix multiplication (MM) is a
classical benchmark for demonstrating the effectiveness of
techniques that aim at improving memory utilization. One
approach towards the cache effective algorithm is to
restructure the matrices into sequence of tiles. The copying
operation is then carried out during multiplication. Also, for a
system AX=B, there are several different methods to obtain a
solution. If a unique solution is known to exist, and the
coefficient matrix is full, a direct method such as Gaussian
Elimination(GE) is usually selected.
LU decomposition (LUD) algorithm is used as the
primary means to characterize the performance of high-end
parallel systems and determine its rank in the Top 500 list[11].
LU Factorization or LU decomposition is perhaps the most
primitive and the most popular matrix factorization techniques
finding applications in direct solvers of linear systems such as
Gaussian Elimination. LU factorization involves expressing a
given matrix as product of a lower triangular matrix and an
upper triangular matrix. Once the factorization is
accomplished, simple forward and backward substitution
methods can be applied to solve a linear system. LU
factorization also turns out to be extremely useful when
computing the inverse or determinant of a matrix because
computing the inverse or the determinant of a lower or an
upper triangular matrix is relatively easy.
III. RELATED WORK
Since multi-core architectures are now becoming
mainstream, to effectively tap the potential of these multiple
units is the major challenge. Performance and power
characteristics of scientific algorithms on multi-core
architectures have been thoroughly tested by many
researchers[7]. Basic linear algebra operations on matrices
and vectors serve as building blocks in many algorithms and
software packages. Loop tiling is an effective optimization
technique to boost the memory performance of a program. The
tile size selection using cache organization and data layout,
mainly for single core systems is discussed by Stephanie
Coleman and Kathryn S. Mckinley [10].
LU decomposition algorithm decomposes the matrix that
describes a linear system into a product of a lower and an
upper triangular matrix. Due to its importance into scientific
computing, it is well studied algorithm and many variations to
it have been proposed, both for uni and multi-processor
systems. LU algorithm is implemented using recursive
methods [5], pipelining and hyperplane solutions [6]. It is also
implemented using blocking algorithms on Cyclops 64
architecture [8]. Dimitrios S. Nikolopoulos, in his paper [4]
implemented dynamic blocking algorithm. Multi-core
architectures with alternative memory subsystems are evolving
and it is becoming essential to find out programming and
compiling methods that are effective on these platforms. The
issues like diversity of these platforms, local and shared
storage, movement of data between local and global storage,
how to effectively program these architectures; are discussed
in length by Ioannis E. Venetis and Guang R. Gao [8]. The
algorithm is implemented using block recursive matrix scheme
by Alexander Heinecke and Michael Bader [1]. Jay
Hoeflinger, Prasad Allavilli, Thomas Jackson and Bob Kuhn
have studied scalability issues using OpenMP for CFD
applications[9]. OpenMP issues in the development of
parallel BLAS and LAPACK libraries have also been
studied[2]. However, the issues, challenges related with
programming and effective exploitation of shared memory
multi-core systems with respect to cache parameters have not
been considered.
Multi-core systems have hierarchical cache structure.
Depending upon the architecture, there can be two or three
layers, with private and shared caches. When implementing
the algorithm, on shared memory systems, cache parameters
must be considered. The tile size selection for any particular
thread running on a core is function of size of L
1
cache, which
is private to that core as well as of L
2
cache which is a shared
cache. If cache parameters like, cache level, cache size, cache
line size are considered, then substantial performance
improvement can be obtained. In this paper, we present the
parallelization of MM, GE and LUD algorithm on shared
memory systems using OpenMP.
IV. IMPLEMENTATION
In this paper we have implemented parallelization of most
widely used linear algebra algorithms, matrix multiplication,
gauss elimination and LU decomposition, on multi-core
systems. Parallelization of algorithms can also be implemented
using message passing interface (MPI). Pure MPI model
assumes that, message passing is the correct paradigm to use
for all levels of parallelism available in the application and
that the application topology can be mapped efficiently to
the hardware topology. However, this may not be true in all
cases. For matrix multiplication problem, the data can be
decomposed into domains and these domains can be
independently passed to and processed by various cores.
While, in case of LU decomposition or GE problem, task
dependency prevents to distribute the work load independently
to all other processors. Since the distributed processors do not
share a common memory subsystem, the computing to
communication ratio for this problem is very low.
Communication between the processors on the same node
goes through the MPI software layers, which adds to
overhead. Hence, pure MPI implementation approach is
useful when domain decomposition can be used; such that, the
total data space can be separated into fixed regions of data or
domains, to be worked on separately by each processor.
For GE and LUD problems, we used the approach of 1D
partitioning of the matrix among the cores and then used
OpenMP paradigm for distributing the work among number of
Vol. 2, No. 6, 2011
107 | P a g e
threads to be executed on various cores. The approach of 2 D
partitioning of data among cores is more suitable for array
processors. For a shared memory platform, all the cores on a
single die share the same memory subsystem, and there is no
direct support for binding the threads to the core using
OpeMP. So, we restricted our experiments with 1D
partitioning technique and applied parallelization for achieving
speedup using OpenMP.
A. Architecture Aware Parallelization
To cope up with memory latency, all data required during
any phase of the algorithm are made available in the cache.
The data sets so chosen, are accommodated into the cache.
Considering the cache hierarchy, the tile size selection
depends upon cache size, cache line size to eliminate self
interference misses. Now depending upon the architecture of
the underlying machine, the computation work is split into
number of cores available. One dimensional partitioning of
data is done, so that, every core receives specific number of
rows (or columns), such that, the data fits in the shared cache.
The blocking technique is then used which ensures that the
maximum block size is equal to the size of private cache
belonging to the core. Parallel computation is carried out
using OpenMP pragmas by individual cores.
B. Determining Block Size
In order to exploit cache affinity, the block size is chosen
such that, the data can be accommodated into the cache. The
experiments were carried out on square matrix of size N. Let
s be the size of each element of matrix and C
s
be the size of
shared cache. Let the block size be .
1. For blocked matrix multiplication, C = A x B, block of
matrix A & B, and one row of matrix C should be
accommodated into the cache. Then the required block
size can be calculated using :

For large cache size, we get,

(1)
2. For GE problem, the size of input matrix is [][
]. The required block size can be calculated with the
following equation:

So, the optimal block size,

(2)
3. For LU decomposition algorithm with same matrix used
for storing lower and upper triangular matrix, the optimal
block size comes out to be

(3)

C. Effect of Cache Line Size
Let cache line size be C
ls
Without the loss of generality,
we assume that the first element of input array falls in the first
position of cache. The number of rows that completely fit in
the cache can be calculated as :
Rows = C
s
/C
ls
(4)
For every memory access, the entire cache line is fetched.
So block size

will lead to self interference misses.
Also if
,
system will fetch additional cache lines,
which may in turn lead to capacity misses; as less number of
rows can be accommodated in the cache. So to take the
advantage of spatial locality, the block sizes chosen were
integral multiple of cache line size C
ls.
We assumed that, every
row in the selected tile is aligned on a cache line boundary.
After finding the row size, block size can be calculated.
Block Size B = k x C
ls
if
(k is integer)
Or

Maximum Speed up is achieved when -

or
when B is multiple of number of Rows
The algorithm for block size selection is presented in Fig.
1.
Further improvement in the performance is achieved by
using the technique of register caching for the array elements,
that are outside the purview of the for loop (like value
a[i][j] shown in Fig. 3). This value is cached, which is then
shared by all the threads executing the for loop The
OpenMP implementation of matrix multiplication and GE
problem is given in Fig. 2 and Fig. 3 respectively.
D. LU Decomposition
The main concept is to partition the matrix into smaller
blocks with a fixed size. The diagonal entry in each block is
processed by master thread on a single core. Then for
calculating the entries in the upper triangular matrix, each row
is partitioned into number of groups equal to number of cores;
so that each group is processed by each core. Similarly, for
calculating the entries in the lower triangular matrix, each
column is partitioned into number of groups equal to number
of cores; so that each group is processed by each core. The
implementation divides the matrix into fixed sized blocks, that
fit into the L1 data cache of the core creating first level of
memory locality. On the shared memory architecture, the
whole matrix is assumed to be in the globally accessible
memory address space. The algorithm starts by processing the
diagonal block on one processor, while all other processors
wait for the barrier. When this block finishes, the blocks on
the same row are processed by various cores in parallel. Then
the blocks on same column are processed by various cores in
parallel. In turn, each processor waits for the barrier again for
the next diagonal block.
The storage space can further be reduced by storing lower
and upper triangular matrices in a single matrix. The diagonal
elements of lower triangular matrix are made 1, hence, they
need not be stored. But this method suffers from the problem
of load imbalance, if number of elements processed in each
row or column by each core is not divisible by number of
cores available. Also, the active portion of the matrix is
reduced after each iteration and hence, load allocation after
each iteration is not constant.
Vol. 2, No. 6, 2011
108 | P a g e
void forwardSubstitution() // GaussElimination loop//
// Matrix size (n x n)
{ int i, j, k, max, kk, p, q; float t;
for (i = 0; i < n; ++i)
{ max = i;
for (j = i + 1; j < n; ++j)
if (a[j][i] > a[max][i]) max = j;
for (j = 0; j < n + 1; ++j)
{ t = a[max][j]; a[max][j] = a[i][j];
a[i][j] = t;
}
for (j = n; j >= i; --j)
{ for (k=i+1; k<n; k=k+B)
{ x=a[i][i]; // Register Caching //
#pragma omp parallel for shared(i,j,k) private(kk)
schedule (static)
for (kk=k; kk<(min(k+B, n)); ++kk)
a[kk][j] -= a[kk][i]/x * a[i][j];
}
}
}
1) LUD computation:
Let A be an matrix with rows and columns
numbered from 0 to (n-1). The factorization consists of n
major steps. Each step consisting of an iteration of the outer
loop starting at line 3 of Fig. 5. In step k, first the partial
column [ ] us divided by [ ]. Then the outer
product [ ] x [ ] is subtracted
from the x sub matrix [
]. For each iteration of the outer loop , the
next nested loop in the algorithm goes from .

Figure 1. Block Size Selection

Figure 2. Parallel Matrix Multiplication
A typical computation of LU factorization procedure in the
k
th
iteration of the outer loop is shown in Fig. 4. The k
th

iteration of the outer loop does not involve any computation
on rows 1 to (k-1) or columns 1 to (k-1). Thus at this stage,
only the lower right (n-k) x (n-k) sub matrix of A is
computationally active. So the active part of the matrix
shrinks towards the bottom right corner of the matrix as the
computation proceeds.
The amount of computation increases from top left to
bottom right of the matrix. Thus the amount of work done
differs for different elements of matrix. The work done by the
processes assigned to the beginning rows and columns would
be far less than those assigned to the later rows and columns.
Hence, static scheme of block partitioning can potentially lead
to load imbalance. Secondly, the process working on a block
may idle even when there are unfinished tasks associated with
that block.

Figure 3. OpenMP parallelization of GE loop
This idling can occur if the constraints imposed by the
task-dependency graph do not allow the remaining tasks on
this process to proceed until one or more tasks mapped onto
other processes are completed.
2) LUD OpenMP parallelization:
For parallelization of LU decomposition problem on
shared memory, we used tiling technique with OpenMP
paradigm. The block size B is selected such that, the matrix
size is accommodated in a shared cache. The actual data block
used by each core is less than the size of private cache so that
locality of memory access for each thread is maintained.
For LUD algorithm, due to the task dependency at each
iteration level, the computation cannot be started
simultaneously on every core. So, algorithm starts on one core.
Diagonal element is executed by master core. After the
synchronization barrier, the computation part of non-diagonal
elements is split over the available cores.
After computing a row and column of result matrix, again
the barrier is applied to synchronize the operations for the next
loop. The size of data computed by each core is determined by
block size.
The size of data dealt by each core after each iteration is
not the same. With static scheduling, the chunk is divided
exactly into the available multiple threads and every thread
works on the same amount of data.
Fig. 5 illustrates the OpenMP parallelization. The size of
input matrix a is N

Procedure BS(CS, CLS,N,B)
Input: CS: Cache Size
CLS: Cache Line Size
s: Size of each element in input
N: Input Matrix Rows
Output : B : Block Size(square)
Total cache lines = CS / CLS
No of rows (NR) from input problem size that can be
accommodated in cache
NR = ( )
The optimal block size B
If (NR > CLS)
B= k xCLS // Where k is integer constant
Else (B = CLS)

void mat-mult() // matrix multiplication //
{ for (i=0; i<N; i=i+B)
{ Read block of a & c;
Read block ofbB;
omp_set_num_threads(Omp_get_num_proc());
#pragma omp parallel for shared(a,b,c,i)
private(r,i1,j) schedule (static)
for (r=i; r<(min(i+B, N)); r++)
for (i1=i; i1<(min(i+B,N)); i1++)
{ for(j=0; j<N; j++)
c[r][i1] += a[r][j] * b[j][i1];
}
Write block of c ;
}}

Vol. 2, No. 6, 2011
109 | P a g e

Figure 6. Matrix Multiplication: Data fits in L
2
Cache

Figure 7. Gauss Elimination on 12 & 16 core machines,
Data accommodated in L
2
cache

Figure 8. Speedup for LU Decomposition
on Dual core CPU

Figure 4.Processing of blocks of LU Decomposition

Figure 5. OpenMP implementation of LU Decomposition algorithm
V. EXPERIMENTAL SETUP & RESULTS
We conducted the experiments to test cache aware
parallelization of MM, GE, LUD algorithms on Intel Dual
core, 12 core and 16 core machines. Each processor had hyper
threading technology such that, each processor can execute
simultaneously instructions from two threads, which appear to
the operating system as two different processors and can run a
multi program workload. The configuration of the systems is
given in Table 1.
Each processor had 32 KB data cache as L
1
cache. Intel
Xeon processors (12 & 16 cores) had an eight way set
associative 256 KB L
2
cache and 12 MB L
3
cache dynamically
shared between the threads. The systems run Linux 2.6.x
Blocked LU decomposition was parallelized at two levels
using OpenMP.
We used relatively large data sets, so that the performance
of the codes becomes more bound to the L
2
and L
3
cache miss
latencies. The programs were compiled with C compiler (gcc
4.3.2). Fig. 6 and Fig. 7 show the speed up achieved when the
block sizes are such that, the data fits in L
2
cache for matrix
multiplication and Gauss elimination algorithm respectively.
Fig. 8 and Fig. 9 show the results of LU decomposition
algorithm for various matrix sizes on dual and 16 core system
respectively

Table 1. System configuration

Processors
Intel(R)
Core2
Duo
CPU
E7500
Intel(R)
Dual
Core
CPU
E5300
Intel(R)
Xeon(R)
CPU
X5650
(12
cores)
Intel(R)
Xeon(R)
CPU
E5630
(16
cores)
Core
frequency
2.93 GHz 2.60 GHz 2.67 GHz 2.53 GHz
L1 Cache
size
32 KB I
cache,
32 KB D
cache
32 KB I
cache,
32 KB D
cache
32 KB I
cache,
32 KB D
cache
32 KB I
cache,
32 KB D
cache
L2 Cache
size
3072 KB,
shared
2048 KB,
Shared
256 KB 256 KB
L3 Cache
size
--- --- 12 MB 12 MB

1. Lu-Fact (a)
2. {
3. for (k=0; k<N; k++)
4. { #pragma omp single
5. for(j=k+1; j<(N); j++)
6. a[j][k]=a[j][k]/a[k][k];
7. #pragma omp parallel for shared (a,k)
private(i) schedule static
8. for(j=k+1; j<(N); j=j+B)
9. for (jj=j; jj<min(jj+B, N), jj++)
10. { v=a[k][jj]; --- caching the value
11. #pragma omp parallel for shared
(a,k,jj) private(i) schedule static
12. for(i=k+1; i<(N); i++)
13. a[i][jj]= a[i][jj]- (a[i][k]*v);
14. }
15. }
16. }

Vol. 2, No. 6, 2011
110 | P a g e
Figure 9. Speedup for LU Decomposition on 16 core
machine
(Matrix size- NxN)

VI. PERFORMANCE ANALYSIS
The strategy of parallelization is based on two
observations. One is that the ratio computation to
communication should be very high for implementation on
shared memory multi-core systems. And second is that the
memory hierarchy is an important parameter to be taken into
account for the algorithm design which affects load and store
times of the data. Considering this, we implemented the
algorithms matrix multiplication and Gauss elimination with a
blocking scheme that divides the matrices into relatively small
square tiles. The optimal block size is selected for each core,
such that the tile is accommodated in the private cache of each
core and thus avoids the conflict misses. This approach of
distributing the data chunks to each core greatly improves the
performance. Fig. 6 and Fig. 7 shows the performance
improvement when block size is multiple of cache line size.
Whenever block size is greater or less than the cache line size,
performance suffers. This is due to reloading overheads of
entire new cache line for the next data chunk. With this
strategy, we got the speedup of 2.1 on 12 core machine and
speed up of 2.4 on 16 core machine. The sub linear
speedups in Fig. 6 and 7 for lower block sizes are attributed to
blocking overheads.
For Gauss elimination and LU decomposition problem, the
OpenMP pragma, splits the data among the available cores.
The size of data dealt by every core, after every iteration is
different. This leads to load imbalance problem. The chunk
scheduling scheme, demands the chunk calculations at every
iteration and hence affects performance. However, static
scheduling ensures equal load to every thread and hence
reduces the load imbalance. For LU decomposition problem
with 1D partitioning of data among the cores, we observed a
speedup of 1.39, & 2.46 for two dual core machines and
speedup of 3.63 on 16 core machine. The maximum speedup
is observed when the number of threads is equal to the number
of (hardware) threads supported by the architecture. Fig. 9
shows the speed up when 16 threads are running on a 16 core
machine. Speed up is directly proportional to the number of
threads. The performance degrades when more software
threads are in execution than the threads supported by
architecture. So, for 18 threads, scheduling overhead increases
and performance is degraded. However, when number of
threads is more than 8, performance degrades due to
communication overheads. This is because, 16 core Intel
Xeon machine comprises of 2 quad cores connected via QPI
link. Fig. 9 shows performance enhancement up to eight
threads and degradation in the performance when number of
threads is ten. When the computation is split across all the
available sixteen threads, speed up is again observed, where
communication overhead is amortized over all cores. Further
enhancement in the performance is achieved when method of
register caching is used for loop independent variables in the
program. Many tiling implementations do not consider this
optimal block size considerations with cache attributes.
However, our implementation considers the hierarchy of
caches, cache parameters and arrives at optimal block size.
The block size calculations are governed by the architecture of
the individual machine and the algorithm under consideration.
Once the machine parameters and input problem size is
available, the tailoring of the algorithm accordingly improves
the performance to a greater extent. Of course, there is a
significant amount of overhead in the OpenMP barriers at the
end of loops; which means that load imbalance and not data
locality is the problem.
VII. CONCLUSION & FUTURE WORK
We evaluated performance effects of exploiting
architectural parameters of the underlying platform for
programming on shared memory multi-core systems. We
studied the effect of private cache L
1
, shared cache L
2
, cache
line size on execution of compute intensive algorithms. The
effect of exploiting L1 cache affinity does not affect the
performance much, but the effects of exploiting L2 cache
affinity is considerable, due its sharing among multiple threads
and high reloading cost for larger volumes. If these factors are
considered and coupled with parallel programming paradigm
like OpenMP, performance enhancement is achieved. We
conclude that, affinity awareness in compute intensive
algorithms on multi-core systems is absolutely essential and
will improve the performance significantly. We plan to extend
the optimization techniques for the performance enhancement
on multi-core systems by considering the blocking technique
at register level and instruction level. We also plan to
investigate and present generic guide lines for compute
intensive algorithms on various multi-core architectures.
REFERENCES
[1] Alexander Heinecke and Michael Bader, Towards many-core
implementation of LU decomposition using Peano Curves, UCHPC-
MAW09, May 18-20, 2009, Ischia, Itali.
[2] C. Addison, Y. Ren, and M. van Waveren, OpenMP issues arising in
the development of Parallel BLAS and LAPACK libraries, In
Scientific Programming Journal, Volume 11, November 2003, pages:
95-104,IOS Press.
[3] Chung-Hsing Hsu, and Ulrich Kremer, A quantitative analysis of tile
size selection algorithms, The Journal of Supercomputing, 27, 279-
294, 2004.
[4] Dimitrios S. Nikolopoulos, Dynamic tiling for effective use of shared
caches on multi-threaded processors, International Journal of High
Vol. 2, No. 6, 2011
111 | P a g e
Performance Computing and Networking, 2004, Vol. 2, No. 1, pp. 22-
35.
[5] F.G. Gustavson, Recursion leads to automatic variable blocking for
dense linear algebra algorithms, IBM Journal of Research and
Development, 41(6):737-753, Nov.,1997.
[6] H. Jin, M. Frumkin, and J. Yan, The OpenMP implementation of NAS
parallel benchmarks and its performance, Technical report nas-99-011,
NASA Ames Research Centre, 1999.
[7] Ingyu Lee, Analyzing performance and power of multi-core
architecture using multithreaded iterative solver, Journal of Computer
Science, 6 (4): 406-412, 2010.
[8] Ioannis E. Venetis, and Guang R. Gao, Mapping the LU decomposition
on a many-core architecture: Chanllenges and solutions, CF09, May
18-20, 2009, Ischia, Itali.
[9] Jay Hoeflinger, Prasad Alavilli, Thomas Jackson, and Bob Kuhn,
Producing scalable performance with OpenMP: Experiments with two
CFD applications, International Journal of Parallel computing.
27(2001), 391-413.
[10] Stephanie Coleman, and Kathryn S. McKinley, Tile size selection using
cache organization and data layout, Proceedings of the ACM SIGPLAN
Conference on Programming Language Design & Implementation,
California, United States, 1995, pages: 279-290
[11] The Top 500 List. http://www.top500.org
[12] Vahid Kazempour, Alexandra Fedorova, and Pouya Alagheband,
Performance implications of cache affinity on multicore processors,
Proceedings of the 14
th
International Euro-Par Conference on Parallel
Processing, Spain, 2008, pages:151-161

AUTHORS PROFILE
S.R. Sathe received M.Tech. in Computer Science from IIT, Bombay (India)
and received Ph.D. from Nagpur University(India). He is currently
working as a Professor in the Department of Computer Science &
Engineering at Visvesvaraya National Institute of Technology,
Nagpur(India). His research interests include paralle and distributed
systems, mobile computing and algorithms.

M.R. Pimple is working as a Programmer in the Department of Computer
Science & Engineering at Visvesvaraya National Institute of
Technology, Nagpur(India) and persuing her M.Tech. in Computer
Science. Her area of interests are computer architecture, parallel
processing.

Vol. 2, No. 6, 2011
112 | P a g e
Implementation of ISS - IHAS (Information Security
System Information Hiding in Audio Signal) model
with reference to proposed e-cipher Method

Prof. R. Venkateswaran
Department of CA, Nehru College of Management,
Research Scholar-Ph.D, Karpagam University
Coimbatore, TN, INDIA
Dr. V. Sundaram, Director
Department of CA, Karpagam College of Engineering
Affiliated to Anna University of Technology,
Coimbatore, TN, INDIA

Abstract This paper shows the possibility of exploiting the
features of E- cipher method by using both cryptography and
Information hiding in Audio signal methods used to send and
receive the message in more secured way. Proposed methodology
shows that successfully using these Poly substitutions methods
(Proposed E-Cipher) for Encode and decodes messages to evolve
a new method for Encrypting and decrypting the messages.
Embedding secret messages using audio signal in digital format is
now the area in focus. There exist numerous steganography
techniques for hiding information in audio medium.
In our proposed theme, a new model ISS-IHAS - Embedding
Text in Audio Signal that embeds the text like the existing system
but with strong encryption that gains the full advantages of
cryptography. Using steganography it is possible to conceal the
full existence of the original text and the results obtained from
the proposed model is compared with other existing techniques
and proved to be efficient for textual messages of minimum size
as the size of the embedded text is essentially same as that of
encrypted text size. This emphasis the fact that we are able to
ensure secrecy without an additional cost of extra space
consumed for the text to be communicated.
Keywords- Encryption; Decryption; Audio data hiding; Mono
Substitution; Poly Substitution.
I. OBJECTIVES OF THE PROJECT
The main purpose of Audio steganography is to hide a
message in some cover media, to obtain new data, practically
indistinguishable from the original message, by people, in such
a way that an eavesdropper cannot detect the presence of
original message in new data. With computers and Networks,
there are many other ways of hiding information, such as
Covert channels, Hidden text within WebPages Hiding files in
Plain sight, Null ciphers.
Today, the internet is filled with tons of programs that use
steganography to hide the secret information. There are so
many medias are used for digitally embedding message such as
plaintext, hypertext, audio/video, still image and network
traffic. There exists a large variety of steganographic
techniques with varying complexity and possessing some
strong and weak aspects.
..
Hiding information in text is the most popular method of
Steganography. It is used to hide a secret message in every nth
character or altering the amount of white space after lines or
between words of a text message [1]. It is used in initial decade
of internet era. But it is not used frequently because the text
files have a small amount of redundant data. But this technique
lacks in payload capacity and robustness. To hide data in audio
files, the secret message is embedded into digitized audio
signal. Audio data hiding method provides the most effective
way to protect privacy. Key aspect of embedding text in audio
files is that, no extra bytes are generated for embedding. Hence
it is more comfortable to transmit huge amount of data using
audio signal. Embedding the secret messages in digital sound is
usually a very difficult process [2].
II. PROPOSED ISS IHAS MODEL
The following IHAS Model provides a very basic
description of the audio steganographic process in the sender
side and receiver side.

Sender Side

Fig 2.1 System Flow ISS- IHAS model
Sender Side Plain Text
E-Cipher
Encryptor

E-Cipher
Decryptor
Plain Text Receiver Side
Encryption
Key
Decryption
Key
Cover Signal
Stegno Signal
Cipher
Text
Cipher
Text
Vol. 2, No. 6, 2011
113 | P a g e
The original text encrypted by E-Cipher using an
encryption key. The model implements E-Cipher encryption as
it proves to be more efficient. The encrypted text is passed on
to Entrencher that embeds the encrypted text inside the cover
signal which is in audio format *.wav resulting in stego signal.
This process happens at sender side. This stego signal is
communicated using Network medium. At the receiver side the
stego signal is passed on to Extorter module that extracts
embedded text from the audio signal that was used a s cover
medium,. The resultant cipher text is then decrypted using E-
cipher Decryptor module. The final plain text can then be used
for further processing.

1001 1000 0011 1100
1101 1011 0011 1000
1000 1000 0001 1111
1101 1100 0111 1000

0011 1100 1001 1000
0011 1000 1101 1011
0001 1111 1000 1000
0111 1000 1101 1100

1001 1000 0011 1101
1101 1011 0011 1000
1000 1000 0001 1111
1101 1100 0111 1001

0011 1100 1001 1001
0011 1001 1101 1010
0001 1110 1000 1000
0111 1001 1101 1100

Fig. 2.2 ISS- IHAS encoding format
To hide a letter A & B to an digitized audio file where each
sample is represented with 16 bits then the LSB bit of sample
audio file is replaced with each bit of binary equivalent of the
letter A & B[4].
III. PERSPECTIVE STUDY ON VARIOUS METHODS
In audio steganography, secret message is embedded into
digitized audio signal which result slight altering of binary
sequence of the corresponding audio file. There are several
methods are available for audio steganography [3]. Some of
them are as follows: -
LSB Coding:
Least significant bit (LSB) coding is the simplest way to
embed information in a digital audio file. By substituting the
least significant bit of each sampling point with a binary
message, LSB coding allows for a large amount of data to be
encoded. The following diagram illustrates how the message
'HEY' is encoded in a 16-bit CD quality sample using the LSB
method:

Fig.3.1. Message 'HEY' is encoded in a 16-bit CD quality sample using the
LSB method

In LSB coding, the ideal data transmission rate is 1 kbps
per 1 kHz. In some implementations of LSB coding, however,
the two least significant bits of a sample are replaced with two
message bits. This increases the amount of data that can be
encoded but also increases the amount of resulting noise in the
audio file as well. Thus, one should consider the signal content
before deciding on the LSB operation to use. For example, a
sound file that was recorded in a bustling subway station would
mask low-bit encoding noise. On the other hand, the same
noise would be audible in a sound file containing a piano solo
[9].
To extract a secret message from an LSB encoded sound
file, the receiver needs access to the sequence of sample indices
used in the embedding process. Normally, the length of the
secret message to be encoded is smaller than the total number
of samples in a sound file. One must decide then on how to
choose the subset of samples that will contain the secret
message and communicate that decision to the receiver. One
trivial technique is to start at the beginning of the sound file
and perform LSB coding until the message has been
completely embedded, leaving the remaining samples
unchanged. This creates a security problem, however in that the
first part of the sound file will have different statistical
properties than the second part of the sound file that was not
modified. One solution to this problem is to pad the secret
message with random bits so that the length of the message is
equal to the total number of samples. Yet now the embedding
process ends up changing far more samples than the
transmission of the secret required. This increases the
probability that a would-be attacker will suspect secret
communication [8].
Parity Coding
Instead of breaking a signal down into individual samples,
the parity coding method breaks a signal down into separate
Audio Sample File

Encryption
01100101 011 00010
Vol. 2, No. 6, 2011
114 | P a g e
regions of samples and encodes each bit from the secret
message in a sample region's parity bit. If the parity bit of a
selected region does not match the secret bit to be encoded, the
process flips the LSB of one of the samples in the region. Thus,
the sender has more of a choice in encoding the secret bit, and
the signal can be changed in a more unobtrusive fashion.
Using the parity coding method, the first three bits of the
message 'HEY' are encoded in the following figure. Even parity
is desired. The decoding process extracts the secret message by
calculating and lining up the parity bits of the regions used in
the encoding process. Once again, the sender and receiver can
use a shared secret key as a seed in a pseudorandom number
generator to produce the same set of sample regions.

Fig.3.2. First three bits of the message 'HEY' are encoded using Parity coding
method

There are two main disadvantages associated with the use
of methods like LSB coding or parity coding. The human ear is
very sensitive and can often detect even the slightest bit of
noise introduced into a sound file, although the parity coding
method does come much closer to making the introduced noise
inaudible. Both methods share a second disadvantage however,
in that they are not robust. If a sound file embedded with a
secret message using either LSB coding or parity coding was
resampled, the embedded information would be lost.
Robustness can be improved somewhat by using a redundancy
technique while encoding the secret message. However,
redundancy techniques reduce data transmission rate
significantly.
Phase Coding
Phase coding addresses the disadvantages of the noise-
inducing methods of audio steganography. Phase coding relies
on the fact that the phase components of sound are not as
perceptible to the human ear as noise is. Rather than
introducing perturbations, the technique encodes the message
bits as phase shifts in the phase spectrum of a digital signal,
achieving an inaudible encoding in terms of signal-to-perceived
noise ratio.

Fig.3.3. Phase Coding

To extract the secret message from the sound file, the
receiver must know the segment length. The receiver can then
use the DFT to get the phases and extract the information.
One disadvantage associated with phase coding is a low
data transmission rate due to the fact that the secret message is
encoded in the first signal segment only. This might be
addressed by increasing the length of the signal segment.
However, this would change phase relations between each
frequency component of the segment more drastically, making
the encoding easier to detect. As a result, the phase coding
method is used when only a small amount of data, such as a
watermark, needs to be concealed.
In a normal communication channel, it is often desirable to
concentrate the information in as narrow a region of the
frequency spectrum as possible in order to conserve available
bandwidth and to reduce power. The basic spread spectrum
technique, on the other hand, is designed to encode a stream of
information by spreading the encoded data across as much of
the frequency spectrum as possible. This allows the signal
reception, even if there is interference on some frequencies.
While there are many variations on spread spectrum
communication, we concentrated on Direct Sequence Spread
Spectrum encoding (DSSS). The DSSS method spreads the
signal by multiplying it by a chip, a maximal length
pseudorandom sequence modulated at a known rate. Since the
host signals are in discrete-time format, we can use the
sampling rate as the chip rate for coding. The result is that the
most difficult problem in DSSS receiving, that of establishing
the correct start and end of the chip quanta for phase locking
purposes, is taken care of by the discrete nature of the signal.
Spread Spectrum

In the context of audio steganography, the basic spread
spectrum (SS) method attempts to spread secret information
across the audio signal's frequency spectrum as much as
possible. This is analogous to a system using an
implementation of the LSB coding that randomly spreads the
message bits over the entire sound file. However, unlike LSB
coding, the SS method spreads the secret message over the
sound file's frequency spectrum, using a code that is
independent of the actual signal. As a result, the final signal
occupies a bandwidth in excess of what is actually required for
transmission [6].
Two versions of SS can be used in audio steganography:
the direct-sequence and frequency-hopping schemes. In direct-
sequence SS, the secret message is spread out by a constant
called the chip rate and then modulated with a pseudorandom
Vol. 2, No. 6, 2011
115 | P a g e
signal. It is then interleaved with the cover-signal. In frequency
-hopping SS, the audio file's frequency spectrum is altered so
that it hops rapidly between frequencies.
Echo Hiding:
In echo hiding, information is embedded in a sound file by
introducing an echo into the discrete signal. Like the spread
spectrum method, it too provides advantages in that it allows
for a high data transmission rate and provides superior
robustness when compared to the noise inducing methods. To
hide the data successfully, three parameters of the echo are
varied:
Amplitude, decay rate, and offset (delay time) from the
original signal. All three parameters are set below the human
hearing threshold so the echo is not easily resolved. In addition,
offset is varied to represent the binary message to be encoded.
One offset value represents a binary one, and a second offset
value represents a binary zero.

Fig.3.4. Echo Hiding
IV. METHODOLOGY
Proposed E-Cipher E & D Algorithm [5]

i. Take three key e1, e2, e3 and assign a character e1 be a
and e2 be D and e3 bes.

ii. Let ASCII value of e1 be 1 and e2 be 2 and e3 be 3 and
take the text , add ASCII value of e1 to value of first
character, and e2 to second character and e3 to third
character, alternatively add the value of e1 , e2, e3 to
consecutive characters.

iii. Three layers to be applied to each three consecutive letters
and same to be continued thru the remaining text.

iv. After adding ASCII value of all values of given text, the
resultant text is an encrypted message, and it generate a
combination of 3* (256 * 256 * 256) letters encrypted
coded text with 128 bit manner.

v. Transposition takes place in each character after all the
process is over that is moves or change one bit either LSB
or MSB, the end result is increasing security.

vi. Reverse process of the above algorithm gives the actual
plain text without any error.

V. METHODOLOGIES
ISS- IHAS SENDER ALGORITHM
Input: Audio file, Key and Original message Output: Mixed
Data.
Algorithm
Step 1: Load the audio file (AF) of size 12 K.
Step 2: Input key for encryption
Step 3: Convert the audio files in the form of bytes and this
byte values are represented in to bit patterns.
Step 4: Using the key, the original message is encrypted using
E-Cipher algorithm.
Step 5: Split the audio file bit patterns horizontally into two
halves.
Step 6: Split the Encrypted message bit patterns vertically into
two halves.
Step 7: Insert the LSB bit of the vertically splitted encrypted
text file (TF) into the LSB bit of the horizontally
splitted audio file.
Step 8: Repeat Step 7 for the remaining bits of encrypted text
file.
Step 9: If size (AF) size (TF) then
embedding can be done as explained above
else
The next higher order bit prior to previous bit position
can be used
Until it is exhausted.

ISS- IHAS ALGORITHM - AT THE RECEIVER SIDE:
Input: Mixed data, Key Output: Original message, audio file.

Algorithm
Step 1: Load the Stegno signal
Step 2: Extract the hidden data and audio files bit patterns
from mixed data [9]
// Reverse process of step 7 of ISS-IHAS algorithm at sender
side.

Step 3: Input key for decryption (as used in encryption)
Step 4: Combine the two halves of audio files bit patterns.
Step 5: Combine the two halves of encrypted messages bit
pattern.
Step 6: Using Key, decrypt the original message.

Vol. 2, No. 6, 2011
116 | P a g e
VI. EVALUATION AND ANALYSIS REPORT

Table 5.1 shows the different levels of satisfaction Level.
VII. CONCLUSION
In this paper we have introduced a robust method of
imperceptible audio data hiding. This system is to provide a
good, efficient method for hiding the data from hackers and
sent to the destination in a safe manner. This proposed system
will not change the size of the file even after encoding and also
suitable for any type of audio file format. Thus we conclude
that audio data hiding techniques can be used for a number of
purposes other than covert communication or deniable data
storage, information tracing and finger printing, tamper
detection
This proposed system provides an efficient method for
hiding the data from the eavesdropper. LSB data hiding
technique is the simplest method for inserting data into audio
signals. ISS- IHAS model is able to ensure secrecy with less
complexity at the cost of same memory space as that of
encrypted text and the user is able to enjoy the benefits of
cryptography and steganography [7] combined together
without any additional overhead. This work is more suitable for
automatic control of robotic systems used in military and
defense applications that can listen to a radio signal and then
act accordingly as per the instructions received. By embedding
the secret password in the audio signal the robot can be
activated only if the predefined password matches with the
incoming password that reaches the robot through audio signal.
It can then start functioning as per the instructions received in
the form of audio signal. More such sort of applications can be
explored but confined to audio medium usage.
REFERENCES

[1] Bethany Delman,'Genetic Algorithms in Cryptography' published in
web; july 2004.

[2] Darrell Whitley,'A Genetic Algorthm Tutorial', Computer Science
Department, Colorado State University, Fort Collins, CO 80523
[3] Nalani N, G. Raghavendra Rao,' Cryptanalysis of Simplified Data
Encryption Standard via Optimisation Heuristics;IJCSNS, Vol.6 No.1B,
January 2006.
[4] Sean Simmons,'Algebric Cryptoanalysis of Simplified AES', October
2009;33,4;Proquest Science Journals Pg.305.
[5] Sujith Ravi, Kevin Knight,'Attacking Letter Substitution Ciphers with
Integer Programming',Oct 2009,33,4; Proquest Science Journals Pg.321.
[6] Verma, Mauyank Dave and R.C Joshi,'Genetic Algorithm and Tabu
Search Attack on the Mono Alphabetic Substitution Cipher in Adhoc
Networks; Journal of COmputer Science 3(3): 134-137, 2007.
[7] William Stallings,"Cryptography and Network Security: Principles
and Practice",2/3e Prentice hall , 2008.
[8] Ingemar J. Cox, Ton Kalker, Georg Pakura and Mathias Scheel.
Information Transmission and Steganography, Springer, Vol.
3710/2005, pp. 15-29.
[9] K.Geetha , P.V. Vanthia muthu , International journal of Computer
Science and Engineering vol 2 No.4 Pg No: 1308-1312 , Year 2010
AUTHORS PROFILE
R. Venkateswaran received his professional degree MCA and MBA (IS)
from Bharathiar University, Tamilnadu, India, He received his M.Phil from
Bharathidasan University, Tamilnadu, India, and He is currently a Ph.D
Scholar in the Karpagam Academy of Higher Education, Karapagam
University, Tamilnadu, India, in the field of Cryptography and Network
Security. Presently he is working as an Asst. Professor of Computer
Applications, Nehru College of Management, Coimbatore, Tamilnadu. He is
the member of CSI and other IT forums. He had published two International
Journals and Presented many papers in national and International conferences,
seminars and workshops. His research interests are in Cryptography and
network security, information security, Software engineering,

Dr. V. Sundaram received his professional degree M.Sc. in Mathematics
from the University of Madras in the year 1967 and he received his
Professional Doctoral Degree Ph. D in Mathematics from the University of
Madras in 1989. And He is currently working as Director, Department of
Computer Applications in Karpagam College of Engineering, Tamilnadu,
India, He is a research Guide for Anna university, Bharathiar university as
well as Karpagam University in the field of Computer applications. He
published several papers in International Journals and Conferences and also
published 13 books in the area of engineering mathematics and he is the life
member of ISTE and ISIAM. His research interests are in Cryptography and
network security, Applied Mathematics, Discrete Mathematics, Network etc.
Plain Text Image Audio Video
Invisibility Medium High High High
Payload
Low Low High High

Capacity

Robustness
against
Low Medium High High

Statistical

Attacks

Robustness
against Text Low Medium High High
Manipulation

Variation in
Medium Medium High Medium

file size

Vol. 2, No. 6, 2011
117 | P a g e
Image Compression using Approximate Matching and
Run Length
Samir Kumar Bandyopadhyay, Tuhin Utsab Paul, Avishek Raychoudhury
Department of Computer Science and Engineering,
University of Calcutta
Kolkata-700009, India

Abstract Image compression is currently a prominent topic for
both military and commercial researchers. Due to rapid growth
of digital media and the subsequent need for reduced storage and
to transmit the image in an effective manner Image compression
is needed. Image compression attempts to reduce the number of
bits required to digitally represent an image while maintaining its
perceived visual quality. This study concentrates on the lossless
compression of image using approximate matching technique and
run length encoding. The performance of this method is
compared with the available jpeg compression technique over a
wide number of images, showing good agreements.
Keywords- lossless image compression; approximate matching; run
length.
I. INTRODUCTION
Images may be worth a thousand words, but they generally
occupy much more space in a hard disk, or bandwidth in a
transmission system, than their proverbial counterpart. So, in
the broad field of signal processing, a very high-activity area is
the research for efficient signal representations. Efficiency, in
this context, generally means to have a representation from
which we can recover some approximation of the original
signal, but which doesnt occupy a lot of space. Unfortunately,
these are contradictory requirements; in order to have better
pictures, we usually need more bits.
The signals which we want to store or transmit are normally
physical things like sounds or images, which are really
continuous functions of time or space. Of course, in order to
use digital computers to work on them, we must digitize those
signals. This is normally accomplished by sampling (measuring
its instantaneous value from time to time) and finely quantizing
the signal (assigning a discrete value to the measurement) [1].
This procedure will produce long series of numbers. For all
purposes of this article, from here on we will proceed as if
these sequences were the original signals which need to be
stored or transmitted, and the ones we will eventually want to
recover. After all, we can consider that from this digitized
representation we can recover the true (physical) signal, as long
as human eyes or ears are concerned. This is what happens, for
example, when we play an audio CD. In our case, we will focus
mainly on image representations, so the corresponding example
would be the display of a picture in a computer monitor.
However, the discussion in this paper, and especially the theory
developed here, apply equally well to a more general class of
signals.
There are many applications requiring image compression,
such as multimedia, internet, satellite imaging, remote sensing,
and preservation of art work, etc. Decades of research in this
area has produced a number of image compression algorithms.
Most of the effort expended over the past decades on image
compression has been directed towards the application and
analysis of different coding techniques to compress the image
data. Here in this paper also, we have proposed a two step
encoding technique that transform the image data to a stream of
integer values. The number of values generated by this
encoding technique is much less than the original image data.
The main philosophy of this encoding technique is based on the
intrinsic property of most images, that similar patterns are
present in close locality of images.
The coding technique makes use of this philosophy and
uses an approximate matching technique along with the
concept of run length to encode the image data into a stream of
integer data. Experimental results over a large number of
images have shown good amount of compression of image
size.
II. RELATED WORKS
Image compression may be lossy or lossless. Lossless
compression is preferred for archival purposes and often for
medical imaging, technical drawings, clip art, or comics. This
is because lossy compression methods, especially when used at
low bit rates, introduce compression artifacts. Lossy methods
are especially suitable for natural images such as photographs
in applications where minor (sometimes imperceptible) loss of
fidelity is acceptable to achieve a substantial reduction in bit
rate. The lossy compression that produces imperceptible
differences may be called visually lossless.
Methods for lossless image compression are:
Run-length encoding used as default method in PCX
and as one of possible in BMP, TGA, TIFF
DPCM and Predictive Coding
Entropy encoding
Adaptive dictionary algorithms such as LZW used in
GIF and TIFF
Deflation used in PNG, MNG, and TIFF
Chain codes

Vol. 2, No. 6, 2011
118 | P a g e
Run-length encoding (RLE) is a very simple form of data
compression in which runs of data (that is, sequences in which
the same data value occurs in many consecutive data elements)
are stored as a single data value and count, rather than as the
original run. This is most useful on data that contains many
such runs: for example, simple graphic images such as icons,
line drawings, and animations. It is not useful with files that
don't have many runs as it could greatly increase the file size.
DPCM or differential pulse-code modulation is a signal
encoder that uses the baseline of PCM but adds some
functionalities based on the prediction of the samples of the
signal. The input can be an analog signal or a digital signal.
Entropy encoding is a lossless data compression scheme
that is independent of the specific characteristics of the
medium.
One of the main types of entropy coding creates and assigns
a unique prefix-free code to each unique symbol that occurs in
the input. These entropy encoders then compress data by
replacing each fixed-length input symbol by the corresponding
variable-length prefix-free output codeword. The length of each
codeword is approximately proportional to the negative
logarithm of the probability. Therefore, the most common
symbols use the shortest codes.
LempelZivWelch (LZW) is a universal lossless data
compression algorithm created by Abraham Lempel, Jacob Ziv,
and Terry Welch. It was published by Welch in 1984 as an
improved implementation of the LZ78 algorithm published by
Lempel and Ziv in 1978. The algorithm is simple to implement,
and has the potential for very high throughput in hardware
implementations.
Deflate is a lossless data compression algorithm that uses a
combination of the LZ77 algorithm and Huffman coding. It
was originally defined by Phil Katz for version 2 of his PKZIP
archiving tool, and was later specified in RFC 1951.
A chain code is a lossless compression algorithm for
monochrome images. The basic principle of chain codes is to
separately encode each connected component, or "blot", in the
image. For each such region, a point on the boundary is
selected and its coordinates are transmitted. The encoder then
moves along the boundary of the image and, at each step,
transmits a symbol representing the direction of this movement.
This continues until the encoder returns to the starting position,
at which point the blot has been completely described, and
encoding continues with the next blot in the image.
III. OUR WORK
The main philosophy behind selecting approximate
matching technique along with run length encoding technique
is based on the intrinsic property of most images, that they have
similar patterns in a localized area of image, more specifically
the adjacent pixels row differ in very less number of pixels.
This property of image is exploited to design a very effective
image compression technique. Testing on a wide variety of
images has provided satisfactory results. The technique used in
this compression methodology is described in this section.
We consider approximate matching algorithm and run
length for our image compression. The approximate matching
algorithm does a comparison between two strings of equal
length and represent the second string with respect to the first
only with the information of the literal position where the string
mismatches.
Replace. This operation is expressed as (p; char) which
means replacing the character at position p by character char.
Let C denote copy", and R denote replace then the
following are two ways to convert the string
11010001011101010" to 11010001111001010" (0,1 are
stored in ASCII)via different edit operation sequences:
C C C C C C C C R C C R C C C C C
1 1 0 1 0 0 0 1 0 1 1 1 0 1 0 1 0
1 1 0 1 0 0 0 1 1 1 1 0 0 1 0 1 0
A list of edit operations that transform a string u to another
string v is called an EditTranscription of the two strings [9].
This will be represented by an edit operation sequence (u; v)
that orderly lists the edit operations. For example, the edit
operation sequence of the edit transcription in the above
example is (\11010001011101010",\11010001111001010") = (
9; 1),(12,0);
Approximate matching method. In this case, the string
\11010001011101010" can be encoded as f(17; 2)= ( 9;
1),(12,0), storing the ASCII characters require 136 bit or 17
byte where as storing 4 characters will require 4 byte. Thus a
compression of approximate 76.4% is achieved. This technique
is very useful in image compression because of the inherent
property of an image because two consecutive rows of an
image has almost same string of pixel values. Only a few pixel
varies. Experimental results prove this hypothesis.
Apart from the concept of approximate matching method,
the concept of run length is also used because using run length
a row of image can be represented using much less literals than
the original.
Run-length Encoding, or RLE is a technique used to reduce
the size of a repeating string of characters. This repeating string
is called a run, typically RLE encodes a run of symbols into
two bytes , a count and a symbol. RLE can compress any type
of data regardless of its information content, but the content of
data to be compressed affects the compression ratio. Consider a
character run of 15 'A' characters which normally would
require 15 bytes to store :
AAAAAAAAAAAAAAA is stored as 15A
With RLE, this would only require two bytes to store, the
count (15) is stored as the first byte and the symbol (A) as the
second byte.
In this compression technique, we have used the
approximate matching method in unison with run length.
Starting from the left uppermost row of image, every three
rows are considered at a time. Of these, the middle row is
represented using run length, and the row above and below it
are matched with the middle row using approximate matching
method. This method is continued iteratively until the whole
image is scanned and compressed.
Vol. 2, No. 6, 2011
119 | P a g e
The algorithms designed as per our technique are as
follows:
A. COMPRESS (Source Raw Image file)
This is the main algorithm for compression. This algorithm
will be used to compress the data part of the Source Image File.
Output: It will output the Compressed-Image file.
Input: This function will take Source Image file as input.
1. Read the Source Image file as input. Obtain its size
(say r*c).Store the data part(pixel values) of the
image in an array A of the same size.
2. Quantize the color palate of the image, i.e array A
with quantization factor 17.
3. If r is not divisible by 3, then duplicate the last row 1
or 2 times at the bottom of A such that the number of
rows become divisible by 3. Reset r with the
corresponding new size of A.
4. Take a blank array say Compress of size n*2(n is a
positive integer).Starting with the 1st row, choose
consecutive 3 rows at a time and perform the
following operations(say, we have chosen row
number k-1,k and k+1) in each iterations:
a. For each column in the array A, if any mismatch
is found in the row k-1 and k, the corresponding
column number and the value at that
corresponding column with row number k-1 in
array A, is stored in array Compress. For every
mismatch, those two values are stored in a single
row of array Compress.
b. For row number k, the corresponding value
(starting from the 1st column) and its runlength
(Number of consecutive pixels with same pixel
value) for kth row is stored in array Compress.
Every set of value and its runlength is stored in a
single row of array Compress.
c. For each column in the array A, if any mismatch
is found in the row k and k+1, the corresponding
column number and the value at that
corresponding column with row number k+1 is
stored in array Compress. For every mismatch,
those two values are stored in a single row of
array Compress.
5. Repeat Step 4 until all the rows are compressed. A
marker should be used to distinguish between the
encrypted versions of each row. Store also the value
of r and c in Compress.
6. Array Compress now constitutes the compressed data
part of the corresponding Source Image File.
B. DECOMPRESS (Compressed-Image file)
This is the main algorithm for decompression or decoding
the image. This algorithm will be used to decompress the data
part of the Source Image File i.e. the image from the
Compress array.
Output: It will output the Decompressed or Decoded Image
file.
Input: This function will take the Compressed-Image file
(Compress array) as input.
1. Read the Compress array. Obtain the size of the image
(say r*c) from the array. Take a blank array say Rec
of the same size for reconstruction of the data part of
the image.
2. Starting from the 1st row, consider the compressed
values of consecutive 3 rows from Temp and perform
the following operations(say, we have chosen row
number k-1,k and k+1) in each iterations:
a. Firstly, construct the kth row of Rec array with
the corresponding positional value and runlength
value in the Compress array, by putting the same
positional value in runlength number of
consecutive places in the same row.
b. Then, construct the (k-1)th row. In the
corresponding Compress array for this particular
row for each column if an entry for column
number v is not present, then Rec[(k-
1),v]=Rec[k,v]. Else, if Compress[i,1]=v then,
Rec[(k-1),v]= Compress[i,2].
c. Then, construct the (k+1)th row. In the
corresponding Compress array for this particular
row for each column if an entry for column
number v is not present, then
Rec[(k+1),v]=Rec[k,v]. Else, if Compress [i,1]=v
then, Rec[(k+1),v]= Compress [i,2].
3. Step 2 is repeated until the full Rec array is filled.
4. Rec array is stored as the Decompressed Image File.
IV. RESULT AND DISCUSSION
A. Complexity analysis of the stated algorithm
Let the size of the image be r*c. Then, at the time of
Compression, 3 rows are considered at a time and for each
compression of rows c number of columns are read. At this
process 3 rows are compressed at a time taking 3*c number of
comparisons. So, for compression of the whole image, total
number of compression required is

*3*c=r*c ,that is
O(r*c).So, for an image of size n*n, the time complexity of the
compression algorithm is O(n2).
In the receiver end, the Compress array is read and the Rec
array is reconstructed which also takes number of
comparisons=r*c ,that is O(r*c).So, for an image of size nxn,
the time complexity of the de-compression algorithm is O(n2).
B. Test Results
Before Compression (For each image) :
Size : 300 x 280 = 84000 [Row x Col]
Size in bytes : 1,68000 byte = 168kb
After Compression (For each image) :
Vol. 2, No. 6, 2011
120 | P a g e

Figure 1. Nature
Size in bytes : 51348 byte = 102.69 kb
Compression percentage : 38.87 %

Figure 2. Library

Figure 3. Landscape

Figure 4. Crowd

Figure 5. Tom_Jerry

Figure 6. Thumbnail

Figure 7. Model_face
C. Conclusion
The algorithm proposed here is for lossless image
compression ass it is evident from the algorithm, that the exact
image data (pixel values) are extracted from the compressed
data stream without any loss. This is possible because the
compression algorithm does not ignores or discards any
Vol. 2, No. 6, 2011
121 | P a g e
original pixel value. Moreover the techniques such as
approximate matching and run length encoding technique are
intrinsically lossless.
This compression technique proves to be highly effective
for images with large similar locality of pixel lay out. This
technique will find extensive use in medical imaging sector
because of its lossless characteristics and the medical images
has large area of similar pixel layout pattern, like in X ray
images large area are black.
REFERENCES
[1] P. S. R. Diniz, E. A. B. da Silva, and S. L. Netto, Digital Signal
Processing: System Analysis and Design. Cambridge University Press,
2002.
[2] J. M. Shapiro, Embedded Image Coding Using Zero-Trees of Wavelet
Coefficients, IEEE Transactions on Signal Processing, vol. 41, pp.
34453462, December 1993.
[3] A. Said and W. A. Pearlman, A New, Fast and Efficient Image Codec
Based on Set Partitioning in Hierarchical Trees, IEEE Transactions on
Circuits and Systems for Video Technology, vol. 6, no. 3, pp. 243 250,
June 1996.
[4] D. Taubman, High Performance Scalable Image Compression with
EBCOT, IEEE Transactions on Image Processing, vol. 9, no. 7, July
2000.
[5] Tse-Hua Lan and A. H. Tewfik, Multigrid Embedding (MGE) Image
Coding, Proceedings of the 1999 International Conference on Image
Processing, Kobe.
[6] Civarella and Moffat. Lossless image compression using pixel
reordering. Proceedings of twenty seventh Australian Computer Science
conference,pp 125-132,2004.
[7] K.Veeraswamy, S.Srinivaskumar and B.N.Chatterji. Lossless image
compression using topological pixel re-ordering. IET international
conference, India, pp 218-222,2006.
[8] Memon and Shende. An analysis of some scanning techniques for
lossless image coding. IEEE Transactions on Image Processing,9 (11),pp
1837-1848,2000.
[9] Memon and Wu X. Recent developments in context based predictive
techniques for lossless image compression. The computer Journal,40,pp
127-136,1997.
[10] K.Sayood. Introduction of data compression. Acdemic press, 2nd
edition,2000.
[11] D.Salomon. Data Compression. Springer,2nd edition, 2000.

AUTHORS PROFILE
Dr. Samir K. Bandyopadhyay, B.E., M.Tech., Ph. D (Computer Science &
Engineering), C.Engg., D.Engg., FIE, FIETE, currently,
Professor of Computer Science & Engineering,
University of Calcutta, visiting Faculty Dept. of Comp.
Sc., Southern Illinois University, USA, MIT, California
Institute of Technology, etc. His research interests
include Bio-medical Engg, Image processing, Pattern
Recognition, Graph Theory, Software Engg.,etc. He has
25 Years of experience at the Postgraduate and under-graduate Teaching &
Research experience in the University of Calcutta. He has already got several
Academic Distinctions in Degree level/Recognition/Awards from various
prestigious Institutes and Organizations. He has published more than 300
Research papers in International & Indian Journals and 5 leading text books
for Computer Science and Engineering. He has visited round the globe for
academic and research purposes.

Tuhin Utsab Paul received his Bachelors degree in Computer science in 2008
and Masters degree in Computer and Information Science
in 2010, both from the University of Calcutta. He is
currently doing his M.Tech course in Computer science
and engineering from the University of Calcutta. His
research interest include Image processing, pattern
recognisation, image steganography and image
cryptography. He has published quite a few Research
papers in International & Indian Journals and
conferences. He is an IEEE member since 2009.

Avishek Raychoudhury received his Bachelors degree
in Computer science in 2008 and Masters degree in
Computer and Information Science in 2010, both from
the University of Calcutta. He is currently doing his
M.Tech course in Computer science and engineering
from the University of Calcutta. His research interest
include Image processing, image steganography, graph
theory and its applications. He has published quite a few
Research papers in International & Indian Journals and conferences.

Vol. 2, No. 6, 2011
122 | P a g e
Interactive Intranet Portal for effective Management
in Tertiary Institution

Idogho O. Philipa
Department of Office Technology
and Management
Auchi Polytechnic, Auchi
Edo State, Nigeria
Akpado Kenneth
Department of Electronics and
Computer Engineering
Nnamdi Azikiwe University
Anambra State Nigeria
James Agajo
Electrical/ Electronics Dept.
Auchi Polytechnic, Auchi
Edo State, Nigeria

Abstract Interactive Intranet Portal for effective management
in Tertiary Institution is an enhanced and interactive method of
managing and processing key issues in Tertiary Institution,
Problems of result processing, tuition fee payment, library
resources management are analyzed in this work. An interface
was generated to handle this problem; the software is an
interactive one. Several modules are involved in the paper, like:
LIBRARY CONSOLE, ADMIN, STAFF, COURSE
REGISTRATION, CHECKING OF RESULTS and E-NEWS
modules. The server computer shall run the portal as well as
OPEN SOURCE Apache Web Server, MySQL Community
Edition RDBMS and PHP engine and shall be accessible by client
computers on the intranet via thin-client browser such as
Microsoft Internet Explorer or Mozilla Firefox. It shall be
accessible through a well-secured authentication system. This
Project will be developed using OPEN SOURCE technologies
such XAMMP developed from WAMP (Windows Apache
MySQL and PHP)
Keywords- Portal; Database; webA; MYSQL; Intranet; Admin.
I. INTRODUCTION
Interactive intranet portal for effective management in
tertiary institution seeks to address the Problems arising from
result processing, tuition fee payment, library resources
management are analyzed in this work. An interface was
generated to handle this problem, the software is an interactive
one. An intranet is a network inside an organization that uses
internet technologies (such as web browsers and servers,
TCP/IP networks protocols, HTML hyper media document
publishing and databases, etc) to provide an intranet-like
environment with the organization for information sharing,
communications, collaboration, and the support of business
processes. An intranet is protected by security measures such as
passwords, encryption, and firewalls, and thus can be accessed
by authorized users through the intranet. Secure intranets are
now the fastest-growing segment of the internet because they
are less expensive to build and manage than private networks
based on proprietary protocols.
Internets appeared in the mid-1990s and were perceived as
the answer to the need for the integration of existing
information systems into organizations. Despite the fact that
has been extensive research regarding implementation,
development processes, policies, standardization vs. creativity
and so forth, the potentiality of intranets has not been fully
exploited. Intranets offer many advantages in the form of
working networks that support and enables empowered
employees to participate in the development of the
organization, to enable the measurement of essential functions
and to monitor industries conditions and find suitable functions
that support doing work.[1]
II. BACKGROUND
On our Institution like many other universities there is an
intense need for communication and co-operation between the
administrational staff and department. This is because most of
the department resources like student course registration,
Result management, staff management and student
management have to be managed partly by one or the other
group, different resources for different reasons.
III. OBJECTIVES
The main objective of this project is to develop on intranet
portal software that will adequately manage records
A. Benefits
1. Enable the library to keep inventory of all books
2. Keep track of borrowed books
3. Save information of all student id cards
4. Show student detail and books borrowed by these
students
5. Provide easy collation of students information for
clearance
6. Connect with oxford university library for additional
materials
7. Web based access for e-books, papers, journals and
students projects which promote Auchi Polytectnic
worldwide via the internet.
8. Access journals and papers from Oxford University
press.
9. A computerized easy and fast clearance process in the
library
B. Justification
Need for efficient, effective and adequate
management of students records[2]
Vol. 2, No. 6, 2011
123 | P a g e

Figure 1 BLOCK DIAGRAM OF PORTAL

Need for adequate protection and security of vital
information
Providing academics the ability to manage and
communicate more effectively with students data.
Helping academics spend less time on processing
students data
Need for easy and past means of information
dissemination within the department
IV. METHODOLOGY
Being a web-based portal, first the web development is
determined using an open sources platform that will be more
flexible and have lower cost due to free licensing.
Dreamweaver: This is a unique tool expressly designed for
the development and optimization of web pages. All the coding
in this project is done in the code view of the Dreamweaver.
Hypertext preprocessor (PHP): PHP is the web
development language written by and for web developers .PHP
is a server-side scripting language, which can be embedded in
the HTML. In this project PHP is used to write all the forms
script codes which made the software very interactive and more
dynamic.
MYSQL: This is an open source, SQL Relational Database
Management System (RDBMS) that is free for many uses.
MYSQL is used for database management in this project.
V. CORPORATE PORTAL DEFINITIONS
Figure 1 show a block diagram describing various stages in
the portal, a portal was referred to as search engine, whose
main goal was to facilitate access to information contained in
documents spread throughout internet. Initially, search engines
enabled internet users to locate documents with the use of
Boolean operators or associative links between web pages. To
reduce even more the searching time and to help inexperienced
users, some search engines have included categories, that is,
they started to filter sites and sports, metrology, tourism,
finance, news, culture etc. The succeeding steps were the
integration of other functions, such as virtual communities and
real time chats; the ability to personalize search engine
interfaces (my yahoo, my Excite, etc); and to access specialized
and commercial contents. This new concept of search engine is
now called a portal.
VI. MAJOR CHARACTERISTICS OF A CORPORATE PORTAL
Since corporate portals integrate some well-know
technologies, such as intelligence business tools, document
management, office automation, groupware, data warehouse,
and intranet, some suppliers of products on these areas have
also positioned themselves as corporate portal vendors. At the
same time, small companies have viewed the great market
opportunity of corporate portals and have announced new
portal products. Besides, some big computers companies have
established technical and/or commercial alliance to provide
joint solutions and to suit specific needs of their customers.
Therefore, the solution of a particular corporate portal, amongst
all products available on the market today, is not an easy task.
These are some of the characteristics of corporate portal.
The ability to manage the information life cycle,
establishing storage hierarchical levels and discarding
necessary

Admin
module
Staff module
Students Course registration
module
Result
module

E- News

Admin
login

Staff Log in

Student
login

New student
Signup

Check
Result

Create
Account
Database
Vol. 2, No. 6, 2011
124 | P a g e
The ability to locate experts in the organization, in
accordance with the type of knowledge demanded for a
particular task
The ability to satisfy the information needs of all types
of corporate users.
The possibility of information exchange among
customers, employees, suppliers and resellers, providing an
information infrastructure suitable for electronic commerce.
Conclusively, corporate portals, whose ancestors are the
decision support and the management information systems, are
the next step into modern design of user interfaces to corporate
information. Adapting the enterprise environment to suit users
needs and optimize the interaction, distribution and
management of internal and external information resources
the corporate portal allow users to access corporate information
in an easier and customized way, resulting, the theoretically, in
reduced costs, increased, productivity and competitiveness.
VII. SYSTEM DESIGN APPROACH
A. Top-Down Design
In top-down design, the system is designed in such a way
that all necessary steps needed for the realization of the design
will be followed promptly starting from the known to the
unknown. The first stage is the approval of the project
topic/title followed by the necessary information/data gathering
required for the design stage. Next is the designing stage that is
the main part of the work followed by the implementation of
the design work where all the required components are being
applied. The testing stage where the implemented works are
tested in order to determine the output of the system.[3]
B. Benefits
To save time and reduce the stress involved in
students result processing;
To facilitate students result processing activities
with the aid of a program to give rise to effective
production of students results with fewer human
errors;
To handle large volumes of results processing;
To aid the management in the providing results
data in time.
To design a computerized result processing
system that will minimize the amount of manual
input needed in student result processing.
Such a system would help the polytechnic system in
becoming more computer-friendly and technologically-inclined
as staff and students could make use of it and benefit more
from its numerous objectives and advantages.
Figure 2 and figure 3 shows Flow chart of student course
registration Module and Flow chart of Admin module. The
flow diagram represents the stages and sequence this world
pass through.
VIII. STRUCTURAL ANALYSIS
The first point to mention is the users of the system.
Different user can use the portal. The system will be analyzed
in the following [4]
Admin Module: The admin module has interface that
allow the administrator to login into the system and
have over roll control of the system. The Admin has
to create login ID and password
Figure 2. Flow chart of student course registration Module

to the authorized staffs. The admin decides on certain
privileges like update, delete, and edit to only authorized
persons. The logic flow in Admin module is represented
below.
Start
Click On course
Registration
Login
Existing account
holder login
Reg No password
Is
Reg No & and
Password
Valid

Submit

Course registration

Enter the
Courses
Any
Outstanding
courses
New User
Sign up
Create Accounts

Total credit unit
must not exceed
24
Yes
No
Yes
No
End
Database
Vol. 2, No. 6, 2011
125 | P a g e
IX. STUDENT COURSE REGISTRATION MODULE
This module has interface that allow students to register
their courses. A new student who has not registered at all has to
go through student account signup and obtain user name and
password before being able to enroll to the program and select
courses. In the signup step, the student is asked to fill his
profile information such as his first name, last name, middle
name, registration number, and gender, date of birth, email
address, permanent address, phone number and educational
level. The student has to login with the user ID and password
obtained from signup account. The logic flow in the student
course registration is represented below. [6]
X. RESULT MODULE
This module has interface that allow students to login into

Figure 3 Flow chart of Admin module

the system, using their registration number and password. New
users have to signup an account that will enable them to log in
and access their result. The system uses the registration number
to query the database in order to retrieve the result pointing to
the registration number. The logic flow in the result module is
represented below.
XI. DATABASE DESIGN
The database is the long-term memory of the web database
application. Relational database management system
(RDBMS) was considered during the design of this database.
The database structures have tables and their relationships are
defined. The databases have eight (8) tables and they are
related with primary and foreign keys..[7]

Figure 4 Flow chart of Result Modules module

1) Course table
2) Course level table
3) Exam table
4) Student table
5) Staff table
6) Semester table
7) Session table
8) Year table
9) Result table
XII. SYSTEM IMPLEMENTATION
The system implementation was done using open source
software XAMMP developed from WAMP (Window Apache,
MYSQL and PHP) and integrating it with macromedia
Dreamweaver installed on computer. The following and done
for the implementation.[8],[9]

Start
Click on
check result
Existing account
in order login
Reg no and
password
Is
Login ID and
Password
Valid

Display results

End

NO

YES
New User
Signup
Start
Click On Admin
Login
Enter Login ID and
Password
Login ID and
Password
Is
Login ID and
Password
valid

Perform Operation

End

Data base

No
Vol. 2, No. 6, 2011
126 | P a g e
Figure 5 Proposed interface for Auchi library Management
Figure 6 (SCREEN SHOT) APPLICANT STATISTICS 1
Figure 7(a) (SCREEN SHOT) APPLICANT STATISTICS 2
Figure. 7(b) (SCREEN SHOT) APPLICANT STATISTICS 3
Figure. 8 Staff module login interface
Figure 9 Student course Registration interface

Vol. 2, No. 6, 2011
127 | P a g e
Figure 10 Student course Registration interface
1) The modules image was created using macromedia
flash 8.0
2) The admin module was created by following the logic
flow diagram using dream weaver 8.0
3) The staff module was also created as in figure 8
4) Student course registration / check result modules was
also created
5) E-news module was created and linked to Admin
module
Admin login was created and the login ID and password is
also created and store in the database
Finally common database was designed for all the modules
XIII. SUB MODULES IMPLEMENTATION
A. Admin Module
The following are done on implementation Admin Area
was created and it has the following features
1) Customize semester courses
2) Enter student Exam scores
3) Enter student Record
4) Check student result
5) View students record
6) Logout
7) Back to Index page
8) Assign login details to staff
B. Staff Module
The following are done on Implementation
1) Obtain login ID and password from admin
2) Staff Area was created and it has the following features
, see figure 8

Figure 11 Student Sign up Block interface [13,14]

a) Customize semester courses
b) Enter student Exam scores
c) Enter student Record
d) Check student result
e) View students record
f) Logout
g) Back to Index page
XIV. AUCHI POLYTECTNIC WEB BASED ADMISSION
PROCESSING CENTER
The system will help speed up admission processing within
24hr. The proposed system will automatically categorize every
applicant within their various choices and arrange them
according to their performance (Jamb Score) in an ascending
order. More Explanation can be provided if requested. [8]
A. Student course registration
The following are done on implementation
1) Student course registration as shown in figure. 10 have
two sub modules was created which account is stored in the
signup database table.
a) Existing Account Holder
2) Existing account holder login was also created
3) New student signup here was created and the data
account is stored in the signup database table.
4) Existing account holder login was also created.
XV. DATABASE IMPLEMENTATION
Implementation of the database is carried out on MYSQL
5.0 DB sever engine. The data retrieval is analyzed in the
implementation phase. The basic operation of a web server is in
fig 13 and 14.
A communication link is required between them. A web
browser makes a request of the server. The servers send back a
response. The web database implementation in this project
follow a general web database structure as shown.
The basic web database architecture consists of the web
browser, webserver, scripting engine and database server
The following stages are involved in data retrieval from the
database
1) A users web browser issues an HTTP request for a
particular web page
2) The web server receives the request for the page,
retrieves the files, and passes it to the PHP engine for
processing
3) The PHP engine begins parsing the script. Inside the
script is command to connect to the database and execute a
query. PHP opens a connection to the MYSQL server and
sends on the appropriate query.
4) The MYSQL server receives the database query,
processes it and sends the results back to the PHP engine the
database as shown below.

Browser

Request

Web server
Vol. 2, No. 6, 2011
128 | P a g e
Figure 12 Student Enhanced Sign up Block interface [15,16]
A. Database entity relationship model
This is the relationship model in MYSQL database as
shown in the figure 15
a) Creating Database table
All the database tables in my design was created in
MYSQL by writing the SQL requires save in the root of my
web server called SQL.Main. Proper SQL command was used
to insert it into
XVI. SYSTEM TESTING AND EVALUATION
A. Test Plan
In testing this work, one should adopt, first the bottom-up
approach to test the various modules before finally testing the
complete system with the control program as was done in this
work.
Furthermore, ones main concerned was the fact that the
system should meets its functional and requirements.
B. Testing the Admin Module Interface
The module was really tested with the xampp server user
Interface in figure 16. The admin was tested with
the valid login ID and password that allow the admin into
database and the admin area was display on the interface.
C. Testing the Staff Module Interface
This module was really tested with the xampp server user
interface. The login ID and password obtained from admin was
used to test the staff module and it was successful
D. Testing the Student Course Registration Module Interface
interface. The module was tested with the login ID and
password obtained by student signup account. It interacted with
the database
E. Testing the Check Result Module Interface
interface.
The module was also tested with the login and password
obtained from student check result signup account. The testing
was successful.
F. Software Testing and Debugging
The program development was carried out in modules
using dreamweaver8.0 connected to xampp server containing
Apache, MYSQL and PHP. The first task was to write all the
PHP codes in Dreamweaver and save it in root of my server.
Next was to check if the Apache server is running which acts as
the web server and if it connected to MYSQL database.
Finally, all the modules were linked into the database to
check if they are all running.
Figure 13 Student Sign up interface
Figure. 14 :Client/Server relationship between and server
Figure. 15 Database entity relationship model interface
XVII. ACTUAL VS EXPECTED RESULTS
During random testing of all the modules some errors came
up and I discovered that it was not inserting data into the
database.

Browser

Web server

PHP Engine

MYSQL server
1
6
2
5
3
4
Vol. 2, No. 6, 2011
129 | P a g e
TABLE 1. ACTUAL RESULT VS. EXPECTED RESULT

ACTUAL RESULT EXPECTED RESULT
It does not insert data into the
database.
Expected to insert data into the
database.
Could not execute SQL queries. It is supposed to execute SQL queries.
Figure. 16 Database tables interface
XVIII. PERFORMANCE EVALUATION
The performance of the Results did not meet the actual
design of this project. It is expected that the project can be
continued upon after the defense to meet the required design of
this project.
The proliferation of digital information resources and
electronic databases challenges libraries and demands that
libraries develop new mechanisms to facilitate and better
inform user selection of electronic databases and search
tools.[11,12]
XIX. CONCLUSION
The fundamental idea of intranet portal-based designs is to
reduce the amount time spent in data processing. Companies
and organizations with intranet portal have attracted much
stock market investors interest because portals are viewed as
able to command a large audience.
Finally, intranet portal bring vast information and services
resources available from many sources to many users within
the same organization in an effective manner.
REFERENCES
[1] Horton Jr., F., Infotrends, Prentice Hall 2
nd
edition, pp. 185-191, 1996.
[2] Oliveira, D., So Paulo: Atlas, IEICE Trans.Commun., pp 123-144, 1996.
[3] Cronin, B., & Davenport, E., Elements of Information management.
New Jersey: Scarecrow Press, 1991.
[4] Taylor, A., & Farrell, S., Information management in context. Aslib
proceedings, 44(9), pp. 319-322, 1992.
[5] Butcher, D., & Rowley, J., The 7 Rs of Information management
Managing Information, 5(3), pp. 34-36, 1998.
[6] Shilakes, C. C., & Tylman, J., Enterprise Information portals. New
York: Merril Lynch, (November 16). [online], October 1999.
[7] Collins, D., Data warehouses, enterprise in information portal, and the
SmartMart meta directory, Information Builders Systems Journal,
12(2), pp. 53-61, 1999.
[8] Murray, G., The Portal in the desktop. Intraspect,
(May/June)[http://archives.grouping
computing.com//index.cfm?fuseaction=viewarticle&contentID=166].
October 1999.
[9] Viador, Enterprise Information portals: Realizing the vision of
ourfingertips.January)[http://www.viador.com/pdfs/EIP_white_paper
_1_99 .pdf]. April 2000.
[10] White, C., Enterprise information portal requirements. Morgan
Hill,CA:DatabaseaAssociatesinternational,(January)[http://www.decisio
n processing.com/papers/eip2.doc]. April 2000.
[11] Wei Ma, Journal of the American Society for Information Science and
Technology archive Volume 53 Issue 7, John Wiley & Sons, Inc. New
York, NY, USA July 2002
[12] Dutta S.B.Wierenga and A. Dalebout, Designing Management support
systems using an integrative perspective Communication of ACM Vol.
40 No 6.1997 june.
[13] Mintzberg H., et al, The strategic Portal process, 4
th
. Edition upper
saddle River, NJ: Prentice Hall (2002).
[14] Mora M. , Management and organizational issues(information Resources
management journal special issue October/December 2002.
[15] McNurlin B.C and R.H. Information system Management , 7
th
. Edition
upper saddle River Prentice Hall.(2005 February)
[16] Joch A.(2005,January/February)Eye on Information Oracle Magazine
(2005,January/February)
ABOUT THE AUTHORS
Dr. (Mrs.)Philipa Omamhe Idogho, a scholar,
received the Ph.D in Educational Administration from
Ambrose Alli University Ekpoma, and a Master Degree
in educational Management from University of Benin.
With over 28 years of teaching/research experience, she
is currently the Rector of a first generation Federal
Polytechnic in Nigeria (Auchi Polytechnic, Auchi). She
has more than 50 papers in National/International Conferences/Journals to her
credit. She is a fellow, Nigeria Institute of Management, Institute of
Administrative Management of Nigeria and member, Association of Business
Educators of Nigeria, Association for Encouraging Qualitative Education in
Nigeria, and other recognized professional organizations. Outside her
academic life, she runs a Non-governmental Organisation (NGO) known as
Women Enhancement Organization. It is a charitable, non-profit making, non-
partisan and non-religious organization which works in three thematic areas of
gender works, HIV/AIDS and literacy education for rural women and
vulnerable children.
Engr. James Agajo is into a Ph.D Programme in the
field of Electronic and Computer Engineering, He has a
Masters Degree in Electronic and Telecommunication
Engineering from Nnamdi Azikiwe University Awka
Anambra State, and also possesses a Bachelor degree in
Electronics and Computer Engineering from the Federal
University of Technology Minna Nigeria. His interest is
in intelligent system development with a high flare for Engineering and
Scientific research. He has Designed and implemented the most resent
computer controlled robotic arm with a working grip mechanism 2006 which
was aired on a national television , he has carried out work on using blue tooth
technology to communicate with microcontroller. Has also worked on thumb
print technology to develop high tech security systems with many more He is
presently on secondment with UNESCO TVE as a supervisor and a resource
person. James is presently a member of the following association with the
Nigeria Society of Engineers(NSE), International Association of
Engineers(IAENG) UK, REAGON, MIRDA,MIJICTE.
Engr. Kenneth Akpado is a Ph.D Research student, he holds an M. Eng,
B. Eng. in Electronics and Computer Engineering, a Member of IAENG,MNSE
and COREN, he is presently a Lecturer in the Department of Electrical
Electronics Engineering in Nnamdi Azikiwe University Awka Anambra State,
Nigeria.
Vol. 2, No. 6, 2011
130 | P a g e
Iris Recognition Using Modified Fuzzy Hypersphere
Neural Network with different Distance Measures

S. S. Chowhan
Dept. of Computer Science
COCSIT
Latur, India
U. V. Kulkarni
SGGS IET
Nanded, India
G. N. Shinde
Dept. of Elec. & Computer Science
Indira Gandhi College
Nanded, India

AbstractIn this paper we describe Iris recognition using
Modified Fuzzy Hypersphere Neural Network (MFHSNN) with
its learning algorithm, which is an extension of Fuzzy
Hypersphere Neural Network (FHSNN) proposed by Kulkarni et
al. We have evaluated performance of MFHSNN classifier using
different distance measures. It is observed that Bhattacharyya
distance is superior in terms of training and recall time as
compared to Euclidean and Manhattan distance measures. The
feasibility of the MFHSNN has been successfully appraised on
CASIA database with 756 images and found superior in terms of
generalization and training time with equivalent recall time.
Keywords- Bhattacharyya distance; Iris Segmentation; Fuzzy
Hypersphere Neural Network.
I. INTRODUCTION
Iris recognition has become the dynamic theme for security
applications, with an emphasis on personal identification based
on biometrics. Other biometric features include face,
fingerprint, palm-prints, iris, retina, gait, hand geometry etc.
All these biometric features are used in security applications
[1]. The human iris, the annular part between pupil and sclera,
has distinctly unique features such as freckles, furrows, stripes,
coronas and so on. It is visible from outside. Personal
authentication based on iris can obtain high accuracy due to
rich texture of iris patterns. Many researchers work has also
affirmed that the iris is essentially stable over a persons life.
Since the iris based personal identification systems can be more
noninvasive for the users [2].
Iris boundaries can be supposed as two non-concentric
circles. We must determine the inner and outer boundaries with
their relevant radius and centers. Iris segmentation is to locate
the legitimate part of the iris. Iris is often partially occluded by
eyelashes, eyelids and shadows. In segmentation, it is desired
to discriminate the iris texture from the rest of the image. An
iris is normally segmented by detecting its inner areas (pupil)
and outer (limbus) boundaries [3] [4]. Well-known methods
such as the Integro-differential, Hough transform and active
contour models have been successful techniques in detecting
the boundaries. In 1993, Daugman proposed an integro-
differential operator to find both the iris inner and outer borders
Wildes represented the iris texture with a laplacian pyramid
constructed with four different resolution levels and used the
normalized correlation to determine whether the input image
and the model image are from the same class [5]. O. Byeon and
T. Kim decomposed an iris image into four levels using
2DHaar wavelet transform and quantized the fourth-level high
frequency information to form an 87-bit code. A modified
competitive learning neural network (LVQ) was used for
classification [6]. J. Daugman used multiscale quadrature
wavelets to extract texture phase structure information of the
iris to generate a 2048-bit iris code and he compared the
difference between a pair of iris representations by computing
their Hamming distance [7]. Tisse. used a combination of the
integro-differential operators with a Hough Transform for
localization and for feature extraction the concept of
instantaneous phase or emergent frequency is used. Iris code is
generated by thresholding both the models of emergent
frequency and the real and imaginary parts of the instantaneous
phase[8]. The comparison between iris signatures is performed,
producing a numeric dissimilarity value. If this value is higher
than a threshold, the system generates output as a nonmatch,
meaning that each iris patterns belongs to different irises [9].
In this paper, we have applied MFHSNN classifier which is a
Modification of Fuzzy Hypersphere Neural Network (FHSNN)
proposed by Kulkarni et al. [10]. Ruggero Donida Labati, et al.
had represented the detection of the iris center and boundaries
by using neural networks. The proposed algorithm starts by an
initial random point in the input image, and then it processes a
set of local image properties in a circular region of interest
searching for the peculiar transition patterns of the iris
boundaries. A trained neural network processes the parameters
associated to the extracted boundaries and it estimates the
offsets in the vertical and horizontal axis with respect to the
estimated center [12].
II. TOPOLOGY OF MODIFIED FUZZY HYPERSPHERE
NEURAL NETWORK
The MFHSNN consists of four layers as shown in Fig 1(a).
The first, second, third and fourth layer is denoted as ,
R
F ,
M
F
,
N
F and
o
F respectively. The
R
F layer accepts an input pattern
and consists of n processing elements, one for each dimension
of the pattern. The
M
F layer consists of q processing nodes that
are constructed during training and each node represents
hypersphere fuzzy set characterized by hypersphere
membership function [11]. The processing performed by each
node of
M
F layer is shown in Fig 1(b). The weight between
R
F
and
M
F layers represents centre points of the hyperspheres.
Vol. 2, No. 6, 2011
131 | P a g e
Figure 1. (a) Modified Fuzzy Hypersphere Neural Network

Figure1. (b) Implementation Modified Fuzzy Hypersphere Neural Network
As shown in above Fig 1(b),
1 2
( , ,....... )
j j j jn
C c c c =
represents center point of the hypersphere
j
m . In addition to
this each hypersphere takes one more input denoted as
threshold, , T which is set to one and the weight assigned to this
link is
j
. .
j
.

represents radius of the hypersphere,
j
m which
is updated during training. The center points and radii of the
hyperspheres are stored in matrix C and vector , . respectively.
The maximum size of hypersphere is bounded by a user
defined value , where 0 l s s

is called as growth parameter
that is used for controlling maximum size of the hypersphere
and it puts maximum limit on the radius of the hypersphere.
Assuming the training set defined as { | 1, 2,......, },
k
R R h P e =
where
1 2
( , ,...... )
h h h hn
R r r r = e
n
I is the th h pattern, the
membership function of hypersphere node
j
m is
) , , ( 1 ) (
, ,
. .
j j j h j
l f C R m = (1)

Where ( ) f is three-parameter ramp threshold function
defined as ( , , ) 0
j
f l . = if 0 ,
j
l . s s else ( , , ) ,
j
f l l . = if
,
j
l l . < s < l 1, and argument l is defined as:

=
=
n
i
hi ji
r c l
1
(2)
The membership function returns 1,
j
m = if the input
pattern
k
R is contained by the
k
R hypersphere. The parameter
, 0 , l s s is a sensitivity parameter, which governs how fast
the membership value decreases outside the hypersphere when
the distance between
h
R and
j
C increases The sample plot of
membership function with centre point [0.5 0.5] and radius
equal to 0.3 is shown in as shown in Fig 2.
Figure 2. Plot of Modified Fuzzy Hypersphere membership function for =1
Each node of
N
F and
o
F

layer represents a class. The
N
F
layer gives fuzzy decision and output of th k
N
F node
represents the degree to which the input pattern belongs to the
class
k
n . The weights assigned to the connections between
M
F and
N
F layers are binary values that are stored in matrix U
and updated during learning as, 1
jk
u = if
j
m is a HS of class
k
n , otherwise 0
jk
u = , for 1, 2,........., k p = and
1, 2,........., j q = , where
j
m is the th j
M
F node and
k
n is the
th k
N
F node. Each
N
F node performs the union of fuzzy
values returned by HSs described as
1,2.....p k max
1
= =
=
for u m n
jk j
q
j
k
(3)
Each
o
F node delivers non-fuzzy output described as,
0,
k
O = if ,
k
n T < otherwise, 1,
k
O = if ,
k
n T = for
1, 2,........., k p = where max( ),
k
T n = for 1, 2,........., k p = .
III. MFHSNN LEARNING ALGORITHM
The supervised MFHSNN learning algorithm for creating
fuzzy hyperspheres in hyperspace consists of three steps.
A. Creating of HSs
Given the th h training pair ( , )
h h
R d find all the
hyperspheres belonging to the class
h
d . These hyperspheres
are arranged in ascending order according to the distances
between the input pattern and the center point of the
Vol. 2, No. 6, 2011
132 | P a g e
hyperspheres. After this following steps are carried sequentially
for possible inclusion of input pattern
h
R .
Step 1: Determine whether the pattern
h
R is contained by
any one of the hyperspheres. This can be verified by using
fuzzy hypersphere membership function defined in equation
(4). If
h
R is contained by any of the hypersphere then it is
included, therefore in the training process all the remaining
steps are skipped and training is continued with the next
training pair.
Step 2: If the pattern
k
R falls outside the hypersphere,
then the hypersphere is expanded to include the pattern if the
expansion criterion is satisfied. For the hypersphere
j
m to
include R
h
the following constraint must be met.
s

=
n
i
hi ji
r c
1
(4)
Here we have proposed a new approach for testing
expansion of new hyperspheres based on Bhattacharyya
distance which is sum of the absolute differences yields
superior results as compared to Euclidian distance and
Manhattan distance.
If the expansion criterion is met then pattern
k
R is included
as

=

=
n
i
hi ji
new
j
r c
1
. (5)

Step 3: If the pattern
k
R is not included by any of the above
steps then new hypersphere is created for that class, which is
described as

0 and = =
new h new
R C . (6)
B. Overlap Test
The learning algorithm allows overlap of hyperspheres
from the same class and eliminates the overlap between
hyperspheres from different classes. Therefore, it is necessary
to eliminate overlap between the hyperspheres that represent
different classes. Overlap test is performed as soon as the
hypersphere is expanded by step 2 or created in step 3.
Overlap test for step 2: Let the hypersphere is expanded to
include the input pattern
k
R and expansion has created overlap
with the hypersphere,
v
m which belongs to other class, which
belongs to other class. Suppose,

1 1
[ , ,........ ]
u n
C x x x = and
u
.
represents center point and radius of the expanded hypersphere
and
' ' '
1 2
[ , ,....... ]
v n
C x x x = and ,
v
. are centre point and radius
of the hypersphere of other class as depicted in Fig 3(a). Then
if

v u
n
i
hi ji
r c . . + s

=1
(7)
means those hyperspheres from separate classes are
overlapping.
a) Overlap test for step 3
If the created hypersphere falls inside the hypersphere of
other class means there is an overlap. Suppose
p
m represents
created hypersphere to include the input
h
R and
q
m represents
the hypersphere of other class as shown in Fig 4b The
presence of overlap in this case can be verified using the
membership function defined in equation (1). If
( , , ) ( , , ) 1
q h p p q h q q
m R C m R C . . = = means two hyperspheres
from different classes are overlapping.
Figure 3. (a) Status of the hypersphere before removing an overlap in step 2
Figure 3. (b) Status of the hyperspheres after removing an overlap in step 2
b) Removing Overlap
If step 2 has created overlap of hyperspheres from separate
classes then overlap is removed by restoring the radius of just
expanded hypersphere. Let,
u
m

be the expanded hypersphere
then it is contracted as

new
u
new
u
. . = (8)
and new hypersphere is created for the input pattern as
described by equation (11). This situation is shown in Fig 3(b).
If the step 3 creates overlap then it is removed by modifying of
other class. Let
1 1
[ , ,........ ]
p n
C x x x = and
p
. , represents centre
point and radius of the
' ' ' '
1 2 3
[ , , ,........ ]
q n
C x x x x = and ,
q
. are
center point and radius of the
Vol. 2, No. 6, 2011
133 | P a g e
o .
=

=
n
i
hi ji
new
j
r c
1
(9)
Figure 4. (a) Status of hypersphere before removing an overlap in step 3.

Figure 4. (b) Status of hypersphere after removing an overlap in step 3.
IV. IRIS SEGMENTATION AND FEATURE EXTRACTION
Iris Segmentation plays very important role for detecting
the iris patterns, Segmentation is to locate valid part of the iris
for iris biometric [3]. Finding the inner and outer boundaries
(pupillary and limbic) are as shown in the Fig 5(a). Localizing
its upper and lower eyelids if they occlude, and detecting and
excluding any overlaid occlusion of eyelashes and their
reflection. The best known algorithm for iris segmentation is
Daugmans intergro-differential operator to find boundaries of
iris as defined.

0 0
0 0
( , , )
( , )
max( , , )
2
r x y
I x y
r x y G ds
r r
o
t
c
-
c
(10)
Figure 5. (a) Inner and Outer boundaries are detected with different radius

Figure 5. (b) ROI is extracted
Iris has a particularly interesting structure and provides rich
texture information. Here we have implemented principal
component analysis method for feature extraction which
captures local underlying information from an isolated image.
The result of this method yields high dimension feature vector.
To improve training and recalling efficiency of the network,
here we have used Singular value decomposition (SVD)
method to reduce the dimensionality of the feature vector, and
MFHSNN is used for Classification. SVD is a method for
identifying and ordering the dimensions along which the
feature exhibit the most variation [13]. The most variation are
identified the best approximation of the original data points
using less dimensions. This can be described as:

T
X UTV = (11)
the covariance matrix is be defined as:

2
1 1
T T
C XX UT U
n n
= = (12)
U is a n m matrix. SVD performs repetitively order the
singular values in descending order, if n m < , the first n
columns in U corresponds to the sorted eigenvalues of C and
if the first mcorresponds to the sorted non-zero eigenvalues
of . C The transformed data can thus be written as:

T T
T
Y U X U UTV = = (13)
Where
T
U U is a simple n x m matrix which is one on the
diagonal and zero. Hence Equation 13 is decomposition of
equation 11.
V. EXPERIMENTAL RESULTS
CASIA Iris Image Database Version 1.0 (CASIA-IrisV1)
includes 756 iris images from 108 eyes. For each eye, 7 images
are captured in two sessions, where three samples are collected
in the first session and four samples in second. For each iris
class, we choose two samples from each session for training
and remaining as testing samples. Therefore, there are 540
images for training and 216 images for testing. The timing
analysis of training and recall, recognition rates in terms of
number of hyperspheres, radius and recognition rate are
depicted in Table 1 and Table 2.
Figure 6. Specimen Feature Vectors of 20 Iris Pattrns
Vol. 2, No. 6, 2011
134 | P a g e
TABLE I. PERFORMANCE EVALUATION OF DIFFERENT DISTANCE
MEASURES ON MFHSNN
TABLE II. RECOGNITION RATE OF MFHSNN ON CASIA IMAGES

VI. CONCLUSION AND FUTURE WORK
In this paper, we described an iris recognition algorithm
using MFHSNN which has ability to learn patterns faster by
creating /expanding HSs. It has been verified on CASIA
database the result is as shown in Table 1 and Table II.
MFHSNN can also be adapted for applications in some other
pattern recognition problems. Our future work will be to
improve the iris recognition rate using fuzzy neural networks.
VII. REFERENCES
[1] Li Ma, T Tan and D.Zhang and Y.Wang. Personal Identification Based
on Iris Texture Analysis, IEEE Trans. Pattern Anal. Machine Intell,
25(12):1519-1533, 2003.
[2] Dey, Somnath, Samanta Debasis. Improved Feature Processing for Iris
Biometric Authentication System, International Journal of Computer
Science and Engineering, 4(2):127-134.
[3] Daugman. J.G, How Iris Recognition Works, Proceedings of 2002
International Conference on Image Processing, Vol. 1, 2002.
[4] Li Ma, Tieniu Tan, Yunhong Wang, and Dexin Zhang. Efficient Iris
Recognition by Characterizing Key Local Variations, IEEE
Transactions on Image Processing, 13(6):739750, 2004.
[5] Wildes R P. Iris Rcognition: An Emerging Biometric Technology. In
Proceeding of the IEEE, Piscataway, NJ, USA, 1997, 85:1348-1363.
[6] S.Lim, K.Lee, O.Byeon and T.Kim, Efficient iris recognition through
improvement of feature vector and classifier, ETRI J., 23(2): 1-70,
2001.
[7] Daugman J.G, Demodulation by complex-valued wavelets for
stochastic pattern recognition, International Journal of Wavelets,
Multiresolution, and Information Processing, 1( 1):117, 2003.
[8] Tisse C.-L. and Torres L. Michel, Robert, Person Identification
Technique Using Human Iris Recognition, Proceedings of the 15th
International Conference on Vision Interface, 2002, pp. 294-299.
[9] Poursaberi A. and Arrabi B.N, Iris Recognition for Partially occluded
images Methodology and Sensitive Analysis, Hindawi Publishing
corporation journal on Advances in Signal Processing, 2007.
[10] U.V. Kulkarni and T.R. Sontakke, Fuzzy Hypersphere Neural Network
Classifier, Proceedings of 10th International IEEE Conference on
Fuzzy Systems, University of Melbourne, Australia, December 2001.
[11] Krishna Kanth B. B. M, Kulkarni U. V. and Giridhar B. G. V., Gene
Expression Based Acute Leukemia Cancer Classification: a Neuro-
Fuzzy Approach International Journal of Biometrics and
Bioinformatics, (IJBB), Volume (4): Issue (4) pp. 136146, 2010.
[12] Ruggero Donida Labati et.al, Neural-based Iterative Approach for Iris
Detection in Iris recognition systems, Proceedings of the 2009 IEEE
Symposium on Computational Intelligence in Security and Defense
Applications (CISDA 2009).
[13] Madsen, R., Hansen, L., &Winther, O. Singular value decomposition
and principal component analysis, isp technical report, 2004.

AUTHORS PROFILE
Santosh S. Chowhan received the M.Sc.(CS) degree from Dr. BAM
University, Aurangabad, Maharashtra, India in the year 2000. He is
currently working as lecturer in the College of Computer Science and
Information Technology, Latur, Maharastra. His current research
interests include various aspects of Neural Networks and Fuzzy Logic,
Pattern Recogntion and Biometrics
Uday V. Kulkarni received Ph.D degree in Electronics and Computer
Science Engineering from S. R. T. M. University , Nanded in the year
2003. He is currently working as a professor in Dept of Computer Scince
and Engineering in SGGS Institute of Engineering and Technology,
Nanded, Maharashtra, India.
Ganesh N. Shinde received M. Sc. & Ph.D. degree from Dr. B.A. M.
University, Aurangabad. He is currently working as Principal in Indira
Gandhi College, Nanded and Maharashtra, INDIA.. He has published 27
papers in the International Journals and presented 15 papers in
International Conferences. In his account one book is published, which
is reference book for different courses. at Honolulu, U.S.A. He is
member of Management Council & Senate of S.R.T.M. University,
Nanded, INDIA. His research interest includes Filters, Image processing,
Pattern recognition, Fuzzy Systems, Neural Network and Multimedia
analysis and retrieval system.

Distance
Measures
Training Recall Hypersphere Radius
Euclidean 0.830694 31.780012 347 0.008
Manhattan 0.867152 31.660341 364 0.022
Bhattacharya 0.750051 23.801397 377 0.0000001
Methodology Recognition rate
Daugman HD,SVM 99.25
Wildes Normalized Correlation 97.43
Y.Wang WED 99.57
Ali HD, WED 95.20
MFHSNN 94.00
Vol. 2, No. 6, 2011
135 | P a g e
Multi-Agent System Testing: A Survey

Zina Houhamdi
Software Engineering Department, Faculty of Science and IT
Al-Zaytoonah University
Amman, Jordan

AbstractIn recent years, agent-based systems have received
considerable attention in both academics and industry. The
agent-oriented paradigm can be considered a natural extension to
the object-oriented (OO) paradigm. Agents differ from objects in
many issues which require special modeling elements but have
some similarities. Although there is a well-defined OO testing
technique, agent-oriented development has neither a standard
development process nor a standard testing technique. In this
paper, we will give an introduction to most recent works
presented in the area of testing distributed systems composed of
complex autonomous entities (agents). We will provide pointers
to work by large players in the field. We will explain why this
kind of system must be handled differently than less complex
systems.
Keywords-Software agent; Software testing; Multi-agent system
testing.
I. INTRODUCTION
As the technology evolving, the more we are driven
towards abstraction and generalization. The increasing use of
Internet as the spine for all interconnected services and
devices makes software systems highly complex and in
practice open in scale. These systems nowadays need to be
adaptive, autonomous and dynamic to serve different user's
community and heterogeneous platforms. These systems are
developed very fast in past few decades. They are changed
continuously to satisfy the business and technology
modifications.
Software agents are key technologies to meet modern
business needs. They offer also an efficient conceptual
methodology to design such complex systems. In practice,
research on software agents' development and Multi-Agent
System (MAS) has become too large and used in different
active area focusing mainly on architectures, protocols,
frameworks, messaging infrastructure and community
interactions. Thus, these systems receive more industrial
attention as well.
Since these systems are increasingly taking over operations
and controls in organization management, automated vehicles,
and financing systems, assurances that these complex systems
operate properly need to be given to their owners and their
users. This calls for an investigation of appropriate software
engineering frameworks, including requirements engineering,
architecture, and testing techniques, to provide adequate
software development processes and supporting tools.
Software agents and MAS testing is a challenging task
because these systems are distributed, autonomous, and
deliberative. They operate in an open world, which requires
context awareness. In particular, the very particular character
of software agents makes it difficult to apply existing software
testing techniques to them. There are issues concerning
communication and semantic interoperability, as well as
coordination with peers. All these features are known to be
hard not only to design and to program [3], but also to test.
There are several reasons for the increase of the difficulty
degree of testing MAS:
Increased complexity, since there are several
distributed processes that run autonomously and
concurrently;
Amount of data, since systems can be made up by
thousands of agents, each owning its own data;
Irreproducibility effect, since we cant ensure that two
executions of the systems will lead to the same state,
even if the same input is used. As a consequence,
looking for a particular error can be difficult if it is
impossible to reproduce it each time [22].
They are also non-deterministic, since it is not
possible to determine a priori all interactions of an
agent during its execution.
Agents communicate primarily through message
passing instead of method invocation, so existing
object-oriented testing approaches are not directly
applicable.
Agents are autonomous and cooperate with other
agents, so they may run correctly by themselves but
incorrectly in a community or vice versa.
As a result, testing software agents and MAS asks for new
testing methods dealing with their specific nature. The
methods need to be effective and adequate to evaluate agent's
autonomous behaviors and build confidence in them. From
another perspective, while this research field is becoming
more advanced, there is an emerging need for detailed
guidelines during the testing process. This is considered a
crucial step towards the adoption of Agent-Oriented Software
Engineering (AOSE) methodology by industry.
Several AOSE methodologies have been proposed [17,
34]. While some work considered specification-based formal
verification [11, 14], others borrow object-oriented testing
techniques, taking advantage of a projection of agent-oriented
abstractions into object-oriented constructs, UML for instance
Vol. 2, No. 6, 2011
136 | P a g e
[9, 33]. However, to the best of our knowledge, none of
existing work provides a complete and structured testing
process for guiding the testing activities. This is a big gap that
we need to bridge in order for AOSE to be widely applicable.
II. SOFTWARE TESTING
Software testing is a software development phase, aimed at
evaluating product quality and enhancing it by detecting errors
and problems. Software testing is an activity in which a
system or component is executed under specified conditions,
the results are observed or recorded and compared against
specifications or intended results, and an estimation is made of
some aspect of the system or component. A test is a set of one
or more test cases.
The principal test goal is to find faults different from
errors. An error is a mistake made by the developer
misunderstanding something. A fault is an error in a program.
An error may lead to one or more faults. When a fault is
executed then an execution error may occur. An execution
error is any result or behavior that is different from what has
been specified or is expected by the user. The observation of
an execution error is a failure. Notice that errors may be
unobservable and as a consequence may play severe disrupt
with the left over computation and use of the results of this
computation. The greater the period of unobserved operation,
the larger is the probability of serious damage due to errors
that is caused by unnoticed failures.
As showed in Figure 1, software testing consists of the
dynamic verification of the program behavior on a set of
suitably selected test cases. Different from static verification
activities, like formal proofing or model checking, testing
implies running the system under test using specified test
cases [22].

Figure 1. Kinds of Tests
There are several strategies for testing software and the
goal of this survey is not to explain all of them. Nevertheless,
we will describe the main strategies found in literature [22,
35]. Here they are:
Black-box testing: also know as functional testing or
specification-based testing. Testing without reference
to the internal structure of the component or system.
White-box testing: testing based on an analysis of the
internal structure of the component or system. Test
cases are derived from the code e.g. testing paths.
Progressive testing: it is based on testing new code to
determine whether it contains faults.
Regressive testing: it is the process of testing a
program to determine whether a change has
introduced faults (regressions) in the unchanged code.
It is based on reexecution of some/all of the tests
developed for a specific testing activity.
Performance testing: verify that all worst case
performance and any best-case performance targets
have been met.
There are several types of tests. The most frequently
performed are the unit test and integration test. A unit test
performs the tests required to provide the desired coverage for
a given unit, typically a method, function or class. A unit test
is white-box testing oriented and may be performed in parallel
with regard to other units. An integration test provides testing
across units or subsystems. The test cases are used to provide
the needed system, as a whole, coverage. It tests subsystem
connectivity. There are several strategies for implementing
integration test:
Bottom-up, which tests each unit and component at
lowest level of system hierarchy, then components that
call these and so on;
Top-down, which tests top component and then all
components called by this and so on;
Big-bang, which integrates all components together;
Sandwich, which combines bottom-up with top-down
approach.
On the other hand, the goal of software testing is also to
prevent defects, as it is clearly much better to prevent faults
than to detect and correct them because if the bugs are
prevented, there is no code to correct. This approach is used in
cleanroom software development [22]. Designing tests is
known as one of the best bug prevention activities. Tests
design can discover and eliminate bugs at every stage in the
software construction process [2]. Therefore, the idea of "test
first, then code" or test-driven is quite widely discussed today.
To date, several techniques have been defined and used by
software developers [1].
Recently, a new testing technique called Evolutionary
testing (ET) [27, 41] has been presented. The technique is
inspired by the evolution theory in biology that emphasizes
natural selection, inheritance, and variability. Fitter individuals
have a higher chance to survive and to reproduce offspring;
and special characteristics of individuals are inherited. In ET,
we usually encode each test case as an individual; and in order
to guide the evolution towards better test suites, a fitness
measure is a heuristic approximation of the distance from
achieving the testing goal (e.g., covering all statements or all
branches in the program). Test cases having better fitness
values have a higher chance to be selected in generating new
test cases. Moreover, mutation is applied during reproduction
in order to generate more different test set. The key step in ET
is the transformation from testing objective to search problem,
specifically fitness measure. Different testing objective gives
rise to different fitness definitions. Once a fitness measure has
been defined, different optimization search techniques, such as
local search, genetic algorithm, particle swarm [27] can be
used to generate test cases towards optimizing fitness measure
(or testing objective, i.e. finding faults).
Vol. 2, No. 6, 2011
137 | P a g e
III. SOFTWARE AGENTS AND MAS TESTING
A software agent is a computer program that works toward
goals in a dynamic context on behalf of another entity (human
or computational), perhaps for a long period of time, with
discontinuous direct supervision or control, and exhibits a
significant flexibility and even creativity degree in how it tries
to transform goals into action tasks [18].
Software agents have (among others) the following properties:
1. Reactivity: agents are able to sense contextual changes
and react appropriately;
2. Pro-activity: agents are autonomous, so they are able
to select which actions to take in order to reach their
goals in given situations;
3. Social ability: that is, agents are interacting entities,
which cooperate, share knowledge, or compete for
goal achievement
A multi-agent system (MAS) is a computational context in
which individual software agents interact with each other, in a
collaborative (using message passing) or competitive manner,
and sometimes autonomously trying to attain their individual
goals, accessing resources and services of the context, and
occasionally producing results for the entities that initiated
those software agents [25]. The agents interact in a concurrent,
asynchronous and decentralized manner [21] hence MAS turn
out to be complex systems [23]. Consequently, they are
difficult to debug and test.
Due to those peculiar characteristics of agents and MAS as
a whole, testing them is a challenging task that should address
the following issues. (Some of them were stated in [37]):
Distributed/asynchronous: Agents operate concurrently
and asynchronously. An agent might have to wait for other
agents to fulfill its intended goals. An agent might work
correctly when it operates alone but incorrectly when put into
a community of agents or vice versa. MAS testing tools must
have a global view over all distributed agents in addition to
local knowledge about individual agents, in order to check
whether the whole system operate accordingly to the
specifications. In addition, all the issues related to testing
distributed systems are applied in testing software agent and
MAS as well, for example problems with controllability and
observability [6].
Autonomous: Agents are autonomous. The same test inputs
may result in different behaviors at different executions, since
agents might modify their knowledge base between two
executions, or they may learn from previous inputs, resulting
in different decisions made in similar situations.
Message passing: Agents communicate through message
passing. Traditional testing techniques, involving method
invocation, cannot be directly applied.
Environmental and normative factors: Context and
conventions (norms, rules, and laws) are important factors that
govern or influence the agents' behaviors. Different contextual
settings may affect the test results. Occasionally, a context
gives means for agents to communicate or itself is a test input.
Scaled agents: In some particular cases, agents could be
seen as scaled in that they provide no or little observable
primitives to the outside world, resulting in limited access to
the internal agents' state and knowledge. An example could be
an open MAS that allows third-party agents to come in and
access to the resources of the MAS, how do we assure that the
third-party agents with limited knowledge about their
intentions behave properly?
The agent oriented methodologies provide a platform for
making MAS abstract, generalize, dynamic and autonomous.
However, many methodologies like MASE, Prometheus, and
Tropos do exist for the agent oriented framework but on
contrary to it the testing techniques for the methodologies are
not clearly supported [10].
A. Test Levels
Over the last years, the view of testing has evolved, and
testing is no longer seen as a step which starts only after the
implementation phase is finished. Software testing is now seen
as a whole process that filters in the development and
maintenance activities. Thus, each development phase and
maintenance phase should have a corresponding test level.
Figure 2 shows V model in which the correspondence between
development process phases and test levels are highlighted
[28].

Figure 2. V model
Work in testing software agents and MAS can be classified
into different testing levels: unit, agent, integration, system,
and acceptance. Here we use general terminologies rather than
using specific ones used in the community like group, society.
Group and society, as called elsewhere, are equivalent to
integration and system, respectively. The testing objectives,
subjects to test, and activities of each level are described as
follows:
Unit testing tests all units that make up an agent,
including blocks of code, implementation of agent
units like goals, plans, knowledge base, reasoning
engine, rules specification, and so on; make sure that
they work as designed.
Agent testing tests the integration of the different
modules inside an agent; test agents' capabilities to
fulfill their goals and to sense and effect the
environment.
Integration or Group testing tests the interaction of
agents, communication protocol and semantics,
interaction of agents with the environment, integration
of agents with shared resources, regulations
enforcement; Observe emergent properties, collective
behaviors; make sure that a group of agents and
environmental resources work correctly together.
Vol. 2, No. 6, 2011
138 | P a g e
System or Society testing tests the MAS as a system
running at the target operating environment; test the
expected emergent and macroscopic properties of the
system as a whole; test the quality properties that the
intended system must reach, such as adaptation,
openness, fault tolerance, performance.
Acceptance testing tests the MAS in the customer's
execution environment and verifies that it meets
stakeholder goals, with the participation of
stakeholders.
B. MAS Testing Problems
Defining a structured testing process for software agents
and MAS: Currently, AOSE methodologies have been
interesting principally on requirement analysis, design, and
implementation; limited attention was given to validation and
verification, as in Formal Tropos [11, 14]. A structured testing
process that complements analysis and design is still absent.
This problem is determinant because without detailed and
systematic guidelines, the development cost may increase in
terms of effort and productivity.
1. They have their own reasons for engaging in proactive
behaviors that might differ from a user's concrete
expectation, yet are still appropriate.
2. The same test input can give different results in
different executions.
3. Agents cooperate with other agents, so they may run
correctly by themselves but incorrectly in a
community or vice versa.
4. Moreover, agents can be programmed to learn; so
successive tests with the same test data may give
different results.
As a conclusion, defining adequate and effective
techniques to test software agents is, thus, a key problem in
agent development.
IV. A SURVEY OF TESTING MULTI-AGENT SYSTEMS
There is very brief written work that describes agents
software testing. The remainder of this section surveys recent
and active work on testing software agents and MAS, with
respect to previous categories. This classification is intended
only to facilitate easily understand the research work in the
field. It is also interesting to notice that this classification is
incomplete in the sense that some work addresses testing in
more than one level, but we put them in the level they
principally focus.
A. Unit Testing
Unit testing approach calls attention to the test of the
smallest building blocks of the MAS: the agents. Its essential
idea is to check if each agent in isolation respects its
specifications under normal and abnormal conditions. Unit
testing needs to make sure that all units that are parts of an
agent, like goals, plans, knowledge base, reasoning engine,
rules specification, and even blocks of code work as designed.
Effort has been spent on some particular elements, such as
goals, plans. Nevertheless, a complete approach addressing
unit testing in AOSE still opens room for research. An analogy
of expected results can be those of unit testing research in the
object-oriented development. At the unit level,
1. Zhang et al. [42] introduced a model based testing
framework using the design models of the Prometheus
agent development methodology [31]. Different from
traditional software systems, units in agent systems
are more complex in the way that they are triggered
and executed. For instance, plans are triggered by
events. The framework focuses on testing agent plans
(units) and mechanisms for generating suitable test
cases and for determining the order in which the units
are to be tested.
2. Ekinci et al. [13] claimed that agent goals are the
smallest testable units in MAS and proposed to test
these units by means of test goals. Each test goal is
conceptually decomposed into three sub-goals: setup
(prepare the system), goal under test (perform actions
related to the goal), and assertion goal (check goal
satisfaction). The first and last goal prepares pre-
conditions and check post-conditions while testing the
goal under test, respectively. Moreover, they introduce
a testing tool, called as SEAUnit that provides necessary
infrastructure to support proposed approach.
B. Agent testing
At the agent level we have to test the integration of the
different modules inside an agent, test agents' capabilities to
achieve their goals and to sense and effect the context. There
is several works in agent testing level.
1. Agile PASSI [7] proposes a framework to support
tests of single agents. They develop a test suite
specifically for agent verification. Test plans are
prepared before the coding phase in according with
specifications and the AgentFactory tool is also able of
generating driver and stub agents for speeding up the
test of a specific agent. Despite proposing valuable
ideas concerning MAS potential levels of tests, PASSI
testing approach is poorly documented and does not
offer techniques to help developers in the low level
design of unit test cases.
2. Lam and Barber [26] proposed a semi-automated
process for comprehending software agent behaviors.
The approach imitates what a human user (can be a
tester) does in software comprehension: building and
refining a knowledge base about the behaviors of
agents, and using it to verify and explain behaviors of
agents at runtime. Although the work did not deal with
other problems in testing, like the generation and
execution of test cases, the way it evaluates agent
behaviors is interesting and relevant for testing
software agents.
3. Nunez et al. [30] introduced a formal framework to
specify the behavior of autonomous e-commerce
agents. The desired behaviors of the agents under test
are presented by means of a new formalism, called
utility state machine that embodies users' preferences
in its states. Two testing methodologies were proposed
Vol. 2, No. 6, 2011
139 | P a g e
to check whether an implementation of a specified
agent behaves as expected (i.e., conformance testing).
In their active testing approach, they used for each
agent under test a test (a special agent) that takes the
formal specification of the agent to facilitate it to
reach a specific state. The operational trace of the
agent is then compared to the specification in order to
detect faults. On the other hand, the authors also
proposed to use passive testing in which the agents
under test were observed only, not stimulated like in
active testing. Invalid traces, if any, are then identified
thanks to the formal specifications of the agents.
4. Coelho et al. [8] proposed a framework for unit testing
of MAS based on the use of Mock Agents. Even
though they called it unit testing but their work
focused on testing roles of agents at agent level
according to our classification. Mock agents that
simulate real agents in communicating with the agent
under test were implemented manually; each
corresponds to one agent role. Sharing the inspiration
from JUnit [15] with Coelho et al. [8], Tiryaki et al.
[40] proposed a test-driven MAS development
approach that supported iterative and incremental
MAS construction. A testing framework called SUnit,
which was built on top of JUnit and Seagent [12], was
developed to support the approach. The framework
allows writing tests for agent behaviors and
interactions between agents.
5. Gomez-Sanz et al. [16] introduced advances in testing
and debugging made in the INGENIAS methodology
[33]. The meta-model of INGENIAS has been
extended with concepts for defining tests to
incorporate the declaration of testing, i.e., tests and
test packages. The code generation facilities are
augmented to produce JUnit-based test case and suite
skeletons based on these definitions with respect to
debugging and it is the developer's task to modify
them as needed. The work also provided facilities to
access mental states of individual agents to check
them at runtime. The system is integrated with
ACLAnalyzer [4], a data mining facility for capturing
agent communication and exploring them with
different graphical representations.
6. Houhamdi [18] introduces a suite test derivation
approach for Agent testing that takes goal-oriented
requirements analysis artifact as the core elements for
test case derivation. The proposed process has been
illustrated with respect to the Tropos development
process. It provides systematic guidance to generate
test suites from agent detailed design. These test
suites, on the one hand, can be used to refine goal
analysis and to detect problems early in the
development process. On the other hand, they are
executed afterwards to test the achievement of the
goals from which they were derived.
C. Integration Testing
Integration testing test the interaction of agents,
communication protocol and semantics, interaction of agents
with the context, integration of agents with shared resources,
regulations enforcement; observe emergent properties; make
sure that a group of agents and environmental resources work
correctly together.
Only a few of methodologies define an explicit verification
process by proposing a verification phase based on model
checking to support automatic verification of inter-agent
communications. Only some iterative methodologies propose
incremental testing processes with supporting tools. At the
integration level, effort has been put in agent interaction to
verify dialogue semantics and workflows.
1. Agile [24] defines a testing phase based on JUnit test
framework [15]. In order to use this tool, designed for
OO testing, in MAS testing context, they needed to
implement a sequential agent platform, used strictly
during tests, which simulates asynchronous message-
passing. Having to execute unit tests in an
environment different from the production
environment results in a set of tests that does not
explore the hidden places for failures caused by the
timing conditions inherent in real asynchronous
applications.
2. The ACLAnalyser [4] tool runs on the JADE [39]
platform. It intercepts all messages exchanged among
agents and stores them in a relational database. This
approach exploits clustering techniques to build agent
interaction graphs that support the detection of missed
communication between agents that are expected to
interact, unbalanced execution configurations,
overhead data exchanged between agents. This tool
has been enhanced with data mining techniques to
process results of the execution of large scale MAS
[5].
3. Padgham et al. [32] use design artifacts (e.g., agent
interaction protocols and plan specification) to provide
automatic identification of the source of errors
detected at run-time. A central debugging agent is
added to a MAS to monitor the agent conversations. It
receives a carbon copy of each message exchanged
between agents, during a specific conversation.
Interaction protocol specifications corresponding to
the conversation are fired and then analyzed to detect
automatically erroneous conditions.
4. Also at the integration level but pursuing a deontic
approach, Rodrigues et al. [36] proposed to exploit
social conventions, i.e. norms, rules, that prescribe
permissions, obligations, and/or prohibitions of agents
in an open MAS to integration test. Information
available in the specifications of these conventions
gives rise to a number of types of assertions, such as
time to live, role, cardinality, and so on. During test
execution a special agent called Report Agent will
observe events and messages in order to generate
analysis report afterwards.
5. Ekinci et al. [13] view integration testing of MAS
rather abstract. They considered system goals as the
source cause for integration and use them as driving
Vol. 2, No. 6, 2011
140 | P a g e
criteria. They apply the same approach for testing
agent goals (unit according to their view) to test these
goals. They define the concept of test goal. This
concept represents the group of tests needed in order
to check if the system goal is achieved correctly.
6. Nguyen et al. [29] propose using ontologies extracted
from MAS under test and a set of OCL constraints,
which act as a test oracle. Having as input a
representation of the ontologies used, the idea is to
construct an agent able to deliver messages whose
content is inspired by these ontologies. The resulting
behaviors are regarded as correct using the input set of
OCL constraints: if the message content satisfies the
constraints, the message is correct. The procedure is
support by eCAT, a software tool.
7. Houhamdi and Athamena [19] introduced a novel
approach for goal-oriented software integration
testing. They propose a test suite derivation approach
for integration testing that takes goal-oriented
requirements analysis artifact for test case derivation.
They have discussed how to derive test suites for
integration test from architectural and detailed design
of the system goals. These test suites can be used to
observe emergent properties resulting from agent
interactions and make sure that a group of agents and
contextual resources work correctly together. This
approach defines a structured and comprehensive
integration test suite derivation process for
engineering software agents by providing a systematic
way of deriving test cases from goal analysis.
D. System and Acceptance Testing
System testing tests; test for quality properties, such as
adaptation, openness, fault-tolerance, performance.
At the system level of testing MAS, one has to test the
MAS as a system running at the target operating environment;
test the expected emergent and macroscopic properties and/or
the expected qualities that the intended system as whole must
reach. Some initial effort has been devoting to the validation
of macroscopic behaviors of MAS.
Sudeikat and Renz [38] proposed to use the system
dynamics modeling notions for the validation of MAS.
These allow describing the intended, macroscopic
observable behaviors that originate from structures of
cyclic causalities. System simulations are then used to
measure system state values in order to examine
whether causalities are observable.
Houhamdi and Athamena [20] introduced a suite test
derivation approach for system testing that takes goal-
oriented requirements analysis artifact as the core
elements for test case derivation. The proposed
process has been illustrated with respect to the Tropos
development process. It provides systematic guidance
to generate test suites from modeling artifacts
produced along with the development process. They
have discussed how to derive test suites for system test
from late requirement and architectural design. These
test suites, on the one hand, can be used to refine goal
analysis and to detect problems early in the
development process. On the other hand, they are
executed afterwards to test the achievement of the
goals from which they were derived.
Acceptance testing tests the MAS in the customer
execution environment and verifies that it meets the
stakeholder goals, with the participation of stakeholders. To
the best of our knowledge, there is no work dealing explicitly
with testing MAS at the acceptance level, currently. In fact,
agent, integration, and system test harnesses can be reused in
acceptance test, providing execution facilities. However, as
testing objectives of acceptance test differ from those of the
lower levels, evaluation metrics at this level, such as metrics
for openness, fault-tolerance and adaptivity, demand for
further research.
V. CONCLUSION
In summary, most of the existing research work on testing
software agent and MAS focuses mainly on agent and
integration level. Basic issues of testing software agents like
message passing, distributed/asynchronous have been
considered; testing frameworks have been proposed to
facilitate testing process. And yet, there is still much room for
further investigations, for instance:
A complete and comprehensive testing process for
software agents and MAS.
Testing MAS at system and acceptance level: how do
the developers and the end-users build confidence in
autonomous agents?
Test inputs definition and generation to deal with open
and dynamic nature of software agents and MAS.
Test oracles, how to judge an autonomous behavior?
How to evaluate agents that have their own goals from
human tester's subjective perspectives?
Testing emergent properties at macroscopic system
level: how to judge if an emergent property is correct?
How to check the mutual relationship between
macroscopic and agent behaviors?
Deriving metrics to assess the qualities of the MAS
under test, such as safety, efficiency, and openness.
Reducing/removing side effects in test execution and
monitoring because introducing new entities in the
system, e.g., mock agents tester agents, and
monitoring agent as in many approaches, can
influence the behavior of the agents under test and the
performance of the system as a whole.
REFERENCES
[1] K. Beck, Test Driven Development: By Example, Addison-Wesley
Longman Publishing Co., Boston, USA, 2005.
[2] B. Beizer, Software Testing Techniques, 2nd edition, Van Nostrand
Reinhold Co., New York, NY, USA, 1990.
[3] F. Bergenti, M. Gleizes, and F. Zambonelli, Methodologies and Software
Engineering for Agent Systems, The Agent-Oriented Software
Engineering Handbook, Springer, Vol. 11, 2004.
Vol. 2, No. 6, 2011
141 | P a g e
[4] J. Botia, A. Lopez-Acosta, and G. Skarmeta, ACLAnalyser: A tool for
debugging multi-agent systems, Proceeding of the 16th European
Conference on Artificial Intelligence, pp. 967-968, IOS Press 2004.
[5] J. Botia, J. Gomez-Sanz, and J. Pavon, Intelligent Data Analysis or the
Verification of Multi-Agent Systems Interactions, 7th International
Conference of Intelligent Data Engineering and Automated Learning,
Burgos, Spain, September 20-23, pp. 1207-1214, 2006.
[6] L. Cacciari, and O. Rafiq, Controllability and observability in
distributed testing, Information and Software Technology, Vol. 41, 11-
12, pp. 767-780. 1999.
[7] G. Caire, M. Cossentino, A. Negri, and A. Poggi, Multi-agent systems
implementation and testing, Proceedings of the 7th European Meeting
on Cybernetics and Systems Research - EMCSR2004, Vienna, Austrian
Society for Cybernetic Studies, pp. 14-16, 2004.
[8] R. Coelho, U. Kulesza, A. Staa, and C. Lucena, Unit testing in multi-
agent systems using mock agents and aspects, Proceedings of the
international workshop on Software engineering for large-scale multi-
agent systems, ACM Press, New York, pp. 8390, 2006.
[9] M. Cossentino, From Requirements to Code with PASSI
Methodology, In Vijayan Sugumaran (Ed.), Intelligent Information
Technologies: Concepts, Methodologies, Tools, and Applications, USA,
2008.
[10] K.H. Dam, and M. Winikoff, Comparing Agent-Oriented
Methodologies, 5th International Bi-Conference Workshop, AOIS
2003 at AAMAS 2003, Melbourne, Australia, July 14, pp. 78-93, 2003.
[11] A. Dardenne, A. Lamsweerde, and S. Fickas, Goal-directed
requirements acquisition, Science of Computer Programming 20(1-2),
pp.3-50, 1993.
[12] O. Dikenelli, R. Erdur, and O. Gumus, Seagent: a platform for
developing semantic web based multi agent systems, AAMAS'05
Proceedings of the fourth International Joint Conference on
Autonomous agents and multi-agent systems, ACM Press, New York,
pp. 12711272, 2005.
[13] E. Ekinci, M. Tiryaki, O. Cetin, and O. Dikenelli, Goal-Oriented
Agent Testing Revisited, Proceeding of the 9th International
Workshop on Agent-Oriented Software Engineering, pp. 85-96, 2008.
[14] A. Fuxman, L. Liu, J. Mylopoulos, M. Pistore, and M. Roveri,
Specifying and analyzing early requirements in Tropos, Requirement
Engineering, Springer Link, Vol. 9, 2, pp. 132-150, 2004.
[15] E. Gamma, and K. Beck, JUnit: A Regression Testing Framework,
http://www.junit.org, 2000.
[16] J. Gomez-Sanz, J. Botia, E. Serrano, and J. Pavon, Testing and
Debugging of MAS Interactions with INGENIAS, Agent-Oriented
Software Engineering IX, Springer, Berlin, pp. 199-212, 2009.
[17] B. Henderson-Sellers, and P. Giorgini, Agent-Oriented
Methodologies, Proceedings of the 4th International Workshop on
Software Engineering for Large-Scale Multi-Agent Systems -
SELMAS'05, Idea Group Incorporation, 2005.
[18] Z. Houhamdi, Test Suite Generation Process for Agent Testing,
Indian Journal of Computer Science and Engineering, Vol. 2, 2, 2011.
[19] Z. Houhamdi, and B. Athamena, Structured Integration Test suite
Generation Process for Multi-Agent System, Journal of Computer
Science, Vol. 7, 5, 2011.
[20] Z. Houhamdi, and B. Athamena, Structured System Test Suite
Generation Process for Multi-Agent System, International Journal on
Computer Science and Engineering, Vol.3, 4, pp.1681-1688, 2011.
[21] M. Huget, and Y. Demazeau, Evaluating multi agent systems: a
record/replay approach, Intelligent Agent Technology, IAT 2004,
Proceedings IEEE/WIC/ACM International Conference, pp. 536 539,
2004.
[22] I. Sommerville, Software Engineering, 9th edition, Addison Wesley,
2011.
[23] N.R. Jennings, An Agent-Based Approach for Building Complex
Software Systems, Communications of the ACM, Vol. 44, 4, pp. 35-
41, 2001.
[24] H. Knublauch, Extreme programming of multi-agent systems,
International Joint Conference on Autonomous Agent and Multi-Agent
Systems, Bologna. ACM Press, pp. 704711, 2002.
[25] J. Krupansky, What is a software Agent?, Advancing the Science of
Software Agent Technology, 2008. http://agtivity.com/agdef.htm.
[26] D. Lam, and K. Barber, Debugging Agent Behavior in an
Implemented Agent System, 2nd International Workshop, ProMAS,
Springer, Berlin, pp. 104-125, 2005.
[27] P. McMinn, and M. Holcombe, The state problem for evolutionary
testing, Proceedings of the International Conference on Genetic and
Evolutionary Computation, Springer, Berlin, pp. 2488-2498, 2003.
[28] G. Myers, The Art of Software Testing, Wiley, 2nd Edition New
Jersey, John Wiley & Sons, 2004.
[29] C. Nguyen, A. Perini, and P. Tonella, Goal-oriented testing for MAS,
Agent-Oriented Software Engineering VIII, Lecture Notes in Computer
Science, Volume 4951, pp. 58-72, 2008.
[30] M. Nunez, I. Rodriguez, and F. Rubio, Specification and testing of
autonomous agents in e-commerce systems, Software Testing,
Verification and Reliability, Vol. 15, 4, pp. 211-233, 2005.
[31] L. Padgham, and M. Winikoff, Developing Intelligent Agent Systems:
A Practical Guide, John Wiley and Sons, 2004.
[32] L. Padgham, M. Winikoff, and D. Poutakidis, Adding debugging
support to the Prometheus methodology, Engineering Applications of
Artificial Intelligence, Vol. 18, 2, pp. 173-190, 2005.
[33] J. Pavon, J. Gomez-Sanz, and R. Fuentes-Fernandez, The INGENIAS
Methodology and Tools, In Agent Oriented Methodologies (eds.
Henderson-Sellers and Giorgini), Idea group, pp. 236-276, 2005.
[34] A. Perini, Agent-Oriented Software, Wiley Encyclopedia of
Computer Science and Engineering, John Wiley and Sons, Chapter 1,
pp. 1-11, 2008.
[35] R.S. Pressman, Engenharia de Software, 6
th
edition, Rio de Janeiro,
McGraw-Hill, 2002.
[36] L. Rodrigues, G. Carvalho, P. Barros, and C. Lucena, Towards an
integration test architecture for open MAS, 1st Workshop on Software
Engineering for Agent-Oriented Systems/SBES. pp. 60-66. 2005.
[37] C. Rouff, A test agent for testing agents and their communities,
Aerospace Conference Proceedings IEEE, Vol. 5. pp. 5-2638, 2002.
[38] J. Sudeikat, and W. Renz, A systemic approach to the validation of
self-organizing dynamics within MAS, Proceeding of the 9th
International Workshop on Agent-Oriented Software Engineering, pp.
237-248, 2008.
[39] TILAB. Java agent development framework. http://jade.tilab.com/.
[40] A. Tiryaki, S. Oztuna, O. Dikenelli, and R. Erdur, Sunit: A unit testing
framework for test driven development of multi-agent systems,
AOSE'06 Proceedings of the 7
th
International Workshop on Agent-
Oriented Software Engineering VII, Springer, Berlin, pp. 156-173,
2007.
[41] J. Wegener, Stochastic Algorithms: Foundations and Applications, In
Evolutionary Testing Techniques, Springer Berlin, Heidelberg, Chapter
9, pp. 82-94, 2005.
[42] Z. Zhang, J. Thangarajah, and L. Padgham, Automated unit testing for
agent systems, 2nd International Working Conference on Evaluation of
Novel Approaches to Software Engineering, ENASE'07, Spain, pp. 10-
18, 2007.
AUTHORS PROFILE
Dr. Zina Houhamdi received the M.Sc. and PhD. degrees in Software
Engineering from Annaba University in 1996 and 2004, respectively. She is
currently an Associate Professor at the department of Software Engineering,
Al-Zaytoonah University of Jordan. Her research interest includes Agent
Oriented Software Engineering, Software Reuse, Software Testing, Goal
Oriented Methodology, Software Modeling and Analysis, Formal Methods.
She has published several Research Papers in referred National/ International
journals. She has referred presentations in National / International
Conferences and Seminars.

Vol. 2, No. 6, 2011
142 | P a g e
Video Compression by Memetic Algorithm

Pooja Nagpal
M.Tech Student,CE deptt.
Yadavindra College of Engineering
Guru Khashi Campus,Talwandi Sabo.

Seema Baghla
Assistant Professor, CE deptt.
Yadavindra College of Engineering
Guru Khashi Campus, Talwandi Sabo

AbstractMemetic Algorithm by hybridization of Standard
Particle Swarm Optimization and Global Local Best Particle
Swarm Optimization is proposed in this paper. This technique is
used to reduce number of computations of video compression by
maintaining same or better quality of video. In the proposed
technique, the position equation of Standard Particle Swarm
Optimization is modified and used as step size equation to find
best matching block in current frame. To achieve adaptive step
size, time varying inertia weight is used instead of constant
inertia weight for getting true motion vector dynamically. The
time varying inertia weight is based up on previous motion
vectors. The step size equation is used to predict best matching
macro block in the reference frame with respect to macro block
in the current frame for which motion vector is found. The result
of proposed technique is compared with existing block matching
algorithms. The performance of Memetic Algorithm is good as
compared to existing algorithms in terms number of
computations and accuracy.
Keywords- Memetic Algorithm (MA); Standard Particle Swarm
Optimization (PSO); Global Local Best Particle Swarm
Optimization (GLBest PSO); Video Compression; Motion Vectors;
Number of Computations; Peak Signal to Noise ratio (PSNR).
I. INTRODUCTION
With the increasing popularity of technologies such as
Internet streaming video and video conferencing, video
compression has became an essential component of broadcast
and entertainment media. Motion Estimation (ME) and
compensation techniques, which can eliminate temporal
redundancy between adjacent frames effectively, have been
widely applied to popular video compression coding standards
such as MPEG-2, MPEG-4.
Motion estimation has been popularly used in video signal
processing, which is a fundamental component of video
compression. In motion estimation, computational complexity
varies from 70 percent to 90 percent for all video compression.
The exhaustive search (ES) or full search algorithm gives the
highest peak signal to noise ratio amongst any block-matching
algorithm but requires more computational time [1]. To reduce
the computational time of exhaustive search method, many
other methods are proposed i.e. Simple and Efficient Search
(SES)[1], Three Step Search (TSS)[2], New Three Step Search
(NTSS)[2], Four step Search (4SS)[3], Diamond Search
(DS)[4], Adaptive Road Pattern Search (ARPS)[5], Novel
Cross Diamond search [6], New Cross-Diamond Search
algorithm [7], Adaptive Block Matching Algorithm [8],
Efficient Block Matching Motion Estimation [9], Content
Adaptive Video Compression [10] and Fast motion estimation
algorithm [11] . Soft computing tool such as Genetic
Algorithm (GA) has also been used for fast motion estimation
[12].
Traditional fast block matching algorithms are easily
trapped into the local minima resulting in degradation on
video quality to some extent after decoding. Since
evolutionary computing techniques are suitable for achieving
global optimal solution. In this paper, we propose a Memetic
Algorithm to reduce number of computations of video
compression by maintaining same or better quality of video.
The paper is divided into five sections. The review of Particle
Swarm Optimization techniques is discussed in section 2. The
proposed Memetic Algorithm for video compression is
discussed in section 3. Section 4 provides experimental results
comparing MA with other methods for video compression.
The conclusion is given in section 5.
II. PARTICLE SWARM OPTIMIZATION
The Particle Swarm Optimizer (PSO) is a population-based
optimization method developed by Eberhart and Kennedy in
1995[13]. PSO is inspired by social behavior of bird flocking
or fish schooling. In PSO, a particle is defined as a moving
point in hyperspace. It follows the optimization process by
means of local best (Lbest), global best (Gbest), particle
displacement or position and particle velocity. In PSO, particle
changes their positions by flying around in a multi-
dimensional search space until computational limitations are
exceeded. The two updating fundamental equations in a PSO
are velocity and position equations. The particle velocity is
expressed as Eq (1) and the particle position is expressed as
Eq. (2)
) ( ) ( *
2 2 1 1 1
i
k
i
k
i
k
i
K
S Gbest r C S Lbest r C V W V + + =
+
(1)
i
k
i
k
i
k
V S S
1 1 + +
+ =
(2)
Where, V= Particle Velocity
S= Particle Position
Lbest = Local best
Gbest = Global best
W = Inertia weight
C
1
and C
2
are acceleration constant
r
1
and r
2
are random values [0 1]
k = Current iteration
Vol. 2, No. 6, 2011
143 | P a g e
i = Particle number
In first parts, W plays the role of balancing the global
search and local search. Second and third parts contribute to
the change of the velocity. The second part of Eq. (1) is the
cognition part, which represents the personal thinking of the
particle itself. The third part of Eq (1) is social part, which
represents the collaboration among the particles. Without the
first part of Eq. (1), all the particles will tend to move toward
the same position. By adding the first part, the particle has a
tendency to expand the search space, that is, they have ability
to explore new area. Therefore, they acquire a global search
capability by adding the first parts.
In GLBestPSO [14], inertia weight (w) and acceleration
co-efficient (c) are proposed in terms of global best and local
best position of the particles as given in Eq. (3) and (4). The
modified velocity equation for the GLBest PSO is given in Eq.
(5).
GLBest PSO is given in Eq. (5).
|
|
.
|
\
|
=
i
i
i
pbest
gbest
w 1 . 1
(3)
|
|
.
|
\
|
+ =
i
i
i
pbest
gbest
c 1
(4)
)) ( 2 ( ) ( ) 1 ( ) ( t x gbest pbest t r c t v w t v
i i i i i i
+ - - + - = (5)
III. MEMETIC ALGORITHM FOR VIDEO COMPRESSION
Memetic Algorithm is developed by hybridization of
Standard Particle Swarm Optimization and Global Local Best
Particle Swarm Optimization. This technique is used to reduce
number of computations of video compression by maintaining
the same or better quality of video.
We have modified velocity and position equations of PSO
to achieve step size for video compression, which is used to
predict best matching macro block in the reference frame with
respect to macro block in the current frame for which motion
vector is found.
To get the step size, the velocity and position equations of
PSO are modified as given below. The velocity equation is
expressed as Eq. (6).
V (t) =W*C*r (6)

Where W is the inertia weight, C is the acceleration
constant, r is random number between 0 to 1 and t is
generation number. To get the adaptive step size, the time
varying inertia weight (W) is used instead of constant inertia
weight similar to GLBestPSO for getting the true motion
vector dynamically. The time varying inertia weight is based
up on previous motion vectors as given in Eq. (7)
W= (1.1-Gbest+Pbest) (7)

Gbest=X+Y
Pbest= X-Y

Where, X and Y is the x and y coordinates of the predicted
motion vector. The velocity term in Eq. (6) is added with
previous motion vector to predict the next best matching block
as given in Eq (8)
S (t+1) =S (t) +V (t) (8)

In Memetic Algorithm, a search is made in an earlier frame
of the sequence over a random area of the frame. The search is
for the best matching block viz. the position that minimizes a
distortion measured between the two sets of pixels comprising
the blocks. The relative displacement between the two blocks
is taken to be the motion vector. Usually the macro block is
taken as a square of side consists of 16 pixels. The
compression ration is 128:1 or 256:2. The each block size of
16 x 16 is compressed into two pixels which are nothing but
motion vectors.
In Memetic Algorithm, five swarms are used to find best
matching block. The initial position of block to be searched in
reference frame is the predicted motion vector as expressed in
Eq. (8). In Memetic Algorithm, the number of generations is
taken as 2. The cost required for finding best matching block
in the reference frame is ten blocks, which is less than existing
methods.
The mean absolute difference (MAD) is taken as objective
function or cost function in Memetic Algorithm and is
expressed as in Eq. (9).
] ) , ( Re ) , ( [
1
1 1

= =
=
N
Q
M
P
Q P ck ferenceBlo Q P ck CurrentBlo
MN
MAD
(9)
Where, M = Number of rows in the frame and N = Number
of columns in the frame. The objective quality obtained by
Memetic Algorithm has been measured by the peak signal-to-
noise ratio (PSNR), which is commonly used in the objective
quality comparison. The performance of the proposed method
is evaluated by following Eq (10)

(10)
A further small improvement in the Memetic Algorithm is
to check for Zero Motion Prejudgment (ZMP). If current
macroblock matches with macroblock in the reference frame
i.e. cost is zero then motion vector are directly stored as zero
motion vector instead of gaining the motion vector through
Memetic Algorithm. The zero motion prejudgment saves
considerable amount of computational time.
Zero Motion Prejudgment is the procedure to find the
static macro blocks which contains zero motion. In real world
video sequences more than 70% of the MBs are static which
do not need the remaining search. So, significant reduction of
computation is possible if we predict the static macro blocks
by ZMP procedure before starting motion estimation
procedure and the remaining search will be faster and saves
memory. We first calculate the matching error (MAD)
between the macro block in the current frame and the macro
block at the same location in the reference frame and then
2
10
,
2
, 1
255
10log
1
( ( , ) ( , ))
M N
P Q
PSNR
OrigionalFrame P Q CompensatedFrame P Q
MN
=
=

Vol. 2, No. 6, 2011
144 | P a g e
compare it to a predetermined threshold. If the matching error
is smaller than predetermined threshold we consider this
macro block static which do not need any further motion
estimation, and return a [0, 0] as its motion vector (MV).
IV. EXPERIMENTAL RESULTS AND DISCUSSIONS
The presented method Memetic Algorithm has been tested
for two videos sequences. The performance of the proposed
method is measured in terms of the PSNR and percentage of
saving number of computations of video compression. To test
the efficiency of the proposed algorithm with existing
methods, the algorithms are executed in single machine. Video
sequence with distance of two frames between current frame
and reference frame are used to generate frame-by-frame
results of the Memetic Algorithm. The performance of
Memetic Algorithm is compared with other existing methods
such as ES, SESTSS, TSS, NTSS and 4SS and the results are
presented in Table 1 and Table 2. Figure 1 to Figure 2 shows
the comparison of PSNR with other existing methods for two
video sequences respectively. Figure 3 to Figure 4 shows the
comparison of number of computations with other existing
methods for two video sequences respectively. The speed of
Memetic Algorithm is found to be faster than that of already
published methods and PSNR is close to published methods as
shown in Table 3 and Table 4.
The Memetic Algorithm saves number of computations up
to 95.36% to 94.95 % with PSNR degradation of -0.1222 to -
0.0537 as compared to ES. Similarly, Memetic Algorithm
saves number of computations up to 58.95% to 35.73% with
PSNR gain of +1.08 to +0.03as compared to SESTSS, TSS,
NTSS and 4SS.
TABLE I. COMPARISON OF PSNR
Sequence ES SESTSS TSS NTSS 4SS Memetic
Video 1
32.3762 31.1686 31.4379 32.2046 31.8304 32.2540
Video 2
37.2776 37.1786 37.2081 37.1939 37.1480 37.2239

TABLE II. COMPARISON OF NUMBER OF COMPUTATIONS
Sequence ES SESTSS TSS NTSS 4SS Memetic
Video 1
207.4109 23.6349 16.2827 19.7889 18.7977 10.4642
Video 2
209.8687 23.7264 16.8776 18.7248 17.3848 9.7377
TABLE III. PSNR GAIN BY MEMETIC ALGORITHM OVER EXISTING
METHODS
TABLE IV. NUMBER OF COMPUTATIONS SAVING BY MEMETIC
ALGORITHM OVER EXISTING METHODS
Sequence ES SESTSS TSS NTSS 4SS
Video 1 94.95485 55.72564 35.73425 47.12086 44.33255
Video 2 95.3601 58.95838 42.304 47.99571 43.98728

Figure1. Comparison of PSNR for Video 1

Figure2. Comparison of PSNR for Video 2

Figure3. Comparison for Number of Computations for Video 1

5 10 15 20 25
29.5
30
30.5
31
31.5
32
32.5
33
33.5
34
34.5
Frame Number
P
S
N
R

i
n

d
b
Frame Number Vs PSNR
ES
SESTSS
TSS
NTSS
4SS
MA
5 10 15 20
20
22
24
26
28
30
32
34
36
38
40
Frame Number
P
S
N
R

i
n

d
b
Frame Number Vs PSNR
ES
SESTSS
TSS
NTSS
4SS
MA
5 10 15 20 25
0
50
100
150
200
250
Frame Number
N
u
m
b
e
r

o
f

C
o
m
p
u
t
a
t
i
o
n
s
Frame Number Vs No. of Computations
ES
SESTSS
TSS
NTSS
4SS
MA
5 10 15 20
-50
0
50
100
150
200
250
Frame Number
N
u
m
b
e
r

o
f

C
o
m
p
u
t
a
t
i
o
n
s
Frame Number Vs No. of Computations
ES
SESTSS
TSS
NTSS
4SS
MA
Sequence ES SESTSS TSS NTSS 4SS
Video 1 -0.1222 1.0854 0.8161 0.0494 0.4236
Video 2 -0.0537 0.0453 0.0158 0.03 0.0759
Vol. 2, No. 6, 2011
145 | P a g e
Figure 4 Comparison for Number of Computations for Video 2

V. CONCLUSION
The soft computing tool such as Memetic Algorithm is
used for video compression. The performance of Memetic
Algorithm is good as compared to existing algorithms except
ES for video compression because it find best matching block
with less computational cost by maintaining same accuracy of
video. Memetic Algorithm is faster and accurate tool for video
compression. The results shows promising improvement in
terms of accuracy, while drastically reducing the number of
computations. The step size equation of MA predicts the best
matching block with less computational requirement. Memetic
Algorithm uses only two-steps for video compression. An
approximately more than 96% of computational time saving in
the motion estimation coding is achieved as compared to ES
algorithm. This saving comes with less degradation in the
PSNR. This technique can easily used as block matching
motion estimation algorithm or video compression algorithm.
REFERENCES
[1] Y. Wang, B. Li and T. Weise, Estimation of Distribution and
Differential Evolution Cooperation for Large Scale Economic Load
Dispatch Optimization of Power Systems, Information Sciences,
Elsevier, 2010
[2] Swagatam Das and Sudeshna Sil, Kernel-induced fuzzy clustering of
image pixels with an improved differential evolution algorithm,
Information Sciences, Elsevier, vol. 180, pp. 12371256, 2010.
[3] M. Senthil Arumugam, M. V. C. Rao and Alan W.C. Tan, A new novel
and effective particle swarm optimization like algorithm with
extrapolation technique, International Journal of Applied Soft
Computing, Elsevier, vol. 9, pp. 308-320, 2009
[4] Dan Simon, Biogeography-Based Optimization, IEEE Trans. on Evol.
Comput. vol. 12, no.6, pp. 712-713, 2008.
[5] Jiancong Luo, Ishfog Ahmed, Yong Fang Liang and Vishwanathan
Swaminathan, Motion Estimation for Content adaptive Video
Compression, IEEE Transactions on Circuits and Systems for video
Technology, vol. 18, no.7, July 2008, pp.900-909.
[6] Chun-Man Mak, Chi keung Fong, and Wai Khen Chan, Fast Motion
Estimation For H.264/AVC in Walsh Hadmard Domain, IEEE
Transactions on Circuits and Systems for Video Technology, vol. 18,
no.6, June 2008, pp. 735-745.
[7] M. Senthil Arumugam, M. V. C. Rao and Aarthi Chandramohan, A
New and Improved Version of Particle Swarm Optimization Algorithm
with Global-Local Best Parameters, Journal of Knowledge and
Information System (KAIS), Springer, vol. 16, no.3, pp. 324-350, 2008
[8] Srinivas Pasupuleti and Roberto Bhattiti, The Gregarious Particle
Swarm Optimizer (G-PSO), GECCO 2006, Seattle, Washington, USA,
July 8-12, 2006, pp. 92-98.
[9] Humaira Niasr, Tae-Sun Chol, An Adaptive Block Motion Estimation
Algorithm Based on Spatio Temporal Correlation, Digest of Technical
papers, International conference on consumer Electronics, Jan 7-11,
2006, pp.393-394.
[10] Viet-Anh Nguyen, Yap-peng Tan, Efficient Block Matching Motion
Estimation Based on Integral Frame Attributes, IEEE Transactions on
Circuits and Systems for Video Technology, vol.16, no. 2, March 2006,
pp.375-385.
[11] Ratnaweera, S. Halgamuge and H. Watson, Self Organizing
Hierarchical Particle Swarm Optimization with time varying
acceleration coefficients, IEEE Transactions on Evolutionary
Computation, vol. 8, pp. 240-255, 2004.
[12] C.W. Lam., L.M. Po and C.H. Cheung, A New Cross- Diamond Search
Algorithm foe Fast Block Matching Motion Estimation, 2003 IEEE
International Conference on Neural Networks and Signal Processing,
Nanjing, China, Dec. 2003, pp.1262-1265.
[13] Yao Nie, Kai-Khuang Ma, Adaptive Rood Pattern Search for Fast
Block-Matching Motion Estimation, IEEE Trans. Image Processing,
vol.11, no.12,December 2002, pp. 1442-1448.
[14] Chun-Ho Cheung, Lai-Man Po, A Novel Cross Diamond Search
Algorithm for Fast Block Estimation, IEEE Trans. Circuits and
Systems, vol.11, no.12, December 2002, pp. 1442-1448.
[15] Shan Zhu, Kai-Khuang Ma, A New Diamond Search Algorithm for
Fast Matching Motion Estimation, IEEE Trans. Image Processing,
vol.9, no.2, February 2000, pp.287-290.
[16] Shen Li., Weipu Xu, Nanning Zheng, Hui Wang , A Novel Fast Motion
Estimation Method Based on Genetic Algorithm, ACTA
ELECTRONICA SINICA, vol.28 no.6, June 2000, pp.114-117.
[17] Jianhua Lu., Ming L. Liou, A Simple and Efficient Search Algorithm
for Block Matching Motion Estimation, IEEE Trans. Circuits and
Systems for Video Technology, vol.7, no.2, April 1997, pp. 429- -433.
[18] Lai-Man Po., Wing Chung Ma, A Novel Four- Step Search Algorithm
for Fast Block Motion Estimation, IEEE Trans. Circuits and Systems
for Video Technology, vol.6, no. 3, June 1996, pp. 313-317 .
[19] J. Kennedy and R. Eberhart, Particle Swarm Optimization, IEEE
International Conference on Neural Networks, Perth, Australia. vol. 4,
Dec. 1995, pp. 1942-1948.
[20] Renxiang Li, Bing Zeng, and Ming L. Liou, A New Three- Step
Search Algorithm for Block Motion Estimation, IEEE Trans. Circuits
and Systems for Video Technology, vol. 4, no. 4, August 1994, pp. 438-
442.

AUTHORS PROFILE
Pooja Nagpal received diploma in Computer Engineering in 2004 from
Technical Board,Chandigarh. B. Tech degree in Computer Science and
Engineering in 2007 under Punjab Technical University,Jalandhar.Since
2008 she is pursuing her M.Tech from Yadavindra College of
Engineering,Guru Khashi Campus,Talwandi Sabo,Bathinda(Punjab) Her
interests include, Image Processing and Swarm Intelligence .She has
contributed near about 8 technical papers in various national and
international conferences. She is a life member of ISTE.
Seema completed M.Tech. in Computer Engineering from punjabi
University Patiala in year 2007and B.Tech. in Computer Science &
Engineering from Punjab Technical University, Jalandhar in year 2004 and
holds NINTH merit position in University. She is presently working as
Assistant Professor in Computer Engineering at Yadavindra College of
Engineering, Punjabi University Guru Kashi Campus, Talwandi Sabo
(Bathinda). Her research areas include Genetic Algorithms, Optimization,
Image processing, motion estimation, Encryption etc.

Vol. 2, No. 6, 2011
146 | P a g e
Impact of Cloud Computing on ERP implementations
in Higher Education
Ms. Shivani Goel, Dr Ravi Kiran , Dr Deepak Garg
Thapar University, Patiala

Abstract Penetration of Higher Education in all regions is
increasing all over the globe at a very fast pace. With the increase
in the number of institutions offering higher education, ERP
implementations has become one of the key ingredient to achieve
competitiveness in the market. Many researchers have given their
inputs specifying the different nature of ERP implementation in
Educational institutions then the corporate organizations.
Recently Cloud computing has become a buzzword and it is
having applications in many domains. Researchers have already
started applying cloud computing in ERP implementations of
Higher education. This paper gives an insight into the nature of
cloud computing impact on ERP implementations and discusses
various issues related to this. Paper comes up with guidelines
regarding the use of cloud computing technology in the ERP
implementations of Higher Technical Education institutions.
Keywords- Cloud Computing; ERP; Higher Technical Education;
ERP implementation.
I. INTRODUCTION
Cloud Computing has got different meaning to people
working in different areas of computer science. Basically it is
all about how the services are managed and delivered.
Definitely it works on the principle of economies of scale at
application, software and hardware level. It results
in many benefits like service provisioning, reduced costs,
optimum resource utilization. Customer is now not tense about
the robustness and continuity of software services in case of
increasing demand of services embedded in cloud computing
applications. Cloud handles elastic demand and the scalability
limits go to infinity.
Good news is that good ERP systems should also have the
characteristics fault tolerance, data sharing, privacy in
communications and data handling, sharing data in diverse
application modules.
Going a little deeper into both it will definitely strike in our
mind that cloud computing technology can meet several of the
advance and futuristic requirements of ERP implementations in
organizations of higher technical education and can cater to
increasing volume and range of services.[1]
Virtualization: It is related to autonomic computing in the
context of utilizing the hardware or software resources
including the processing power of a machine. There are
different types of virtualizations like Hardware virtualization,
Desktop virtualization, OS-level virtualization, Network
virtualization and application virtualization etc. The similarity
in all these is that virtualization is all about saving costs and
running parallel virtual tasks from a single source increasing
the utility of that source.
Service Oriented Architecture: In this kind of setting
customer is the king. Customer is not aware of the
heterogeneity of the different hardware architectures in the
back of service providers. Customer is also not concerned with
the different software applications running in different tiers of
the web environment. SOA uses loose coupling concept and
interfaces are defined in terms of protocols.[2]
SaaS (Software as a service) : In our context SaaS means
that when the infrastructure need not to be developed in the
organization but there is a service provider managing all the
infrastructure at some remote location and providing the
services in real time through a thin client to the users of the
organization.
Ubiquitous computing: With this traditional desktop model has
been left behind and users can now access the services through
the device of their choice. It can be a mobile phone, ipad or any
other thin client. The users are many times not even aware of
this that they are using ubiquitous computing. Young
generation now has grown and has left behind the desktop
era.[3]
II. EVOLUTION OF ERP IMPLEMENTATIONS IN HIGHER
EDUCATION INSTITUTIONS
A. Traditional ERP implementation
Still more than half of the institutions in the developing
countries are following traditional approach due to different
reasons. In these kinds of implementations different modules
like admissions, fee collection, attendance, grading, Feedback
system, Billing, Smart classes, Inventory, Human resource
management etc are implemented in an Adhoc manner without
having an overall objective of a comprehensive ERP. Many of
the modules are generally developed at different times and
these modules dont talk to each other. There is no concept of
service architecture being used in these kinds of
implementations. Cloud computing theory is out of scope for
these kinds of implementations.
B. On-Campus implementation
Most of the implementations in the 90s stick to this
category, where organisations have developed in-house
infrastructure for ERP implementations and dont have any
outside dependence.
C. Implementations with Internet Host Provider
In these kinds of implementations virtualization, Server
side scripts or tools, Storage and networking component is
looked after by the host who has been hired for providing the
Vol. 2, No. 6, 2011
147 | P a g e
services. It leads to lower cost of hardware by the organization.
It also dilutes the requirement of highly skilled professions to

Fig 1.1 Cloud Components
maintain the system. Databases and applications are still run
and managed at the campus. These implementations take some
advantages of ubiquitous computing and SoA but bottlenecks
still remain.[4]
D. Cloud implementations
Apart from the applications part everything is managed by the
cloud server provider. Applications are anyway to be installed
or made available at the user end. Applications can run in
different variety of devices like TV, mobile phones, web
kiosks, ipads, laptops etc. Databases, Servers, networking and
storage components are still handled by the cloud provider.
III. BENEFITS OF CLOUD COMPUTING IN ERP
IMPLEMENTATIONS AT HIGHER TECHNICAL EDUCATION
INSTITUTIONS[5-8]
Reduced cost: Whole of the hardware is now the
responsibility of the cloud service provider. There is no
requirement of any hardware at the campus,
whatsoever. Even the cost of software decreases. You
dont need to buy the fix time licenses or user based
licenses. Now you only need to pay based on the usage
of a particular service.
Unfettered Access: All the faculty members, students,
parents, management, recruiters, vendors can access
the system according to the privileges granted to them.
The access can be through any wired or wireless
protocols using multiple devices of their choice.
Security: Standard encryption and decryption
techniques will be used and there is no need to worry
about the security of the applications.
Uptime: Almost zero downtime can be expected.
However it may also depend upon the trust factor of
your service provider and the feedback from the
existing customers.
Manpower: No manpower is required to be recruited at
the institution.
Futuristic Needs: Scalability, extension, peak load
performance work becomes very simple. Generally this
kind of support is available with every cloud provider.
Customization: Any institution is not bound to use all
the services being provided in the cloud application.
The customer can choose the services based on specific
requirements and the budgetary allocations.
Group organizations: It is even more cost effective for
institutions which have multiple courses and multiple
campuses at different places. The access is all in real
time through the web through any explorer like safari,
Mozilla or IE.
Integration : It can also integrate with the biometric,
fingerprint, swipe and other machines already working
in your organisations.
There is no additional risk in case of any emergency or
natural disasters because of third party
involvement.The abbreviation i.e. means that is,
and the abbreviation e.g. means for example.
IV. CHALLENGES IN IMPLEMENTING CLOUD COMPUTING
FOR ERP IMPLEMENTATION IN HIGHER TECHNICAL
EDUCATION[9-12]
Elasticity Complexity: The demands and services in
the cloud computing environment are elastic and
integrated with the inherent distributed and parallel
architectural characteristics. It involves highly skilled
technical labour to develop and maintain such
applications.
Superstructure Emergence: Cloud is emerging as a
superstructure and bypassing the boundaries of
functional and vertical domains. Cloud server hosting
different domain applications may become an issue
because of overlapping demand and supply
requirements.
Technological Bottlenecks : Emergence of clouds also
require corresponding technology upgradation in terms
of new data structures for handling massive and
dynamic data, new file systems, storage technologies
etc.
Serializability and consistency: Clouds are servicing
the users in 24-hr format from different regions. We
need to address the issue of trade-off between the
consistency and serializability in sophisticated
applications.
Applications
Storage and
Networking
Databases
Server
Vol. 2, No. 6, 2011
148 | P a g e
Programming model: New Programming paradigms
are need of the hour which offers something between
virtualization and physical reality.
Monitoring, Analysis and Building Trust: Due to the
scale of the applications running and the criticality of
the services constant monitoring of the cloud servers
for different parameters of operation is required.
Regular reporting after proper analysis is required,
which will in turn help in building the trust of the users
in cloud systems.
Mobility and provisioning: Most of the users are now
accessing the services through the devices which are
fast moving. So it requires the integration of cutting
edge mobile routing technologies with the cloud
services technologies.
V. CONCLUSION
It has been established that cloud computing will be helpful
in improving the cost, maintenance and technical efficiency of
ERP implementations in technical education institutions.
However, it is seen that apart from the benefits there are many
issues which need to be dealt with. The issues can be overcome
with proper planning. The association of right people at the
time the system is evolving is very crucial.
REFERENCES
[1] Brian Hayes. 2008. Cloud computing. Commun. ACM 51, 7 (July 2008),
9-11
[2] Mache Creeger. 2009. CTO Roundtable: Cloud Computing. Queue 7, 5,
Pages 1 (June 2009), 2 pages.
[3] Aaron Weiss. 2007. Computing in the clouds. netWorker 11, 4
(December 2007), 16-25
[4] Kemal A. Delic and Martin Anthony Walker. 2008. Emergence of the
Academic Computing Clouds. Ubiquity 2008, August, Article 1 (August
2008)
[5] Sanjay Ghemawat, Howard Gobioff, and Shun-Tak Leung. 2003. The
Google file system. In Proceedings of the nineteenth ACM symposium
on Operating systems principles (SOSP '03). ACM, New York, NY,
USA, 29-43.
[6] Roxana Geambasu, Cherie Cheung, Alexander Moshchuk, Steven D.
Gribble, and Henry M. Levy. 2008. Organizing and sharing distributed
personal web-service data. In Proceeding of the 17th international
conference on World Wide Web (WWW '08). ACM, New York, NY,
USA, 755-764.
[7] Michael Vrable, Stefan Savage, and Geoffrey M. Voelker. 2009.
Cumulus: Filesystem backup to the cloud. Trans. Storage 5, 4, Article 14
(December 2009), 28 pages.
[8] Michael Isard and Yuan Yu. 2009. Distributed data-parallel computing
using a high-level programming language. In Proceedings of the 35th
SIGMOD international conference on Management of data (SIGMOD
'09), Carsten Binnig and Benoit Dageville (Eds.). ACM, New York, NY,
USA, 987-994.
[9] Faouzi Kamoun. 2009. Virtualizing the Datacenter Without
Compromising Server Performance. Ubiquity 2009, August, pages
[10] Jeanna Matthews, Tal Garfinkel, Christofer Hoff, and Jeff Wheeler.
2009. Virtual machine contracts for datacenter and cloud computing
environments. In Proceedings of the 1st workshop on Automated control
for datacenters and clouds (ACDC '09). ACM, New York, NY, USA,
25-30
[11] Harold C. Lim, Shivnath Babu, Jeffrey S. Chase, and Sujay S. Parekh.
2009. Automated control in cloud computing: challenges and
opportunities. In Proceedings of the 1st workshop on Automated control
for datacenters and clouds (ACDC '09). ACM, New York, NY, USA,
13-18.
[12] Albert Greenberg, James Hamilton, David A. Maltz, and Parveen Patel.
2008. The cost of a cloud: research problems in data center networks.
SIGCOMM Comput. Commun. Rev. 39, 1 (December 2008), 68-73.

Vol. 2, No. 6, 2011
149 | P a g e
Algorithms for Content Distribution in Networks

Dr Deepak Garg, Senior Member, IEEE
Computer Science and Engineering Department
Thapar University, Patiala

Abstract In this paper an algorithm is presented which helps us
to optimize the performance of content distribution servers in a
network. If it is following the pay-as-you-use model then this
algorithm will result in significant cost reduction. At different
times the demand of different kind of content varies and based on
that number of servers who are serving that demand will vary.
Keywords- Content Distribution; Efficient Algorithms; Capacitated.
I. INTRODUCTION
The problem of content distribution in Networks is defined
in the context of demand and supply paradigm. There may be
downloadable software with a considerable size. At peak loads
10,000 persons may be downloading that software at the same
time. However, downloads will come to an average number of
15, 00 downloads at any given point of time. Whenever the
new versions of that software are launched, peak is again
reached. There can be a number of such software downloads
that you want to make available through your website. The
demands will also be coming from different regions and
different IP addresses. There will be certain pattern that can be
found out by the analysis. These demand number will also keep
on changing based on the time of the day in that region. The
distance of the servers from that client who is requesting the
software is crucial in determining the time taken to download
that software. Overall performance of the site in terms of
average time being taken to download one software is crucial to
the image and working of the company. This is a dynamic
problem, where new servers need to be made active once the
active server reaches a threshold value. Similarly once there is
a drop in demand from a particular region, than the server
servicing that region must be relieved from service to save
money. The input in this kind of problem is given in terms of a
matrix containing the cost of opening of each server location
and a set of locations generating the demand.[1-4]
The demands from the clients will come one by one in the
form of the http request and must be handled by our algorithm.
These client demands must be assigned to some server based
on the location of that client and the nearest server from that
position. However, it is not possible to open a new or passive
demand location for a few demands. In that case these demands
will be transferred to the nearest active server. When the new
demands crosses a particular threshold then only new service
points can be opened.
In Incremental Content Distribution Algorithm when an
existing server serves number of clients less than a defined
value then that server is stopped and existing demand services
are transferred to another content distribution servers.
II. CONTENT DISTRIBUTION SYSTEM ALGORITHM FOR
DYNAMIC AND INCREMENTAL CONTENT
Content Distribution Server problem can be characterized
as:
A universe, U, from which a set C of client input
positions is selected,
A distance metric, d: U U R
+,
defined over the
universe R
+
,
An integer, p 1, denoting the number of servers to
be located, and
An optimization function g that takes as input a set of
client positions and p server positions and returns a
function of their distances as measured by the metric
d.
III. APPROACHES FOR HANDLING LOCATION MODELS
Formulating an appropriate model is only one step in
analysing a location problem. Another much challenging task is
identifying the optimal solution. Attempting a solution with
well-known branch and bound optimization methods often
consume unacceptable computational resources. The reason
behind is that even the most basic location models are NP Hard
[6]. As a result, the location analyst must have to devise other
methods to identify optimal solution or at least near optimal
ones. Some of the most common approaches used by location
analysts are discussed below.
A. Greedy heuristic
A sequential approach that begins by evaluating each site
individually and selecting the one facility site that yields the
greatest impact on the objective. That facility site is then fixed
open. The location of the next facility is then identified by
enumerating all remaining possible locations and choosing the
site that provides the greatest improvement in the objective.
Each subsequent facility is located in an identical manner. The
method stops when the required number of facilities has been
sited.
B. Improvement heuristic
While greedy heuristics are effective at identifying a
feasible solution with modest computational effort, they cant
be relied upon to produce consistently good solutions. One of
the earliest improvements heuristic is neighborhood search
algorithm. In 1968, the most widely known improvement
method was introduced [7]. The basic idea is to move a facility
from the location it occupies in the current solution to an
unused site. Each unused location is tried in turn and when a
Vol. 2, No. 6, 2011
150 | P a g e
move produces a better objective function value, then that
relocation is accepted and we have a new (improved) solution.
The search procedure is repeated on the new solution and stops
when no better solution can be found via this method. A
variable neighborhood search algorithm was presented for
solving the p-median problem [8]. The algorithm performs an
intensive local search on the current solution until it settles in
local optima. It then repeats the process by randomly selecting
a solution from a neighborhood at a distance k from the current
best solution. The process continues, increments k, until some
predefined maximum value of k is attained.[9-11]
C. Lagrangian relaxation
When using any heuristic we are trading on savings in
solution time against the quality of the solution while the
heuristic often find good solutions to a variety o location
problems, it is difficult to evaluate the trade off since we have
no way of knowing how far from optimality those solutions
are. Without having the optimal value of the objective function
available for comparison, we can sometimes approximate the
difference between a heuristics solution and the optimal
solution by finding bounds. One of the primary attraction of the
technique known as Lagrangian relaxation is that it provides
upper and lower bound on the value of objective function [12].
This is done by eliminating i.e relaxing one or more of the
constraint of the original model and adding these constraints
multiplied by an associated Lagrange multiplier to the objective
function. The role of these multipliers is to derive the
Lagrangian problem towards a solution that satisfies the
relaxed constraints. The primary challenge in applying such
technique is in selecting which constraint to relax. Ideally the
relaxed problem ought to be solvable by inspection or by a
simple sorting the objective function coefficients.
Algorithm
Input : Cost of starting a new server and a queue of client
requests {cr1,cr2,,crn}
A set S is maintained which consists of currently active
servers to process that demand. The set will look like { ,
s1,s2,sn}. Sets of Client requests which are being handled by
individual servers will also be maintained.
Step 1: When a new client request for a download to start,
either it can be assigned to the existing active server or a new
server should be started.
Step 2. Let si is the server nearest to the client request and
number of client requests assigned to si< threshold value then
the download request is assigned to that server
Step 3: If si is the server nearest to the client request and
number of client requests assigned to si=threshold value then
the new server needs to be assigned. New server will be started
in the cloud of regions being represented by the incoming
request for download.
Step 4: Another thread will keep a check on the number of
requests that are being services by a server because some of
them will be completed at any given time.
Step 5: Once the number of requests are below a threshold
then it will check for the possibility of transferring the existing
download request of one server to another server keeping in
mind that the total load does no increases beyond the limits
after such transfer.
The Total cost will be calculated as
(Number of active servers) * cost of starting a server + Sum
of all client requests for download taken over all distances of
existing servers
The evaluation of this algorithm tells that it will give results
close to the optimum values that can be reached.
IV. RESULTS AND DISCUSSION
By the nature of the problem it can be seen that essentially
it an NP-complete problem in which the number of options can
be exponential. Because a particular request canbe assigned to
any of the active set of n servers giving rise to n different
options, similarly if there are m different requests that will
come to exponential possibilities. By the above algorithm, the
solution will be close to the actual solution and limited to 1.5
times the value of the exact solution.
V. CONCLUSION
Content management algorithms initially were only dealing
with static data. Now improved algorithms can also handle
dynamic nature of the data, locations, demands and resources.
The resultant saving is significant in terms of time and cost.
There is also a possibility of applying the concepts of linear
programming to further improve the approximation ratio.
REFERENCES
[1] N. Alon, B. Awerbuch, Y. Azar, N. Buchbinder, and J. Naor. A General
Approach to Online Network Optimization Problems. ACM
Transactions on Algorithms, 2(4):640 to 660, 2006.
[2] V. Arya, N. Garg, R. Khandekar, A. Meyerson, K. Munagala, and V.
Pandit. Local Search Heuristics for k-Median and Facility Location
Problems. SIAM J. on Computing, 33(3):544 to 562, 2004.
[3] A. Borodin and R. El-Yaniv. Online Computation and Competitive
Analysis. Cambridge University Press, 1998.
[4] M. Charikar and R. Panigrahy. Clustering to Minimize the Sum of
Cluster Diameters. In Proc. of the 33rd ACM Symposium on Theory of
Computing (STOC '01), pages 1 to 10, 2001.
[5] M. Chrobak and C. Kenyon-Mathieu. Online Algorithms Column 10:
Competitiveness via Doubling. SIGACT News, 37(4):115 to 126, 2006.
[6] J. Fakcharoenphol, S. Rao, and K. Talwar. Approximating Metric
Spaces by Tree Metrics. Encyclopedia of Algorithms, pages 51 to 53,
2008.
[7] D. Fotakis and C. Tzamos. Winner-Imposing Strategyproof Mechanisms
for Multiple Fa-cility Location Games. In Proc. of the 6th Workshop on
Internet and Network Economics (WINE '10), volume 6484 of LNCS,
pages 234 to 245, 2010.
[8] G. Frahling and C. Sohler. Coresets in Dynamic Geometric Data
Streams. In Proc. of the 37th ACM Symposium on Theory of Computing
(STOC '05), pages 209 to 217, 2005.
[9] P. Lu, X. Sun, Y.Wang, and Z.A. Zhu. Asymptotically Optimal
Strategy-Proof Mechanisms for Two-Facility Games. In Proc. of the
11th ACM Conference on Electronic Commerce (EC '10), pages 315 to
324, 2010.
[10] R.R. Mettu and C.G. Plaxton. The Online Median Problem. SIAM J. on
Computing,32(3):816 to 832, 2003.
[11] A. Meyerson. The Parking Permit Problem. In Proc. of the 46th IEEE
Symposium on Foundations of Computer Science (FOCS '05), pages
274-284, 2005.
Vol. 2, No. 6, 2011
151 | P a g e
[12] R.L. Francis (Editors) P.B. Mirchandani. Discrete Location Theory.
Willey, 1990.
[13] B. Krishnamurthy, C. Wills, Y.Zhang On the use and performance of
content distribution networks IMW01 poceedings of the 1
st
ACM
SIGCOMM Workshop on Internet Measurement,2001
[14] G.Pallis, A.Vakali Insight and perspectives for content delievery
networks , Communications of the ACM, 2006

AUTHORS PROFILE
Dr Deepak Garg, Senior Member of ACM and is Chair, ACM North India
SIGACT. He is Senior Member IEEE and Secretary IEEE Delhi Section
Computer Society. He has more than 50 publications to his credit. His
research interests are Algorithm design and complexity issues. He is
having 15 years of research experience

Vol. 2, No. 6, 2011
152 | P a g e
Generation of Attributes for Bangla Words for
Universal Networking Language(UNL)

Muhammad Firoz Mridha
Stamford University Bangladesh
Dhaka, Bangladesh
Manoj Banik
Dept. of Computer Science and Engineering
Ahsanullah University of Science and Technology,
Dhaka, Bangladesh
Kamruddin Md. Nur
Stamford University Bangladesh
Dhaka, Bangladesh
Mohammad Nurul Huda
Dept. of Computer Science and Engineering
United International University
Dhaka, Bangladesh

Abstract The usage of native language through Internet is
highly demanding now a day due to rapidly increase of Internet
based application in daily needs. It is important to read all
information in Bangla from the internet. Universal Networking
Language (UNL) addressed this issue in most of languages. It
helps to overcome the language barrier among people of different
nations to solve problems emerging from current globalization
trends and geopolitical interdependence. In this paper we
propose a work that aims to contribute with morphological
analysis of those Bangla words from which we obtain roots and
Primary suffixes and developing of grammatical attributes for
roots and Primary suffixes that can be used to prepare Bangla
word dictionary and Enconversion/Deconversion rules for
Natural Language Processing(NLP).
Keywords- Morphology; Bangla Roots; Primary Suffix;
Grammatical Attributes; Universal Networking Language.
I. INTRODUCTION
In the last few years, machine translation techniques have
been applied to web environments. The growing amount of
available multilingual information on the Internet and the
Internet users has led to a justifiable interest on this area.
Hundreds of millions of people of almost all levels of education
and attitudes, of different jobs all over the world use the
Internet different purposes [1]. English is the main language of
the Internet. Understandably, not all people know English.
Urgent need, therefore, arose to develop interlingua translation
programs. The main goal of the UNL system is to provide
Internet users access to multilingual websites using a common
representation. This will allow users to visualize websites in
their native languages. The UNL system has a growing
relevance since the usage of the WWW is generalized across
cultural and linguistic barriers. Many languages [10] such as
Arabic, French, Russian, Spanish, Italian, English, Chinese or
Brazilian Portuguese have already been included in the UNL
platform. Lexical knowledge representation is a critical issue in
natural language processing systems. Recently, the
development of large-scale lexica with specific formats capable
of being used by several different kinds of applications has
been given special focus; in particular to multilingual systems.
Our aim is to introduce Bangla into this system. In order to
implement this project with the lowest time and human effort
costs, we will reuse linguistic resources already available as
much as possible.
In this paper we present the UNL system for Bangla. The
major components of our research works touches upon i)
development of grammatical attributes for Bangla root and Krit
Prottoy to construct Bangla Word Dictionary and use of
morphological analysis ii) UNL Expression of the Bangla
attributes and iii) Selecting scheme of attributes. In section 2
we describe the UNL system. In sections 3, 4 and 5 we present
our main works that include all the above three components.
II. UNIVERSAL NETWORKING LANGUAGE
UNL is an artificial language that allows the processing of
information across linguistic barriers [10]. This artificial
language has been developed to convey linguistic expressions
of natural languages for machine translation purposes. Such
information is expressed in an unambiguous way through a
semantic network with hyper-nodes. Nodes (that represent
concepts) and arcs (that represent relations between concepts)
compose the network. UNL contains three main elements:
Universal Words: Nodes that represent word meaning.
Relation Labels: Tags that represent the relationship
between Universal Words. Tags are the arcs of UNL
hypergraph.
Attribute Labels: Additional information about the
universal words.
These elements are combined in order to establish a
hierarchical Knowledge Base (UNLKB) [10] that defines
unambiguously the semantics of UWs. The UNL Development
Set provides tools that enable the semi-automatic conversion of
natural language into UNL and vice-versa. Two of such tools
are the EnConverter and the DeConverter. The main role of
EnConverter [11] is to translate natural language sentences into
UNL expressions. This tool implements a language
Vol. 2, No. 6, 2011
153 | P a g e
independent parser that provides a framework for
morphological, syntactic and semantic analysis synchronously.
This allows morphological and syntactical ambiguities
resolution. The DeConverter [3, 12], on the other hand, is a
language independent generator that converts UNL expressions
to natural language sentences.
A. Universal Words
Universal Words are words that constitute the vocabulary
of UNL. A UW is not only a unit of the UNL syntactically and
semantically for expressing a concept, but also a basic element
for constructing a UNL expression of a sentence or a
compound concept. Such a UW is represented as a node in a
hypergraph. There are two classes of UWs from the viewpoint
in the composition:
labels defined to express unit concepts and called UWs
(Universal Words)
a compound structure of a set of binary relations grouped
together and called Compound UWs.
B. Relational Labels
The relation [1] between UWs is binary that have different
labels according to the different roles they play. A relation
label is represented as strings of three characters or less. There
are many factors to be considered in choosing an inventory of
relations. The following is an example of relation defined
according to the above principles.
Relation: There are 46 types of relations in UNL. For
example, agt (agent), agt defines a thing that initiates an action,
agt(do, thing), agt(action, thing), obj(thing with attributes) etc.
C. Attributes
The attributes represent the grammatical properties of the
words. Attributes of UWs are used to describe subjectivity of
sentences. They show what is said from the speakers point of
view: how the speaker views what is said. This includes
phenomena technically [4, 5] called speech, acts, propositional
attitudes, truth values, etc. Conceptual relations and UWs are
used to describe objectivity of sentences. Attributes of UWs
enrich this description with more information about how the
speaker views these state of affairs and his attitudes toward
them.
III. MORPHOLOGY OF BANGLA WORDS
Morphology is the field of linguistics that studies the
structure of words. It focuses on patterns of word formation
within and across languages, and attempts to formulate rules
that model the knowledge of the speakers of those languages.
Thus morphological analysis is found to be centered on
analysis and generation of word forms. It deals with the
internal structure of words and how words can be formed.
Morphology plays an important[2, 8] role in applications such
as spell checking, electronic dictionary interfacing and
information retrieving systems, where it is important that words
that are only morphological variants of each other are identified
and treated similarly. In natural language processing (NLP) and
machine translation (MT) systems we need to identify words in
texts in order to determine their syntactic and semantic
properties [7]. Morphological study comes here to help with
rules for analyzing the structure and formation of the words. A
Bangla morpheme, besides the root word, is supposed to be
represented in the Bangla-UNL dictionary using the following
UNL format [10].
[HW] UW (ATTRIBUTE 1, ATTRIBUTE 2 ) <FLG,
FRE, PRI>
HW Head Word (Bangla Word)
UW Universal Word
ATTRIBUTE Attribute of the HW
FLG Language Flag
FRE Frequency of Head Word
PRI Priority of Head Word

The attributes describe the nature of the head word
classifying it as a grammatical, semantic or morphological
feature. So, we will be especially concerned about
representation of morphemes using various attributes.
A. Prefixes ( )
Prefixes are the words that are used before words to express
various meanings of the same words. There are around fifty
(50) prefixes used in Bangla sentences. In Shangskrit Bangla
we use twenty (20) prefixes[2] say , ( ), (),
() etc. , in Bangla we use thirteen prefixes (13)
prefixes[2] such as (), (), ()
etc., five(5) foreign prefixes[5] such as (), () etc.,
four English prefixes[14] such as ( ), (),
(), () etc. These prefixes are used before words to
make thousands of meaningful Bangla Words.
In our work, we will make separately Word Dictionary
entries for all of these prefixes and words, so that they can
combinely make meaningful words by applying rules. For
example, if we consider prefix [9] (means
like/similar/every/opposite/against etc.) we can make
, etc. Now we can make the word
for dictionary entry. But the word has two or more
meanings so that we have to represent two or more dictionary
entries for the word as follows.
[]{} every (icl>thing) (ABSTRACT THING) <B,0,0>
[]{} opposite (icl>thing) (ABSTRACT THING)
<B,0,0>

Now if we want to represent the concepts of the words say
, etc., we need not represent the whole words.
We have to represent only the words , in the
dictionary entry as per the following format.
[]{}day (icl>period>time) (N, ABSTRACT THING,
LIGHT) <B, 0, 0>
[] {} sound (icl>occure>thing) (N, ABSTRACT THING)
<B, 0, 0>

Vol. 2, No. 6, 2011
154 | P a g e
B. Noun Morphology
Bangla Nouns have very strong and structural inflectional
morphology base on case. Case of noun may be nominative
(, boy), accusative (-), to the boy) and genetive
(-, of the boy) and so on. Gender and number are also
important for identifying proper categories of nouns. Number
may be singular (, boy or , the boy, , book,
, the book) plural (, boys , the boys
, the books etc.). So, from the word we get
, , , etc. and from the
word we get , etc. Some dictionary entries
may look like. []{} boy (icl>person) (N, HN, C,
ANI)<B,0,0>
Here, boy (icl>person) is the UW for but ,
etc. have no UWs. Therefore, they should be represented
in the dictionary only with grammatical attributes as follows.
[] {} (3P, SUF, N)<B,0,0>
[] {} (3P, SUF, N, HUMN, SG) <B,0,0>
[] {} (3P, PL, SUF, N, HUMN) <B,0,0>
[] {} (N, SG, SUF,3P) <B,0,0>
[] {} (N, PL, SUF,3P) (<B,0,0>
C. Adjective Morphology
As Adjective we can consider Bangla words ,
and meaning bravery, beautiful and good
in English respectively. From the first word we get
(+), (+). And from the second and third
words we get , , etc. We have to have the
dictionary entries for , , , , , , to make
the meaningful words , , , etc. by
combining the morphemes with the root words using analysis
rules.
D. Pronoun Morphology
Here we can consider the word root (he/she). From
this we get -, -, -, --, -
etc. So, we have to consider these morphemes , ,
, for dictionary entries to form words with as
above.
E. Verb Morphology
Diversity of verb morphology in Bangla is very significant.
We can select the head words as the Longest Common Lexical
Unit (LCLU) of all the possible transformations of the word
[8].
We can give the example of the Bangla word (means
read). The corresponding UW in basic form is read. The
dictionary entry is: [] { } read (icl>do), where v & is
the head word and (icl>do) is from the knowledge base. Some
possible transformations of " in the Bangla to UNL
dictionary are given as follows [9, 10]:

If we consider (means read) as a root, we can
represent this root in the dictionary as
[]{} read (icl>do) (V, @present) <B,0,0>

Some transformations based on the persons and tenses are.
For first person:
[] { } read (icl>do) (ROOT, BANJANT)<B, 0, 0>
[] { } read (icl>do) (ROOT, BANJANT, PRESENT
INDEF)<B, 0, 0>
CONT)<B,0, 0>

For second person:
CONT)<B, 0, 0>
[] { } read (icl>do) (ROOT, BANJANT, FUTURE
INDEF)<B, 0, 0>

For third person:
PERF)<B, 0, 0>
CONT)<B, 0, 0>

For resolving the ambiguities of the words ,
, , , , etc. we have to
define them as full words for dictionary entries. For instance
[] {} go (icl>move>do)(V, PAST, INDEF, 1P).
Using the same procedure we can make dictionary entries for
different transformations of other roots such as (do),
(write), (give) etc.
IV. MORPHOLOGY OF BANGLA ROOT WORDS
Bangla Language contains a lot of verbs. The core part of
those verbs is called roots. In another way if we split the verbs
we get two parts Roots and Suffixes. From verbs if we remove
suffixes we get roots. For example Ki (do) is a verb. Its two
parts are: Ki+G; here Ki is a root and G is a suffix. Some
other Bangla roots verbs are Pj& , co&, ai& , Mo&, Nl&,
bvP&, Kuv`& etc.
A. Bangla Primary Suffixes (K cZq)
We know that the core of the verb is called root and if
number of suffixes are added to roots then they form verbs.
When sound or sounds [8] are added with roots and form nouns
or adjectives then the root words are called root verbs and the
sound or sounds are added with root verbs are called Primary
Suffixes. For example Pj& (Root verb)+Ab& (Primary
Suffix)=Pjb (Noun) and Pj& (Root verb)+A (Primary
Suffix)= Pj (Adjective). Some others primary suffixes are
Ab, Abv, Awb, AK, Av etc.
Vol. 2, No. 6, 2011
155 | P a g e
B. Vowel ended and Consonant ended root
Verb roots that are ended with vowel are called SORANTO
root. For example, , , , , , , etc. And verb
roots that are ended with consonant are called BANJANTO
root. For example, ,, , , , .etc.
C. Morphological Analysis of Bangla verbs
Morphological analysis is applied to identify the actual
meaning of the word by identifying suffix or morpheme of that
word. Every word is derived from a root word. A root word
may have the different transformations. This happens because
of different morphemes which are added with it as suffixes. So,
the meaning of the word varies for its different transformations.
For example, if we consider Ki&(do) as a root word then after
adding B we get the word Kwi[6, 8] which means a work
done by some one(first person , present tense). Similarly after
adding Av, we get the word Kiv. Here, this word represents
noun of the root word Ki&. Therefore, by morphological
analysis we get the grammatical attributes of the main word.
Derivational morphology is simple and a word rarely uses the
derivational rule in more than two or three steps. The first step
forms nouns or adjectives from verb roots. The next steps form
new nouns and adjectives [5]. We have examined derivational
morphology for UNL Bangla dictionary too.
V. METHODS OF FINDING GRAMMATICAL ATTRIBUTES
Representing Universal Words (UWs) for each of the
Bangla Head Word we need to develop grammatical attributes
that describe how the words behave in a sentence. Grammatical
Attributes (GA) have to be developed by the rules
(Enconversion and Deconversion) and dictionary developers.
They play very important rules for writing Enconversion and
Deconversion rules because a rule uses GA in morphological
and syntactic analysis, to connect or analyze one morpheme
with another to build a meaningful (complete) word and to
examine or define the position of a word in a sentence.
When we analyze the HWs for representing them in the
word dictionary as UWs, we find all the possible specifications
of the HWs as attributes named grammatical attributes, so that
they can be used in the dictionary for making rules (EnCo and
DeCo). For example, if we consider cvwL meaning bird as a
head word, then we can use attributes N (as it is noun), ANI (as
bird is an animal), SG for singular number and CONCRETE
(as it a concrete thing which is touchable).
So, this word can be represented in the dictionary as
follows:
[cvwL]{}bird(icl>animal>animate thing)(N,ANI,SG,CONCRETE)

Head Universal Word Grammatical Attributes
Word

Same way we can represent the words avb (paddy), bvP&
(dance) etc. as follows:
[avb] {} paddy(icl>plant>thing)(N, PLANT, CONCRETE)
[bvP&]{} dance(icl>do)(ROOT, BANJANT)
Where, N for noun, CONCRETE for any concrete thing
which is touchable, ANI for animal and SG for singular
number.
As we are the initiators of developing rules and word
dictionary for Bangla we are proposing some grammatical
attributes and their descriptions in table 1.
A. Primary suffixes with consonant ended roots
(BANJANTO):
Some primary suffixes that are added with roots (which are
ended with consonant) to form nouns and adjectives are given
in table 2.
Table 2 Consonant ended roots with primary suffixes
Banjanto root + Primary suffix Noun and Adjective
+ ,
+ ,
+
+
+
+
+

The dictionary entries for the roots and their suffixes are as
follows:
[]{} do (icl>do)( ROOT, BANJANT)\
[]{} read (icl>do)( ROOT, BANJANT)
[]{} go (icl>do)( ROOT, BANJANT)
[]{} cut (icl>do)( ROOT, BANJANT)
[ ]{} attack (icl>do)( ROOT, BANJANT)
[ ]{} fly (icl>do)( ROOT, BANJANT)
[]{} (PROT, KPROT, BANJANT, MVNOUN)
[]{} (PROT, KPROT, BANJANT, MNOUN)
[]{} (PROT, KPROT, BANJANT, MADJ)

B. Primary suffix with vowel ended roots (SORANTO):
Some SORANTO is added only with primary suffix Av to
form nouns are shown in table 3.
Table 3 Vowel ended roots with primary suffix
Soranto root + Primary suffix Noun
+
+
+
+

So, the dictionary entries would be:
[]{} want(icl>do) ( ROOT, SORANT) <B,0,0>
Vol. 2, No. 6, 2011
156 | P a g e
[]{} get(icl>obtain) ( ROOT, SORANT) <B,0,0>
[]{} give(icl>do) ( ROOT, SORANT) <B,0,0>
[]{} eat(icl>do) ( ROOT, SORANT) <B,0,0>
[]{} (PROT, KPROT, SORANT, MNOUN) <B,0,0>

C. There are some such primary suffixes which have two
forms. One form is added with vowel ended roots and other
forms are added with consonant ended roots. Those are given
in table 4.
Table 4 Primary suffix with SORANTO roots and BANJANTO roots
Banjanto root + Primary suffix Noun
+
+
+
+
+
+

The UNL form:
[]{} want(icl>do) ( ROOT, SORANT) <B,0,0>
[]{} get(icl>obtain) ( ROOT, SORANT) <B,0,0>
[]{} give(icl>do) ( ROOT, SORANT) <B,0,0>
[]{}(PROT,KPROT,SORANT,MNOUN)<B,0,0>
[]{} read (icl>do)( ROOT, BANJANT) <B,0,0>
[]{} go (icl>do)( ROOT, BANJANT) <B,0,0>
[]{} cut (icl>do)( ROOT, BANJANT) <B,0,0>
[]{}(PROT,KPROT,BANJANT,MNOUN)<B,0,0>

All the Roots are added with Krit Prottoy and form a
meaningful word. For example,
+=
+=

But Some BANJANTA Roots ( ,) etc do not form a
meaningful Bangla word to add with Prottoy. For
example,
+=
+=.

Ratherin place of and in place of is added
with Krit Prottoy and form meaningful word. For
example, +=, +=.
Table 1 some proposed grammatical attributes
D. To solve this problem we divide BANJANTA Roots into
two categories. One is General BANJANTA that is attributed
as BANJNT and another is attributed with URoots.

Grammatical
Attributes
Descriptions Examples (Here we
can use Bangla Words)
ROOT Only for verb root co& (Read), Ki&(do)
etc.
SORANT Verb roots that are ended with
vowel
(want),(go) etc.
BANJANT Verb roots that are ended with
consonant
(read), ai& (catch)
etc.
1P First person vwg (I), Avgiv (we) etc.
2P Second Person Zzwg (you) etc.
3P Third Person m cyil (He), m
gwnjv (She) etc.
PROT For all suffixes Av, Ab, AvB etc.
KPROT For the suffixes that are used
after roots to create Nouns,
Adjectives etc.
BK, Ab etc.

KBIVOKTI

For the suffixes that are used
after roots to create only verbs.
B, BZwQ, e etc.
ADJ For any adjective fvj (good), my`i
(beautiful) etc.
N For any noun Kjg (pen), Avg(mango)
etc.
MNOUN suffixes that are added with
roots to make nouns.
Av etc.
MADJ suffixes that are added with
roots to make adjectives
A etc.
MVERB suffixes that are added with
roots to form verbs
B, BZwQ, e etc.
UROOT For consonant , etc.
PRESENT
INDEFINITE
suffixes that create present
indefinite form of the sentence
B etc.
HPRON For human , etc.
NUM For number ,, etc.
NANI For not amimate , etc.
CONCRETE For solidthing , etc.
HF For human food , etc.
FEMALE For female person () etc.
BIV For Bivokti ,
V For Verb , etc
3PG 3
rd
Person General , etc
IMPS Imperative Sentence , etc
3PR 3
rd
Person Respected , etc
2PR 2
nd
Person Respected , etc
2PG 2
nd
Person General , etc
2PN 2
nd
Person Neglected , etc
PAH Past Habituate , etc
1P 1
st
Person , etc
Vol. 2, No. 6, 2011
157 | P a g e
Table 5 Examples of general BANJANTO and URoots

The UNL form:
[]{} do (icl>do)( ROOT, BANJANT) <B,0,0>
[ ]{} lost (icl>do)( ROOT, BANJANT) <B,0,0>
[]{} (PROT, KPROT, SORANT, MNOUN) <B,0,0>
[]{} swing (icl>do)( ROOT, BANJANT) <B,0,0>
[]{} open (icl>do)( ROOT, BANJANT) <B,0,0>
[ ] {} swing(icl>do)( ROOT, BANJNT ,URoot) <B,0,0>
[] {} open(icl>do)( ROOT, BANJNT ,URoot ) <B,0,0>

We make two entries in Bangla word dictionary for
URoots. For example, and , and . In Bangla
there are some primary suffixes which added with roots and
form a new different words for example, + w = w and
+ w= w. These words are added to word dictionary in
special category.
The dictionaries would be:
[w]{} speech (icl>do)( ROOT, BANJANT, SP) <B,0,0>
[ w] {} free (icl>do)( ROOT, BANJANT, SP) <B,0,0>

The suffixes , , etc. will be in the dictionary only
with grammatical attributes. They will be added with the roots
to form verbs, nouns or adjectives using rules. In the above
examples we have classified suffixes in the basis of adding
either with SORANTO (vowel ended) or BANJANTO
(consonant ended) to give them proper attributes so that they
can be used to make appropriate rules enconversion and
deconversion.
VI. CONCLUSION AND FUTURE WORK
A system capable of understanding natural language
sentences is of potentially unlimited uses in the field of natural
language processing. In this paper we have generated
grammatical attributes of Bangla roots and Krit Prottoy for
developing Bangla Word Dictionary for Universal Networking
Language (UNL).
We have presented some method to select grammatical
attribute using morphological analysis Bangla words that can
be used to make dictionary for converting the Bangla sentences
to UNL documents and vice versa. We have done limited work
so far for Bangla words.
Our future plan is to build a Bangla language server that
will contain a complete Bangla Word Dictionary.
REFERENCES
[1] S. Abdel-Rahim, A.A. Libdeh, F. Sawalha, M. K. Odeh, Universal
Networking Language(UNL) a Means to Bridge the Digital Divide,
Computer Technology Training and Indistrial Studies Center, Royal
Scientific Sciety, March 2002.
[2] M. M. Asaduzzaman, M. M. Ali, Morphological Analysis of Bangla
Words for Automatic Machine Translation, International Conference on
Computer, and Information Technology (ICCIT), Dhaka, 2003, pp.271-
276
[3] Serrasset Gilles, Boitel Christian, (1999) UNL-French econversion as
Transfer & Generation from an Interlingua with Possible Quality
nhancement through Offline Human Interaction. Machine Translation
Summit-VII, Singapore.
[4] M. E. H. Choudhury, M. N.Y. Ali, M.Z.H. Sarkar, R. Ahsan, Bridging
Bangla to Universal Networking Language- A Human Language Neutral
Meta- Language, International Conference on Computer and
Information Technology (ICCIT), Dhaka, 2005,pp.104- 109
[5] M.E.H. Choudhury, M.N.Y. Ali, Framework for synthesis of Universal
Networking Language, East West University Journal, Vol. 1, No. 2,
2008, pp. 28-43
[6] M.N.Y. Ali, J.K. Das, S.M. Abdullah Al Mamun, M. E.H. Choudhury,
Specific Features of a Converter of Web Documents from Bengali to
Universal Networking Language, International Conference on
Computer and Communication Engineering 2008(ICCCE08), Kuala
Lumpur, Malaysia.pp. 726-731
[7] S. Dashgupta, N. Khan, D.S.H. Pavel, A.I. Sarkar, M. Khan,
Morphological Analysis of Inflecting Compound words in Bangla,
International Conference on Computer, and Communication Engineering
(ICCIT), Dhaka, 2005, pp. 110-117
[8] M.N.Y. Ali, J.K. Das, S.M. Abdullah Al Mamun, A. M.
Nurannabi,Morpholoical Analysis of Bangla worfs for Universal
Networking Language,icdim,08.
[9] Bangla Academy (2004), Bengali-English Dictionary, Dhaka.
[10] H. Uchida, M. Zhu, The Universal Networking Language (UNL)
Specification Version 3.0, Technical Report, United Nations
University, Tokyo, 1998
[11] Enconverter Specifications, version 3.3, UNL Center/ UNDL
Foundation, Tokyo, Japan 2002.
[12] Deconverter Specifications, version 2.7, UNL Center/ UNDL
Foundation, Tokyo, JapanA.
[13] D. M. Shahidullah, Bangala Vyakaran, Maola Brothers Prokashoni,
Dhaka, August 2003,pp.110-130
[14] D. C. Shniti Kumar,Vasha-Prokash Bangla Vyakaran, Rupa and
Company Prokashoni, Calcutta, July 1999, pp.170-175
AUTHORS PROFILE
Muhammad Firoz Mridha is now a full time faculty in Stamford
University Bangladesh. He obtained M.Sc. in Computer Science and
Engineering from United International University (UIU) in 2010 and B.Sc.
in Computer Science Engineering 2004 from Khulna University of
Engineering and Technology(KUET). His research interest includes Natural
Language Processing, Artificial Intelligence and Software.

Kamruddin Md. Nur attended Victoria University of Wellington, New
Zealand (VUW) for his bachelor in Computer Science (2004) and United
International University (UIU), Bangladesh for his masters in Computer
Science (2010). He is a permanent faculty member of Computer Science
department at Stamford University Bangladesh and his research interests
include Software Engineering Approaches, Software Visualization, Web
Engineering, Ubiquitous and Mobile Computing, Usability Engineering,
Artificial Intelligence etc.

Manoj Banik was born in Habiganj, Bangladesh in 1972. He completed his
B.Sc. Engineering in Computer Science and Engineering from BUET,
Dhaka, Bangladesh and M.Sc. Engineering in Computer Science and
Engineering from UIU, Dhaka, Bangladesh. Now, he is working as an
Assistant Professor in the Department of Computer Science and Engineering
of Ahsanullah University of Science and Technology, Dhaka,
Banjanto root + Primary suffix Noun
+
+
+
+
+
+
+
Vol. 2, No. 6, 2011
158 | P a g e
Bangladesh. His research interests include Speech Recognition, Phonetics,
Universal Networking Language and Algorithms. He is a member of
Bangladesh Engineers Institute (IEB).

Mohammad Nurul Huda was born in Lakshmipur, Bangladesh in 1973. He
received his B. Sc. and M. Sc. in Computer Science and Engineering degrees
from Bangladesh University of Engineering & Technology (BUET), Dhaka in
1997 and 2004, respectively. He also completed his Ph. D from the
Department of Electronics and Information Engineering, Toyohashi
University of Technology, Aichi, Japan. Now he is working as an Associate
Professor in United International University, Dhaka, Bangladesh. His research
fields include Phonetics, Automatic Speech Recognition, Neural Networks,
Artificial Intelligence and Algorithms. He is a member of International Speech
Communication Association (ISCA).

Vol. 2, No. 6, 2011
159 | P a g e
A Modified Method for Order Reduction of Large
Scale Discrete Systems

Dr. G.Saraswathi
Dept. of EEE, GIT, GITAM University
Visakhapatanam, India

Abstract In this paper, an effective procedure to determine the
reduced order model of higher order linear time invariant
discrete systems is discussed. A new procedure has been proposed
for evaluating Time moments of the original high order system.
Numerator and denominator polynomials of reduced order
model are obtained by considering first few redefined time
moments of the original high order system. The proposed method
has been verified using numerical examples.
Keywords- order reduction; eigen values; large scale discrete
systems; modeling.
I. INTRODUCTION
The rapid advancements in science and technology led to
extreme research in large scale systems. As a result the overall
mathematical complexity increases. The computational
procedure becomes difficult with increase in dimension.
Therefore high-order models are difficult to use for
simulation, analysis or controller synthesis. So it is not only
desirable but necessary to obtain satisfactory reduced order
representation of such higher order models. Here the objective
of the model reduction of high order complex systems is to
obtain a Reduced Order Model(ROM) that retains and reflects
the important characteristics of the original system as closely
as possible.
Several methods are available in the literature for large-
scale system modeling. Most of the methods based on the
original continued fraction expansion technique [1, 2] fail to
retain the stability of the original systems in the reduced order
models. To overcome this major drawback, alternative
methods have been suggested but the common limitation of
such extension is that in some cases, they may generate
models of order even higher than that of the original system
[1-3].
Modal-Pad methods [4] use the concept of the dominant
poles and matching the few initial time moments of the
original systems. The major disadvantage of such methods is
the difficulty in deciding the dominant poles of the original
system, which should be retained in the reduced order models.
Retaining the poles closest to the imaginary axis need not be
always the best choice. Sinha et al, [5] used the clustering
technique but the serious limitation of this method is the
number and position of zeros of the original system sometimes
decides the minimum order of the reduced order model. The
major disadvantage of such methods is in deciding the clusters
of poles hence cannot generate unique models.
Some methods based on Eigen Spectrum which is the
cluster of poles of high order system considered to derive the
approximant. Pal et al proposed [6] pole-clustering using
Inverse Distance Criterion and time-moment matching.
Vishwakarma and R.Prasad [7,8] modified the pole clustering
by an iterative method. The difficulty with these methods is in
selecting poles for the clusters. Mukherjee [9], suggested a
method based on Eigen Spectrum Analysis. The Eigen
Spectrum consists of all poles of the high order system. The
poles of the reduced model are evenly spaced between the first
and last poles. Parmar et al [10] proposed a mixed method
using Eigen Spectrum Analysis with Factor division algorithm
to determine the numerator of the reduced model with known
denominator. Parmar et al [11] proposed another mixed
method using Eigen Spectrum Analysis equation and Particle
Swarm method to find the reduced model. The methods based
on eigen spectrum analysis cannot be applied to high order
systems having complex poles. Saraswathi et al [17] proposed
a method retaining some of the properties of original system
based on eigenspectrum. The method can be applied for
systems having both real and complex poles unlike the other
existing methods [9-11].
Many methods are available for order reduction of high
order continuous systems but very few are extended for
discrete systems. The proposed method is extended to discrete
systems using Tustin approximation [12] to maintain the static
error constants identical in both discrete and continuous
transfer functions. In the proposed reduced order method the
Time moments and Markov parameters are redefined as
function of Residues and poles for strictly proper rational
transfer functions having real and complex poles. Poles of the
Reduced Order Model (ROM) are selected by considering the
highest contribution of each pole in redefined Time Moments
(RTMs) and lowest contribution in Redefined Markov
Parameters (RMPs).
II. PROPOSED METHOD
Let the original high order transfer function of a linear time
invariant discrete system of n
th
order be
()
()
()
; m < n ...(1)
Let the original high order transfer function of a linear time
invariant discrete system of n
th
order in continuous form using
Tustin approximation be
Vol. 2, No. 6, 2011
160 | P a g e
()
()
()
; m < n ...(2)
where () (
; n is the number of poles and

() (
; m is the number of zeros,
)

of high order system G(s). The poles
and zeros may be real and/or complex. If they are complex,
they occur in conjugate pairs.
The reduced order model of k
th
order using proposed
new algorithm in continuous form is defined as

()
()
()
; k<n (3)
where
() (
; k is the number of poles and
() (
; r m, r is the number of zeros,
)

of reduced order
model
(). In the reduced model poles and zeros may be

real and/or complex. If they are complex, they occur in
conjugate pairs as mentioned for original system.
Using Tustin approximation the reduced order linear time
invariant discrete model of k
th
order is
()
()
()
; k<n ...(4)
We know that the power series expansion of G(s) about s = 0
is
()
(5)
Where
()
; i = 0,1,2,3, ..
The expansion of G(s) about s = is
()
()
()
()

()
(6)
Where

I = 0,1, 2, 3, ..
i) Considering the original high order system G(s) with
distinct poles
Define the expressions for

redefined time moments
(RTMs) as

(7)
where
()

Define Redefined Markov Parameters (RMPs) as
(8)

where
and P
j
are residues.
ii) If G(s) is having r repeated poles
()
)(
)

; m < n
Define the expressions for

redefined time moments
(RTMs) as
()
[
()
]
(9)
where
()
()
{
()
()
}] and P
j
are
residues.
Define Redefined Markov Parameters (RMPs) as

] (10)
where
[
()
()
{
()
}] ;
= 0 if
and P
j
are residues.
The denominator polynomial D
k
(s) of the k
th
order
reduced model is obtained by selecting poles with the highest
contribution in RTMs and lowest contribution in RMPs
according to their contribution weight age as shown in Table I.
Table I : Contributions of individual poles
Parameters
1
2

3

j

n
Sum
RTMs x
i1
x
i2
x
i3
x
ij
x
in
RTM
i
RMPs y
i1
y
i2
y
i3
y
ij
y
in
RMP
i
Where x
ij
is the contribution of pole
j
in RTM
i
and y
ij
is the
contribution of pole
j
in RMP
i
. The numerator polynomial,
N
k
(s) of the k
th
order reduced model is obtained by retaining
the first few initial RTMs and RMPs of the original system as
follows:
()
; r = r
1
+ r
2
, r m
and r
1
1. (11)
where
()
; j = 0, 1, 2, 3,, r
1
; r
1
is number
of RTMs (12)
and
()

()
()
; j = 0, 1, 2, 3,, r
2
; r
2
is
number of RMPs (13)
r
2
, the number of RMPs is zero if r
1
is considered. If
, naturally r
2
= 0.
III. EXAMPLE
Considering a sixth order discrete system described by the
transfer function

given as
()

Using the Tustin transformation
() is transformed to
() with sampling time T =1.
()

The poles of G
1
(s) are -
1
= -1, -
2
= -2, -
3
= -3, -
4
= -
4, -
5
= -5 and -
6
= -6. The contributions of individual poles
are derived from equ.(7) for RTMs and from equ.(8) (non zero
terms) for RMPs. These contributions are tabulated in Tables
II and III. Poles having highest contribution in RTMs and
Vol. 2, No. 6, 2011
161 | P a g e
lowest contribution in RMPs according to their contribution
weight age are -
1
z
= -1 and -
3
z
= -4.
Table II Contribution of poles in RTMs and RMPs of HOS
RTMs/RMP
s
2

3

RTM
0
=1.33
24
1.616
7
2.2396 -7.7444
RTM
1
=1.64
48
1.616
7
1.1198 -2.5815
RMP
0
=1.00
00
1.616
7
4.4792 -23.233
RMP
1
= -5.4
-
1.6167
-
8.9583
69.700

Table III Contribution of poles in RTMs and RMPs of HOS
4

5

6

10.8146 -8.4433 2.8493
2.7036 -1.6887 0.4749
43.2583 -42.2167 17.0958
-
173.0333
211.0833 -102.575
For the second order model with two poles in ESZ,
Denominator polynomial is
D
2
(s) = (s + 1)(s + 4)
= s
2
+ 5s + 4
Numerator of the ROM is obtained using equation (11)
matching first two RTMs of the original system from Table II.
N
2
(s) = (1.3324 X 4) + {(1.3324 X 5) + (-1.6448 X 4)}s
= 5.329 + 0.8244s
The transfer function of the second order reduced model is
IV.
()

The conversion of continuous transfer function to discrete
R
1-1
(s) to R
1-1
(z) is done using the Tustin transformation with
Sampling time 1. The second order reduced discrete model is
()

The second order model by Farsi et al [13] is
()

The step response of the proposed second order discrete
model R
2
-
1
(z) and second order discrete model of Farsi et al
[13] R
F
2-1
(z) are compared with original discrete system G
1
(z)
in Fig.1. The step response by proposed method is following
the original very closely when compared Farsi et al[13].

Fig.1 Comparison of step responses of G
1
(z), R
F
2-1
(z) and R
2-1
(z)
V. CONCLUSIONS
Conclusions and Future Research: In this paper, an
effective procedure to determine the reduced order model of
higher order linear time invariant discrete systems is
presented. Numerator and denominator polynomials of
reduced order model are obtained by redefining the time
moments of the original high order system. The stability of the
original system is preserved in the reduced order model as the
poles are taken from the original system. The method produces
a good approximation when compared with other methods.
The method is applied for real, complex and repeated poles of
continuous [14] and discrete systems and the work is in
progress to make it generalize for interval systems.
REFERENCES
[1] Sinha, N.K. and Pille,W., A new method for order reduction of dynamic
systems, International Journal of Control, Vo1.14, No.l, pp.111-118,
1971.
[2] Sinha, N.K. and Berzani, G. T., Optimum approximation of high-order
systems by low order models, International Journal of Control, Vol.14,
No.5, pp.951-959, 1971.
[3] Davison, E. J., A method for simplifying linear dynamic: systems, IEEE
Trans. Automat. Control, vol. AC-11, no. 1, pp.93-101, 1966.
[4] Marshall, S. A., An approximate method for reducing the order of a
linear system, International Journal of Control, vol. 10, pp.642-643,
1966.
[5] Mitra, D., The reduction of complexity of linear, time-invariant
systems, Proc. 4th IFAC, Technical series 67, (Warsaw), pp.19-33, 1969.
[6] J.Pal, A.K.Sinha and N.K.Sinha, Reduced order modeling using pole
clustering and time moments matching, Journal of the Institution of
engineers (India), Pt.EL, Vol pp.1-6,1995.
[7] C.B.Vishwakarma, R. Prasad, Clustering methods for reducing order of
linear systems using Pad Approximation, IETE Journal of Research,
Vol.54, Issue 5, pp.326-330, 2008.
[8] C.B.Vishwakarma, R. Prasad, MIMO system reduction using modified
pole clustering and Genetic Algorithm, personal correspondence.
[9] S.Mukherjee, Order reduction of linear systems using eigen spectrum
analysis, Journal of electrical engineering IE(I),Vol 77, pp 76-79,1996.
[10] G.Parmar, S.Mukherjee, R.Prasad, System reduction using factor
division algorithm and eigen spectrum analysis, Applied Mathematical
Modelling, pp.2542-2552, Science direct, 2007.
[11] G.Pamar, S.Mukherjee, Reduced order modeling of linear dynamic
systems using Particle Swarm optimized eigen spectrum analysis,
International journal of Computational and Mathematical Sciences, pp.45-
52, 2007.
[12] Franklin, G.F.J.D.Powell, and M.L.Workman, Digital control of
Dyanmic systems, second, Addisan-Wesley 1990.
0 5 10 15 20 25 30
0
0.5
1
1.5
Original
Proposed
Farsi et al
Vol. 2, No. 6, 2011
162 | P a g e
[13] Farsi M.,Warwick K. and Guilandoust M. Stable reduced order models
for discrete time systems, IEE Proceedings, Vol.133, pt.D, No.3,pp.137-
141, May 1986.
[14] G.Saraswathi, An extended method for order reduction of Large Scale
Systems, Journal of Computing, Vol.3, Issue.4, April, 2011.
[15] G. Saraswathi, K.A. Gopala Rao and J. Amarnath, A Mixed method for
order reduction of interval systems having complex eigenvalues,
International Journal of Engineering and Technology, Vol.2, No.4,
pp.201-206, 2008.
[16] G.Saraswathi, Some aspects of order reduction in large scale and
uncertain systems, Ph.D. Thesis.

IJACSA Volume2No6

Uploaded by

Copyright:

Available Formats

IJACSA Volume2No6

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

IJACSA Volume2No6

Uploaded by

Copyright:

Available Formats

(IJACSA) International Journal of Advanced Computer Science and Applications,

Vol. 2, No. 6, 2011

is the probability that a respondent has the

is the probability that a respondent in Type of

type of Further Education

for each logit and the matrix xi is the

is the set of explanatory variables for

is the set of explanatory variables for

is the set of explanatory variables for

is the set of explanatory variables

is the set of explanatory variables

is the set of explanatory variables

is the interaction effect of

is the set of explanatory variables

is the interaction effect

is the set of explanatory variables for

is the interaction effect

is the interaction effect of

is the interaction effect of

and the step size was chosen

, which is much larger than the traditional

but still gets a good result.

is the estimate of the

Figure 5. Estimated background image.

(IJACSA) International Journal of Advanced Computer Science and Applications,

; n is the number of poles and

; m is the number of zeros,

; k is the number of poles and

; r m, r is the number of zeros,

(). In the reduced model poles and zeros may be

() with sampling time T =1.

You might also like