AJES Vol.2 No.2 (July-Dec 2013)

Asian Journal of Electrical Sciences
Volume 2 Number 2 July - December 2013
Contents
Sl. No. Title Page No.
1. Information Science (IS) and Its Increasing Interaction with Electronic Science 01
and Mechanical Sciences
Prantosh Kumar Paul and K.S.Shivraj
2. Conditional Power Monitoring of Domestic Loads through Human Machine Interface 08

M.C John Wiselin, C.V. Aravind, Rajparthiban Kumar and Sugendran
3. Vehicle Braking Performance Improvement by SVM Based Road Surface Detection 14

T. Anbalagan , C. Gowrishankar and A. Shanmugam
4. Approaching Vehicle Identification for the Visually Challenged 22

M.C. John Wiselin, C.V.Aravind, R. Rajparthiban and Uthaya Kumar
5. Fixed Frequency Sliding Mode - PI Control for Single Phase Unipolar Inverters 25
M.C. John Wiselin and Riyaz.A.Rahiman
6. Retinal Blood Vessels Segmentation Using the Curvlet Transform 31

M.C. John Wiselin and A. Prabin
7. Hybrid Method for Retinal Image Segmentation and Identifying True Vessels 38
A.Prabin, J.Veerappan, M.C. John Wiselin, S.Binu Sathiya and Parvathy R Nair
ISSN: 2249 - 6297 Vol. 2 No. 2, 2013, pp. 1-7
The Research Publication, www.trp.org.in
Information Science (IS) and Its Increasing Interaction with

Electronic Science and Mechanical Sciences
Prantosh Kumar Paul1 and K.S.Shivraj2
1
FBAS, Bengal Engineering and Science University, Howrah, West Bengal, India
2
EBET Group of Institutions, Tamil Nadu, India
E-mail: prantoshkpaul@gmail.com
(Received on 05 May 2013 and accepted on 10 June 2013)
Abstract Information Science plays an important role in the field

on. Here in all most all the areas of Information Science, the
of applied science and technology. Information Science is truly
main principle may be applicable. Cloud Computing, Grid
responsible for several societal improvements like for removing
Computing is no doubt wonderful tool of Electronic Science
information and digital divide, for improved Information
which may be applied on information science [02,13].
society, creation of information based knowledge Economy
and so on [9]. Information Science is mistakenly considered II. Objectives
as Computer Science branch or Library Science field but it is
totally different and interdisciplinary subject responsible for
The main objective of this research are:
Information Transfer Cycle. Information Science has two foci To know basic about Information Science;
manual and computational. The computational Information To learn the interaction in between Information
Science has several connection and relationship with the Science and Electronic Science;
world of electronics and Electrical Science. In this paper we
To learn the latest about Information Science foci
describe the interaction and interconnection in between
and strategies;
the Information Science and with technology which mainly
depends on Electronics and Electrical Science Emphasizing the
To learn the role of Electronics and technologies
contemporary scenario [10, 13]. for the improvement of Information science and
technology (IST) from the traditional Information
Keywords: Information Science, Electronic Science, Applied field.
Science, Information Science and Technology, IST, Computing,
III. Information Science
Pure Science
Information Science is subject which is centered in
I. Introduction
information activities and associated with computing
Information Science is a subject which mainly deals and technological fundamentals from basic to advance
with several information activities ranging from collection, [11]. Information Science is actually originated as pure
selection, organization and dissemination of information, information field as like Documentation and Information
data and related facet. The interesting fact about Information Studies but depending upon time and advancement come
Science is that, Information Science may be depend on as an interdisciplinary knowledge cluster which is actually
normal knowledge organization tool or may be applied knowledge combination of Computer Science, Management
Information and Computing tools and technologies [09, Science, Information Technology, Documentation, Cognitive
10]. The relationships in between Information Science with Science and other subjects are responsible and associated with
electronic depend technologies are increasing day-by-day. Information Activities. Information Science is considered as
The normal activities of Information Science is mainly field of fields and emerging Applied Science field. Computer
responsible and applied in information centre, information and Information Technology are the key mover of this subject
systems, information networks, information grids and so [05, 12].
1 AJES Vol.2 No.2 July - December 2013

Fig. 1 Main task of information science
A.Information Science: Two Foci Web Technology and Systems;

Information Repositories;
Information Science practically has two foci or
division, however the root of both the foci was actually Information Architecture;
Documentation Studies or Science which always depends Advance Computing;
on manual knowledge organization tools like indexing, Networking Technology;
abstracting, database, classification, cataloguing and so on.
Computer Technology;
These foci are mainly applicable in Information Centre,
Multimedia Information System;
Documentation Centre, Information Offices, Information
Analysis Centre, Information Repositories and other related Database System;
organization [13, 08]. Information Management;
Information Security;
Other hand next foci of Information Science depend
on Mechanical and Electrical Science, particularly in Cloud Computing and so on.
Computer Science, Computer Technology and other The applied Information Science is applicable in almost
Engineering Fundamentals [10]. This foci some times all the areas which are mention in Manual Information
treated as follows:- Science. However it is also applicable in the industries,
Computerized Information Science; organization and other areas.
Applied Information Science;
B. Electronic Science and Its Utilization in Practice Field
Information Science & Computing; of Information Science
Electronic Information Science;
Electronic Science is today widely associated with
Industrial Information Science and so on.
Information Science and allied Sciences. In many areas
The applied Information Science is actually considered application of Electronic Science is possible in Information
as Informatics. This Information Science has some advance Science in the following areas.
Computing and Interdisciplinary topic like
AJES Vol.2 No.2 July - December 2013 2

Information Science (IS) and Its Increasing Interaction with Electronic Science and Mechanical Sciences
Fig. 2 Role and uses of Manual Information Sciences
1. General Communication: For general communication For the intercommunication in between one Information
and more clearly in general networking we need the help centre to another or one Information Network to another
of Electronic Science. The use and utilization we can use in we need to take the help of computer Networking[10].
the following activities.
For building general connection in between one
For in house communication in between one room to international Information Networks to another or
another or one section to another or one floor to another Information System to another we need the help of
in the information centre or Information Networks; that Computer Networks. For share of data, Resources,
means for Local area Communication, we need the help Hardware, Software and so on.
of Computer Networking.
Fig. 3 New opportunities through Information Science

Information Science also takes the help general In computerized Information Services;
Computing and Computer Applications [11]. The general
In Computerized Document Delivery system and
computing Application is including as follows:
services;
General official activities of an office of Information
In Reference and Referral Services;
Centre;
In Online Information Delivery System;
Computerized cataloguing and MARC systems;
In Digital Information Repositories;
General E-mail system and services;
In Office Automation and so on.
General and computerized classification system;
Information Science today increasingly depends on
Official Documentation;
Advance Computing tools and technologies like Cloud
In news paper clipping services, Current Awareness Computing, Green Computing, Usability Computing and so
Services(CAS), Selective Dissemination of on.
Information(SDI) and so on;
Fig. 4 From IS to IST nomenclature through Electronic Science

2. Cloud Computing: This is a kind of virtualization in 3. Green Computing: The main approach of Green
which sharing of hardware, software and packages are computing is actually power management, material
possible like [11] recycling and tele-conferencing and so on. Green Computing
some times refers to Green Technology [12,13]. The main
All the software, hardware and application of information
base of Green Computing in information science practice is
centre or related organization may share with the help of
actually as follows:
cloud computing;
Better algorithm design helps in power managed
Expansion regarding IT and Computing infrastructure;
computing;
To design, development and implementation of
As information practices deals with information and
Information Systems and Information Networks, it is
computing tools so that saving and designing Eco-
useful;
friendly computing is useful;
It is useful in publishing house and journal for sharing
The information foundation may be re-cycling principle
of software and hardware and other packages;
for future saving.
Online information mechanism it is widely applicable.
Fig. 5 Changing Character of IS
4. Usability Computing: Usability computing is Design and development of web pages of Information
actually designing and development of usability experience Centre, Information Analysis Centre, Documentation
designing. This also indirectly helps in information Centre and related organizations;
activities. The main aim of Usability Engineering is
In search engines;
design and development of user-friendly interface, picture
designing so in several areas this is applicable like In Graphical User Interfaces(GUI);
Design and development of information retrieval Mobile Information Services.

system;

5. Quantum Information Service: Quantum Information Cloud Computing-directly relate with Electronic
Science is most important and contemporary application of Science.
Electronic Science to the information services. The main
Green Computing-directly relate with Electronic
benefit of this is including
Science and so on.
Easy and fastest information, data and knowledge
delivery; V. Findings
Correct and current information and data delivery and During this study we find out the following:
sharing;
Information Science is now totally computing and
Reliable information activities Electronic Related discipline.
Easy and fastest decision making process; Information Science has two foci.
Better computing and information practices and so on.
Information Science mistakenly consider as computer
IV. From IS to IST and Electrical Science Science.
Information Science is an interdisciplinary subject Information Science is widely applied in several areas
responsible for information collection, selection, apart from existing held.
organization and management and lastly dissemination.
New nomenclatures are emerged like Information
During the evolution and development information science
Science and Technology, Information Science and
is considered as information field but due to application
Computing, Automated Information Science and so on
of Electronic Science and computing technology it is now
due to emerging interaction with Electronic Science.
totally considered as Applied Science and in some countries
and scientific community as Applied Technology field. Quantum Information Service, Cloud computing, Green
Computing are the most emerging fact of Information
Due to most Engineering IT and computing Application,
Science.
Engineering Fundamentals Information sciences Traditional
look and is totally changed and renamed with many new VI. Suggestion
nomenclatures those we already discussed in this study.
Information Science needs much more Electronic
But most hot and contemporary thought of todays age is
Science and Computer Science interaction world wide
actually nothing but the IST or in full form Information
for its subjective benefit.
Science and Technology. So new area of Electronic Science
make this new wonderful knowledge cluster is including: The Information Science syllabus needs moderation and
changes due to industry requirement.
Quantum Information Science-directly related with
Electronic Science[10] Information Science programme needs much more
merging with Electronic Science for creation of new
Usability System and Engineering-indirectly related
nomenclature like IST In the developing country like
with Electronic Science.
India.
Web Engineering-indirectly related with Electronic
It is essential to launch Information Science programme
Science.
in the department of Computer Science and Electronic
Digitalization-indirectly related with Electronic Science. Science and related discipline.
E-Resource management-indirectly related with

Electronic Science.

VII. Conclusion [7] Mitra, C.R. University-Industry interaction with reference to

RECs. University News 35(25) June 23:14. 1997.
The role and nature of Information Science is changing [8] Prytherch, Ray. Problems of research. Information Management
day-by-day. The Electronic Science and Allied Science Report (Sept): 18. 1997
truly changed the entire arena of Information science [9] Paul, Prantosh Kumar, Dipak Chaterjee and Bhaskar Karn
[13, 14]. The Information Science, Information Systems, Information Science Education and Research: emphasizing
contemporary Indian scenario- an overview in IEM/IEEE sponsored
totally give a new look of Information Science and its
international conference proceedings (IEMCON-12). P-349-353.
industrial practices. The R/D activities need much more [indexed, abstracted in Google Scholar[USA], Cite Ceer, EBSCO]
interaction and cooperation with Electronic Science as far [10] Paul, Prantosh Kumar, Dipak Chaterjee and Bhaskar Karn Cloud
as Information Science is concerned for better IT depended Computing: emphasizing its possible roles and importance in
Information Systems and Centers in IEM/IEEE sponsored
Knowledge Economy [10].
international conference proceedings (IEMCON-12). P-345-348.
[indexed, abstracted in Google Scholar[USA], Cite Ceer, EBSCO]
References
[11] Paul, Prantosh Kumar, Dipak Chaterjee and Bhaskar Karn Cloud
[1] Diener, R. Information Science: What is it?...What Should it Be? in Computing: beyond ordinary Information Transfer Cycle submitted
Bulletin of ASIS, June/July, 1989, pp 17-21. in National Conference on Computing and Systems, Dept of
[2] Buckland, M. Information as Thing. in JASIS, June 1991, vol 42, No Computer Science, Burdwan University. (Status- Accepted) 15
5p 351-359. March
[3] Kaula, P.N. Hundred years of library and information science [12] Saracevic, Tefko, Information Science, JASIS, 50(12):10511063,
education In Kumar, P. S. G. and Vashisth, C.P. Eds., Library and 1999,ASIS,USA,1999
information science in India. New Delhi: Sterling Publishers. 1992 [13] www.en.wikipedia.org
[4] Keren, Carl. On information science. Journal of the American [14] Williams, J. Information Science: definition and Scope in Williams,
Society for Information Science 35(2): 137. 1984 J. and Carbo T. (eds.) Information Science: Still an Emerging
[5] Kumar, P.S.G.. Research in library and information science in India. Discipline. Cathedral Publishing, Pittsburgh, Pa, 1997.
Indexed by A. Tejomurty and H.R. Chopra. New Delhi. 1987
[6] Lahiri, Ramansu. Research in library science in India (195095):

an account of Ph.D. Programme. Annals of library science and
documentation 43(2): 5968. 1996.

ISSN: 2249 - 6297 Vol. 2 No. 2, 2013, pp.8-13
Conditional Power Monitoring of Domestic Loads through

Human Machine Interface
M.C. John Wiselin1, C.V. Aravind2, Rajparthiban Kumar3 and Sugendran4
1
Travancore Engineering College, India
2
CIARG, Taylors University, Malaysia
3
The Nottingham University, Malaysia, 4UCSI University, Malaysia
E-mail: aravindcv@ieee.org
Abstract A real time data acquisition system for load micro processing unit such as the Programmable Logic
monitoring using human machine interface is presented. The Controllers (PLC), capacitor banks and so forth. Advanced
design proposes a real time data acquisition whose parametric control systems are initially installed in the modern RTUs
values are monitored and recorded for analysis purposes in a in order to provide better and efficient system operation. The
typical power system set-up. The parametric measurements communications involved in this system are basically wired
includes the load voltage, load current, active power, reactive transmission where only cables and wires are attached. USB
power, apparent power, power factor and line current. A communication are implemented for the Data Acquisition
perfect interface design is developed using graphical system Module where data from the designed RTU are transmitted
design software. The recorded data is highly helpful in power by wires and the processed informations are then passed
monitoring and analysis for performance evaluation. The to the host computer. The HMI design is basically built in
design gets complete with a manual supervisory control based order to view and display all the informations processed by
on the data acquired through the interface module. the DAQ system from the RTU. LabVIEW software is used
as the interface panel where the software is programmed in
Keywords: Power Monitoring, DAQ, Graphical System Design
order to receive the correct data on the desired slots [2] [3].
I. Introduction
II. Methodology
The essence of power has become the most dominant
The primary objective of this system is to monitor and
asset in humans life. Power distribution has made its premier
control power in a single phase network. As mentioned,
role by contributing to applications from submicron level to
the system relies deeply on measurement and analysis on
all major devices [1]. A supervisory control is always the
non-linear loads where the essence of active power, reactive
need of any power monitoring system which consists of a
power, load voltage, load current, apparent power, power
Remote Telemetry Unit (RTU), communication circuits and
factor and phase. The system is mounted on a wooden board
Human Machine Interface (HMI). The RTU includes the
with proper wiring and electrical grounding as well. Voltage
implementation of two different sensing devices which are
transducer and the Hall Effect current transducer majors the
the voltage transducer and a Hall Effect current transducer
initial design of the system and measurement is made on the
(LTS 15-NP). The voltage transducer is a self made sensor by
hardware before the software measurement for validation of
using two rated resistors to match and calculate the reference
the results. Figure 1 shows the schematic representations of
voltage in order to transport the data into the data acquisition
the design.
system. The general RTU established in the industries are
attached with high current/voltage control relays, advanced

Conditional Power Monitoring of Domestic Loads through Human Machine Interface
Fig. 1 Schematic Representation
The Data Acquisition Module is the heart of the system 6008 DAQ module is used and the mode of communication
where this module holds the responsibility to perform sig- is established by using USB and the current transducer is
nal and information acquisition. The National Instruments from LEM as shown in Figure 2.
Fig. 2 Voltage transducer schematics
III. System Development though these loads are for commercial use. The RTU is
passed through the Hall Effect current transducer and it is
The design of the hardware incorporated with the
connected to the switch wall sockets in order to sense the
project is a rapid prototype where the materials are found
load current. The mathematics that involve in this section are
correlating to the cost as well. Figure 3 shows the system
applied to the current sensing methodology and the voltage
module setup with the HMI interface panel.
transducer because the DAQ system can only take small
Prioritized experimental loads as such the electric iron value of voltage and current. Figure 4 shows the graphical
and electric kettle are tested as the loads manufactured system design interface for the system module under test.
with hot plates and coils draw very high current even The design of the software caters the requirement of the

M.C. John Wiselin, C.V. Aravind, Rajparthiban Kumar and Sugendran
Fig. 3 System Module under test
power monitoring aspect where all the desired parameters design with several important programming functions such
are placed and analyzed. The most important part of the as the indexing array, Boolean function, tab control and
software is to visualize the RMS voltage and RMS current etc. The software design required a very long time in order
where the theoretical explanation of the control system can to search and match the parameters in order to associate
be mentioned. G-programming concept is used in this HMI with the theoretical power equations. Figure 4 shows the
software interface of the module.
Fig. 4 Software interface panel
IV. Testing and Results the scaling factor calculated from the transformer ratio is
about 60. The scaling factor is one of the most important
In this design, the testing section is separated into two
components as this value determines the load voltage.
segments which are the hardware testing and the software
On the other hand, the software testing is established by
testing. Figure 5 shows a typical test pane to acquire the
trial and error and thus, the current formulation is created
data. As for the hardware section, the reference voltage
manually. This is because any output that comes from the
and the scaling factor must be calculated manually before
Hall Effect sensor is a voltage. Therefore, this situation be
programming the software. Based on the measurement
analyzed as voltage in terms of current by re-shuffling the
made, the reference voltage is observed to be 4.007 V and
formula as

Fig. 5 Test Pane in the Graphical System Design
Fig. 6 Sensed current formula
Figure 6 shows the equation representation of the sensed loop until the user stops the simulation. The results were
current formula. This formula indicates the sensed current analyzed for a number of non-linear loads and are tabulated
on the load with a variable X which is generated from the accordingly. A typical for the power monitoring over a
current sensor and the software performs calculation and certain period of time and the comparison in terms of the
finally displayed. The programs designed are not written but parametric identity are compared and presented in the
it is graphically programmed where with the functions are Table I. Figure 7 shows the graphical module for the system
dragged and dropped as long as the function corresponding result identification for more accurate understanding of the
to the parameters are true. While loop condition is used as to readings in terms of monitoring and for analysis purpose.
measure real time signal acquisition and will continuously
Table I Parametric Records For Different Loads

(Single Phase Non-Linear Loads)

M.C. John Wiselin, C.V. Aravind, Rajparthiban Kumar and Sugendran
Fig. 7 System Result Identification
Fig. 8 RMS voltage waveform

V. Discussions obtained. Capacitor bank design and voltage regulation are

best viewed in three phase networks and the precision of the
The results obtained using the system HMI interface
involved measurement will increase. In this project, many
is measured in real time sequence and fluctuations in the
problems were face in terms of measurement aspect when the
power line are observed. There are anonymous overshoots
formula node was incorrect, parameters measured wrongly
in the system which proves instability and erroneous in
and so forth. Some of the problems faced required lots of
the power network. The average voltage reading measured
time and finally managed to obtain results and compared
manually from different switch wall sockets from the same
with the manufacturer ratings. The graphical essence was
residential house shows a reading of 247.2 V. Overvoltage
not depicted much as this only shows fluctuation for the
occurrences would result to high inrush of currents towards
RMS voltage and RMS current. Most measurements made
the loads and damages are observed. From Table I, the
are in numeric and the RMS voltage waveform is observed
active power and the power factor tend to provide the
as shown in Figure 8.
results in negative because the current direction is looping
and reversed in polarity. But, the magnitude measure proves References
to be correct and the power factors of all residential loads [1] Jens Matics, Gerhard Krost, Carsten Roggatz and Udo Spanel,
are nearing unity. One of the sections in this project could Dispersed Generation Modelling in SCADA Time Scale, IEEE
Publication
not be implemented which is the control system aspect and
voltage regulation. Even though experiments was conducted [2] D. Beck, H. Brand, S. Gotte, F. Herfuth, C. Rauth, R. Savreux, S.
Schwarz, Y. Yazidjian, The CS Framework- A LabVIEWBased
using self made capacitor banks and etc, the experiment Approach to SCADA System, 10th ICALEPCS Int. Conf. on
fail to produce result and only the power monitoring was Accelerator & Large Expt. Physics Control System. Geneva, 10
established for the time being. When the power monitoring 14 Oct 2005, 2005, pp.1051-60.
was visualized in depth, a vast amount of informations was [3] S. Sumathi, P. Surekha, LabVIEW Based Advanced
Instrumentation Systems, Springer Publication, 2007

ISSN: 2249 - 6297 Vol. 2 No. 2, 2013, pp.14-21
Vehicle Braking Performance Improvement By

SVM Based Road Surface Detection
T. Anbalagan1, C. Gowrishankar2 and A. Shanmugam3
1
Specialist, Active Safety, Robert Bosch Engineering and Business Solution, Coimbatore, India
2
Assistant Professor, Department of EEE, K.S.R. College of Engineering, Tiruchengode, Tamil Nadu, India
3
Principal, Bannari Amman Institute of Technology, Sathyamangalam, Tamil Nadu, India
E-mail: anbalagan4u@gmail.com, cgsshankarme@gmail.com,, dras_bit@yahoo.com
Abstract Advance information about the road surface a vehicle is in the classification of different kind of road surfaces like
is going to encounter can improve the performance of Antilock cement, asphalt, sand, rough and grass to provide assistance
Braking System (ABS). For e.g. the initial slip cycles caused by to automotive systems. Research work already done relates
the ABS could be avoided, if it is already known that the vehicle is to classification of road surfaces in context with still
on a surface having a low coefficient of friction (). In this paper, images of asphalt or concrete macro texture as in [1] or
an algorithm is developed that detects different road surfaces based on spectrum analysis as in [3]. In [2], fast Fourier
using streaming video acquired from a camera mounted on the transform is used to determine the road surface conditions.
hood of the vehicle. The road surfaces detected here are asphalt However, it has computational complexity and the response
road, cement road, sandy road, rough asphalt road (asphalt may be sluggish. Here we propose a novel idea of image-
road which is deteriorating), grassy road and rough road. based road surface detection system based on simple
The value of coefficient of friction () is also given out with image processing techniques like Canny edge detection,
the detected surfaces to obtain additional information about Hough transform, contours, intensity histogram and
the road surfaces. Split (a road having different conditions image segmentation and compare with the SVM. Simple
on the left and right side of the vehicle) and jump (different image processing algorithms are used on the continuously
conditions on the front and rear of the vehicle) are also pre streaming video thereby reducing time and computational
detected. One method was not sufficient to achieve the goals complexity of the imaging system. It avoids the training
of this algorithm. Here several simple techniques like the schemes givenin [4] as decisions are given out instantly. The
Canny edge algorithm, intensity histogram, contours, Hough algorithm is robust as it works even in shadowy conditions.
transform and image segmentation were employed and In [5], the decision is taken only with histogram, whereas
compared with the Support Vector Machine (SVM). To prevent in the proposed algorithm this technique is used effectively
misdetections, the road surface detection during high motion and judiciously to take decisions when edge detection
blur is prohibited. technique is not sufficient.
Keywords: Computer vision; Vehicle dynamics; Vehicle safety This algorithm makes the early detection of road
system; Signal processing; Automotive systems; Electronics in surfaces which will allow the vehicle dynamic system to
industry and transport react decisively with the changing road conditions. It also
I. Introduction identifies lane marking on cement and asphalt based on
Hough transform as in [9]. Split and jump conditions
Images provide a dense environment perception and are also being pre detected and increases the efficiency of
their benefits especially when used in vision- based systems ABS. So the stopping distance also reduces.
are enormous. One such application of vision based analysis

Vehicle Braking Performance Improvement by SVM Based Road Surface Detection
II. Image Processing Technique to 195, 135 to 170 and 50 to 200 are found to classify sand,
cement/asphalt and grass respectively taken from [7] [8].
Image processing techniques are employed to develop
this algorithm. It is discussed in detail in this section. E. Support Vector Machine (SVM)
A. Canny Edge Detector The basic SVM takes a set of input data and predicts,
for each given input, which of two possible classes forms
Edge detection is used for (i) identification of blurred
the output, making it a non- probabilistic binary linear
frames (ii) broad classification among smooth and rough
classifier.
surface (iii) classification of cement and asphalt. The
Canny edge detection is performed on the frames with III. Proposed Algorithm and Decision Making
the sensitive threshold values (upper threshold 10000 and
lower threshold 4900) and again it is performed with the The six main parts involved with algorithm and
insensitive threshold values (upper threshold 50000 decision making are (i) blur detection, (ii) broad
and lower threshold 9800). If a pixel has a gradient greater classification of road surfaces (iii) classification of smooth
than the upper threshold, then it is an edge pixel. If a pixel roads (iv) Classification of rough roads (v) classification of
has a gradient lower than the lower threshold, it is not an split and jump (vi) co-efficient of friction for different
edge pixel. If the pixels gradient is between the upper and road surfaces. The procedure below is followed to classify
the lower thresholds, then it is considered as an edge, only different kinds of road surfaces. The flowchart of the
if it is connected to a pixel that is above the high threshold proposed algorithm is shown in Fig.14.
value as given in [7] [8]. The number of edges is then
A. Blur Detection
computed.
The frames from the streaming video are converted into
B. Contours
gray scale. Canny edge detection with sensitive threshold
Contours are connected edges and are used to value identifies the motion blurred images since the blurred
(i) identify the lane marking (ii) classify rough asphalt frames have edges less than 1000. Once blur is detected,
from rough road. Contours are drawn on the frame with the those frames are not considered for decision making.
threshold value of 210 and again contours are drawn for the
B. Broad Classification of Road Surfaces
threshold value of 75. The reason for choosing the threshold
values are mentioned in the section III. The threshold values To classify rough and smooth surface, Canny edge
are chosen such that the pixels brighter than the threshold detection with insensitive threshold value is performed.
value are alone identified as in [7] [8]. The number of The insensitive threshold value is chosen such that the
contours for the threshold value of 75 is alone computed. problem of shadow and other disturbances are eliminated
since shadows form weak edges. Man made partitions like
C. Hough Transform
lane marks and patterns on a smooth surface contribute to
Hough transform is used to (i) identify man made more number of edges causing misdetection. So lines are
partitions on the surfaces (ii) find the line segments (lane detected using Hough transform in the frames to eliminate
marking) in an image given in [9]. The Standard Hough man made partitions. The edges are then computed on
transform is used to map each pixel in image space to a line the frames where man made partitions are eliminated.
in Hough space and vice versa as mentioned in [7]. Fig. 7 shows the calculated edges with insensitive threshold
for different types of road surfaces. Smooth s u r f a c e
D. Intensity Histogram is classified from rough surface since it has less number
of edges (less than 1000) when compared to rough surface.
Histogram is used to (i) classify sand from cement/
Cement, asphalt and sand are classified as smooth. Rough
asphalt (ii) classify grass from rough surface. The total
asphalt, rough road and grass are classified as rough.
number of pixels having intensity values ranging from 150

T. Anbalagan, C. Gowrishankar and A. Shanmugam
C. Classification of Smooth Roads 3. Sand

1. Cement and Asphalt Roads with lane marking Intensity histogram is used to find the total number
of pixels in the range of intensity values 150 to 195 and
A lane marking is a straight line in an image and
135 to 170. If the total number of pixels between 150 to
the gray scale value lies in the vicinity of 210, thereby
195 ranges as shown in Fig. 12 is greater than 135 to 170,
using contours with the threshold value of 210 and Hough
the smooth surface is classified as sand since those intensity
transform, lane marking is identified as shown in the
values are typical indicators of sand surface.
Fig. 1. The threshold value of 210 is chosen to identify
lane marking such that the pixels brighter than the D. Classification of Smooth Roads
threshold value 210 are retained as non zero (white pixels
1. Grass
corresponding to the lane marking are alone identified).
Canny edge detection with sensitive threshold is performed Grass is expected to have more number of pixels lying
and the number of edges is computed. Fig. 8 shows the between the intensity values 50 to 200 because the intensity
calculated edges with sensitive threshold for different values for grass is widely distributed as in Fig. 11. So grass
types of road surfaces. Cement has less number of edges is classified from rough road and rough asphalt because
(less than 6000) when compared to asphalt as shown in the intensity distribution of rough asphalt/rough is not as
Fig. 8. Once lane marking is detected, it is sure that the wide as that of grass.
surface is either cement/asphalt. 2. Rough Asphalt
Intensity histogram is used to find the gray scale intensity Contours with the threshold value of 75 is used to
value of the frame having lane marking. To distinguish separate rough asphalt from rough road since the contours
asphalt/cement from sand, adaptive threshold is used. for rough asphalt are lesser than 2000 as in Fig. 9. At the
The intensity value which is already chosen to classify threshold value of 75, rough asphalt has less number of
cement/asphalt is changed to the intensity value of the lane connected edgesbecause it does not have pixels brighter
mark detected frame. This technique can accommodate than the threshold value 75.
changing lighting conditions in the image and also can
be adapted to lane marks. Successive frames are then 3. Rough Road
checked for the intensity levels. If the intensity values
For rough roads, the contours with the threshold
are approximately same, the frame is not checked for the
value of 75 are greater than 2000 as shown in Fig. 9
other parameters like edges and contours, instead they
because the brightness of the pixels on a rough road is
are simply classified as cement or asphalt.
mostly greater than the threshold value 75. And hence the
2. Cement and Asphalt Roads without lane marking connected edges are more prominent.
Asphalt/cement has more number of pixels in the E. Classification of Split and jump
range of 135 to 170 as shown in Fig. 10 since these intensity
Split roads have the coefficient of friction values
values represent the gray color band of cement/asphalt. If
significantly different between the left and the right wheel
the total number of pixels between the intensity values
path. They are identified by dividing each frame into two
135 and 170 is greater than the intensity values ranging
equal halves and the concepts like edges, contours,
from 150 to 195 as in Fig. 12, then it is either cement or
Hough transform and histogram are applied to both the
asphalt. Canny edge detection having sensitive threshold is
divided frames. The total number of edges, total number of
again used to classify cement from asphalt similar to the
contours and the total number of pixels count found from
procedure done after the lane marking +has been detected.
the histogram technique are reduced to approximately half
of the values computed from the entire frame. When the
decision of the road surface taken by the divided frames is
different, then it is split .

Whenever there is a transition from one surface to

another surface, then it is jump. The transition can be
from a surface having low coefficient of friction (rough)
to a surface having high coefficient of friction (smooth) or
vice versa. The former is positive jump and the latter is
negative jump.
F. Coefficient of friction values for different roads

Fig. 4 Sand road
Once the surface is detected the values are also given
out for various kinds of road surfaces as in Table I [6]
Table I Coefficient of Friction
Fig. 5 Rough road
(a) (b)
Fig. 1 (a) Cement road with lane mark, (b) lane
mark identified on Figure 1.(a)
Fig. 6 Grass road
Fig. 2 Asphalt road
Fig. 7 Edges with insensitive threshold value for different types of road
surfaces
Fig. 3 Rough Asphalt Road

Fig. 8 Edges with sensitive threshold value for different types of road surfaces
Fig. 9 Contours with threshold value of 75 for different types of road surfaces
Fig. 10 Intensity histogram for cement

Fig. 11 Intensity histogram for Grass
Fig. 12 Intensity histogram for Sand
Fig. 13 Margins for an SVM trained with samples

Fig. 14 Flow chart for the proposed Algorithm
The stopping distance of the ABS has been reduced

based on the result of the algorithms.
Table II Results of the Algorithm Table III Accuracy Results of The Algorithm
Stopping Stopping
Video Type Distance Distance using
existing SVM ()
technique
Cement 8 8.23 9 8.32
Asphalt 9 2.59 9 9.64
San 9 6.80 9 9.89
Rough asphalt 8 2.50 9 7.62
Rough 98.94 99.92
Gras 7 6.31 9 5.20
IV. Results and Discussion cement as referred in Table III, 75 frames were detected
correctly and rest were either blurred or were wrongly
Streaming video of the road surface ahead of the vehicle
detected. Thus a significant level of accuracy was achieved
was acquired by the camera mounted on the hood of the
through the developed algorithm.
car. This streaming video was processed by the algorithm
to detect and classify various road surfaces simultaneously. In future, it can be extended for night conditions as well,
Table II shows the stopping of the ABS for classification by providing infrared cameras for surface detection. Further
of different kinds of road surfaces. Table III shows the this algorithm can be improved if detection of surfaces such
accuracy of the algorithm for classification of different as ice, snow and water can be done based on the specularity
kinds of road surfaces. For example, if 85 frames are of type of such surfaces.

References [8] Rafael C. Gonzalez, Richard and E. Woods, Digital Image

Processing, Third Edition, 2009.
[1] Ronald Elunai, Vinod Chandran, and Edith Gallagher,
[9] Bin Yu and Anil K. Jain, Lane Boundary Detection Using A
Asphalt Concrete Surfaces Macro texture Determination From
Multiresolution HoughTransform, Dept. of Computer Science,
Still Images, IEEE Transactions on, Intelligent Transportations
Michigan State University, East Lansing, MI48824, IEEE
Systems, vol. 12, no. 3, September 2011, pp .857869.
Conference, October 1997, vol. 2, pp. 748-751.
[2] Yoshiroh Murata and Masashi Takeuchi, Junichi Takagi and
[10] Ahmed Kharrat, Karim Gasmi, Mohamed Ben Messaoud, Nacera
Kouichi Egawa , Road surface recognition sensor using an optical
Benamrane and Mohamed Abid A Hybrid Approach for Automatic
spatial filter Central R&D Lab, IEEE Conference, Shimokaiinj,
Classification of Brain MRI Using Genetic Algorithm and Support
Nagaokakyo, Kyoto,1997, pp. 1000 1004.
Vector Machine, Leonardo Journal of Sciences, No. 17, pp. 71-
[3] Pidwerbesky, Waters, Gransberg, Stemprok, Road surface texture 82, 2010.
measurement using digital image processing and information
[11] Amir Ehsan Lashkari Neural Network based Method for Brain
theory, Land Transport New Zealand Research Report 290, 2006,
Abnormality Detection in MR Images Using Gabor Wavelets,
pp. 6579.
International Journal of Computer Applications, Vol. 4, No. 7, pp.
[4] Raqib Omer , Liping Fu , An Automatic Image Recognition 9-15, 2010.
System for Winter Road Surface Condition Classification, 13th
[12] Beena Ullala Mata, B.N. and Meenakshi, M. A Novel Approach
International IEEE Annual Conference on Intelligent Transportation
for Automatic Detection of Abnormalities in Mammograms IEEE
Systems, Madeira Island, Portugal, September 2010, pp. 1375-
International conference on Recent Advances in I n t e l l i g e n t
1379.
Computational Systems (RAICS), 2011.
[5] Jose M. Alvarez and Antonio M. Lopez, Road Detection Based
[13] Chang, G.S., Bin Yu and Martin Vetterli Adaptive Wavelet
on Illuminant Invariance, IEEE Transactions on Intelligent
Thresholding for Image Denoising and Compression, IEEE
Transportation S ystems, 2011, vol. 12, no. 1, pp.184193.
Transactions Image Processing, Vol. 9, No.9, pp 1532-1546, 2000.
[6] Peter D.Cenek, Niel J.Jamieson and Maurice W.Mclarin, Frictional
[14] Dheeba, J. and Wiselin Jiji, G. Detection of Microcalcification
characteristics of Roadside Grass Types , Opus International
Clusters in Mammograms using Neural Network, International
Consultants, Central Laboratories, Grace field, New Zealand.
Journal of Advanced Science and Technology, Vol. 19, pp 13-22,
[7] Gary Bradski and Adrian Kaehler,Learning OpenCV, OReilly 2010.
Media, Inc., 1005 Gravenstein Highway North, Sebastopol,2008.

ISSN: 2249 - 6297 Vol. 2 No. 2, 2013, pp. 22-24
Approaching Vehicle Identification for the Visually Challenged

M.C. John Wiselin1, C.V.Aravind2, R. Rajparthiban3 and Uthaya Kumar4
1
Travancore Engineering College, India
2
CIARG, Taylors University, Malaysia
3
The Nottingham University, Malaysia, 4UCSI University, Malaysia
E-mail: aravindcv@ieee.org
Abstract The paper presents the development of a novel person is inevitable in any particular day activity. Many
information system for the visually impaired. The objective of commercial products have been developed to sustain the
the developed system is to enable the visually impaired people disabilities substantially thereby improve their quality of
to get information of the incoming source of vehicle without any life. The use of these devices allows the user to experience
other physical help and to travel by vehicle such as buses to their the freedom of certain aspects of life such as mobility,
desired destinations. This device works in such a way, where communication and other important fundamental tasks.
the visually impaired people are able to detect the arrival of a Normally the visually challenged persons find difficulty in
particular vehicle as their only source of transportation. The their mobility to reach their destination. Most of them fall
system incorporated is designed to be simple and portable hand under the lower income group and mostly depend on the
held for the convenient usage. A simple wireless communication public transport [2][3]. The most preferable public transport
system with frequency modulation is employed for the signal of common man is bus as it is cheap and economical. The
transmission. The microcontroller in the system is used to process of catching a bus involves the combined skills of
generate the automatic signal that is transmitted through the hearing, sight and cognition. The task will become nearly
FM transmitter. The signal picked up by a single chip FM impossible for those visually challenged persons without
receiver and announces voice message of the coming vehicle the assistance of another person. The developed system
information using the voice record and playback device. In provides a solution for this problem having the freedom to
this way the visually impaired will manage to catch the vehicle independently commute via public bus transportation. The
by their own ability. However the model developed aims for developed system is tested with a single unit under real time
identify only one vehicle but can be extended to any number condition.
of systems. The FM carrier frequency used for communication
II. System Description
between the transmitter and receiver is 89.3MHz. The maximum
transmission distance for transceivers signal is roughly about The system is a portable handheld device that decodes the
0.5 to 1 meter. radio frequency (RF) signal from the transmitter attached in
the bus and playback the route number of the arriving bus.
Keywords: Embedded Controller, Visually impaired,
The transmitter works automatically in sending signals by
Information systems
adding a microcontroller that generate information signals,
I. Introduction which encodes and modulate by the transmitter to transmit
the information signal at all the time. The receiver is a
Visually challenged is a person who is partially sighted
hand-held device for the visually challenged that receives
or incapable of sight which makes them impossible to see
and decodes the transmitted signal from the public bus.
around and observe the surrounding environment by using
The voice playback system is activated for replaying the
adaptions such as touching [1]. Dependence of the very next
information of the bus which stored in it through speaker or

Approaching Vehicle Identification for the Visually Challenged
earphone. This system consist of TDA7000 single chip FM For the playback of the information as audio, the
receiver, signal activate relay switch, and voice playback software code for the controller is programmed to generate
system. Figure 1 shows the block diagram of the system. signals for every twelve seconds including a two second
time delay for reset. The software programming follows the
system flow-chart as in Figure 2.
III. Design Implementation
A. FM Transmitter
The variable LC tank circuit vibrates at frequencies in the

FM radio band (88 to 108 MHz). The voice is modulated to
transmit at a frequency between 88 and 108MHz by tuning
the variable capacitor value. The variable capacitor and self-
Fig. 1 Block Diagram of the System
made inductor constitute a parallel LC circuit that vibrates
at a resonant frequency to be picked up by the pocket FM
The PIC16F84A microcontroller device generate the radio. The transmitter operating carrier frequency is shown
signal that is encoded and modulated with carrier frequency in Figure 4.
by the FM transmitter and transmit to a small, hand held,
wireless receiver which carried by the visually challenged.
The signal will be received by the FM receiver which
tuned as same as carrier frequency of the FM transmitter.
The FM receiver decode and demodulate the signal from
carrier frequency and the demodulated signal (information
signal) activate the relay switch and the ISD 1110 announce
the information which stored in the ISD 1110 voice record
and playback chip through the speaker or earphone. This
enables visually challenged persons to identify, catch and
board the bus with the same ease, convenience and safety Fig. 3 Transmitter Device
of the average commuter. The software code designed for

PIC16F84A is according to the Voice playback duration.
Fig. 4 FM Transmitter at 89.28MHz
B. FM Receiver
The TDA7000 chip is a FM receiver on a chip that

receives frequency from 70 up to 120MHz. With this small
receiver it is possible to pickup 88-108 MHz FM band. The
demodulated receiver signal carrier frequency is shown in
Fig. 2 Software System Flow Chart Figure 6.

M.C. John Wiselin, C.V.Aravind, R. Rajparthiban and Uthaya Kumar
multilevel storage technology. Voice and audio signals are

stored directly into memory in their natural form, providing
a high-quality, and a solid-state voice reproduction.
IV. Conclusion
A novel embedded controlled handheld device for the

visually challenged person is developed. Communication
between the incoming vehicle and visually challenged
could be established through the use of FM transmitter and
receiver device. The developed system is tested with the
bus as the most common public transport. The automatic
Fig. 5 Receiver Device
signal generation through PIC16F84A microcontroller chip
established the signal generating process. The developed
FM transmitter modulates the signal with produced
carrier frequency at 89.28MHz. The FM receiver picks
the signal to demodulate and activate the voice record
and playback system device to playback the pre-recorded
stored information with the support of signal activate relay
switch to turn ON and OFF. This designed and developed
system able to identify a bus ID for visually challenged.
The carrier frequency which used for transmission is
89.3MHz, the unoccupied FM channel which identified by
trimming the FM radio for empty carrier frequency. The
Fig. 6 Demodulated signal waveform possible distance for FM wireless communication of the
system between the transmitter and receiver is about 0.5 to
C. Signal Activated Relay Switch
1 meter long. However the developed system for a single
The signal activated relay switch allows controlling the approaching vehicle and further scope is in the development
ON & OFF of any electrical appliances by analog signal of information system for many incoming vehicles.
such as tone and is specially designed to filter out ambient
References
noise and respond only analog signal sound. The analog
[1] Economic Development/E-Gov: Wireless Signage Project for
signal sensed and then amplified to activate the relay to ON
Visually Impaired Seattle Travelers.
& OFF.
[2] Role of object identification in sonification system for visually
D. Information Storage Device impaired, TENCON 2003. Conference on Convergent Technologies
for Asia-Pacific Region, Vol. 2, No.2, 15-17 Oct. 2003, pp. 735
The ISD1110 is the voice record/playback device used 739.
for information storage providing high quality, single- [3] Bus Identification system for the Vision Impaired, UQ Innovation
chip record/playback solutions to 10 second messaging Expo 2003.
applications. The CMOS device includes an on-chip

oscillator, microphone preamplifier, automatic gain control,
anti-aliasing filter, smoothing filter and speaker amplifier.
Recordings are stored in on-chip nonvolatile memory cells,
providing zero-power message storage. This unique, single-
chip solution is made possible through ISDs patented

ISSN: 2249 - 6297 Vol. 2 No. 2, 2013, pp. 25-30
Fixed Frequency Sliding Mode - PI Control for Single Phase

Unipolar Inverters
M C John Wiselin1 and Riyaz.A.Rahiman2
1
Academic Director, Travancore Engineering College, Oyoor, Kollam District, Kerala, India
2
Assistant Professor, ACE College of Engineering, Trivandrum, Kerala, India
E-mail: wiselin16@yahoo.com
(Received on 05 June 2013 and accepted on 10 July 2013)
Abstract - Sliding Mode - PI controller is recognized as robust voltage distortion. For inverters with 50-Hz output voltage
controller with a high stability in a wide range of operating frequency and its switching frequency higher than 2 kHz,
conditions, although it suffers from chattering problem. In low-frequency harmonics (2nd to 13th) should be rejected by
addition, it cannot be directly applied to multi switches power a closed-loop controller perfectly. Moreover, the controller
converters. This paper concerns a sliding controller design with must perform a good regulation of the output voltage against
the proportional-integral (PI) control form. First, a theoretical the abruptvariations of the input voltage, output current, and
analysis of the considered sliding PI controller is provided. Our the reference voltage. These demands imply to use a fast
analysis reveals that the proportional control term of the sliding controller with good dynamic responsefor the inverters.
PI controller, i.e., sliding proportional controller, can reduce the
chattering problem and thus, the robustness can be established.
Second, the chattering problem is eliminated by smoothing
the control law in a narrow boundary layer, and a pulse width
modulator produces the fixed frequency switching law for the
inverter. The smoothing procedure is based on limitation of
pulse width modulator. Therefore, the chattering problem of the
proportional gain is resolved and the valuable robust control
property of the sliding integral controller is illustrated again.
Fig. 1 PWM-based single-phase inverter.
The simulation model has been developed and tested using
MATLAB software. Table I ieee Standard 1547 for Maximum Acceptable Harmonicvoltage
Distortion
Keywrods: Pulse Width Modulator, Sliding PIControl, Unipolar

Single Phase Inverter
I. Introduction
Recently, many control methods, like repetitive control,
Now-a-days, single-phase pulse width modulation(PWM)- deadbeat control, multi loop feedback, hysteresis current
based inverter (see Fig. 1), which is used in uninterruptible mode control, and internal model control,have been proposed
power supply (UPS), should supply nonlinear and critical to achieve the aforementioned demands. Occasionally,
step loads. Since the inverter output impedance is not zero, nonlinear observer and harmonic elimination techniques are
these loads can deform the sinusoidal output voltage of the employed to improve the transient response. It is noticeable
inverter. According to the IEEE Standard 1547, the total that these control methods are based on average model (small
harmonic distortion (THD) of the output voltage must be signal model) of the inverter, because the inverter state space
less than 5%, especially for nonlinear load. Table I shows equations vary when the switches state changes. In this
the standard details for maximum acceptable harmonic model, discontinuous control quantity, duty ratio of switches,

M C John Wiselin and Riyaz A.Rahiman
is approximated and modelled by a continuous variable over III. SMC

a number of switching cycles. Thus, it can only well describe
One effective control tool compliant with the switching
the system behaviour close to its operating points. In fact,
nature of the inverter is represented by SMC, which is derived
the instantaneous behaviours of the inverter are ignored in
from the variable structure system theory. This control
the average model and the model is only valid around the
method has individual advantages for control of converters
operating point.
such as
II. Existing Method
1) Stability against severe variations of load and the line;
The inherent switching nature of power converters is 2) Robustness;
compatible with sliding mode control (SMC) feature. The
3) Good dynamic response;
SMC is well known for its robustness, stability, and good
4) Very simple implementation.
regulation properties in a wide range of operating conditions;
moreover, it is not necessary to use average model of the Based on SMC theory, a discrete control law can be
inverter. According to the SMC theory, the main subject of the defined for each system in which the system states X follow
power converter control (with an inherent switching action) desired states Xd.
is the definition of a good sliding surface and switching
condition to guarantee the output voltage regulation with low The discrete control law is as follows:
THD in different conditions. U= sign (S(X)) = +1 if S(X) > 0
In spite of the SMC excellent performance, it suffers from = -1 if S(X) < 0
chattering problem which leads to variable and high frequency Where S(X) is a scalar function andS(X)=0 is called
switching in the converter. This phenomenon increases switching or sliding surface. This function is defined as
power losses and also produces severe electromagnetic
compatibility (EMC) noise. To fix the switching frequency
against line variation for step-down dc/dc converter, Tan et (2)
al. have introduced an adaptive feed forward control scheme
where n is the system order (n _= 1) and is a strictly
that varied the hysteresis band in the hysteresis modulator of
positive constant. According to Lyapunov theorem, to ensure
the SM controller in the event of any change of the line input
the stability of the system, the following inequality should be
voltage.
established
In addition to chattering problem, discontinuous control
while is a strictly positive constant:
law (switching condition) that is generated by a sign function
or hysteresis modulator in the SMC is only suitable for single (3)
switch converter such as buck converter due to two stable
This means that
states of the sign function output. Consequently, the SMC
method could not be applied to multi switch converters such (4)
as the single-phase inverters, directly. At least, four stable
states are needed toobtain suitable switching pulses for
unipolar single phase.
In this paper, based on SMC theory, a fixed frequency and

high-performance controller is proposed to apply to unipolar If conditions (4) or sliding conditions are established,
single-phase inverters. A pulse width modulator is employed starting from any initial condition, the discontinuous control
to fix the switching frequency and to generate the suitable law u makes the state trajectory to reach the sliding surface
switching law for the four switch inverter. in a finite time, and then slides along the surface towardsXd

Fixed Frequency Sliding Mode - PI Control for Single Phase Unipolar Inverters
exponentially. In the slidingmode, the system dynamic respectively. Assuming ideal elements, the state equations of
response is represented by S(X) = 0. the inverterare
IV. Unipolar Single-Phase Inverter
There are two popularPWMtechniques (see Fig. 2), (6)

applying to single-phase inverters, unipolar and bipolar
PWM. The unipolar PWM employs +VDC and zero to
provide positive outputs, and VDC and zero to provide where inductor current iLand output voltage Voutare
negative outputs, but the bipolar PWM only uses+VDC and selected as state variables. uis the discontinuous input of the
VDC to make either positive or negative outputs. Hence, system. It is 0 or 1 to provide positive output and, 0 or1
carrier harmonic content in the bipolar scheme is twice of to provide negative output. In addition, ic,io, and R are the
unipolar scheme. Notice that, in the unipolar technique, the capacitor current, output current, and load, respectively. To
output voltage does not contain even harmonics and also the implement the SMC along with PI for the inverter, it is more
first high-frequency harmonic appears around twice of the convenient to use a system description, which involves the
switching frequency and not at the switching frequency as output error and its derivative.
it would have been in the case of bipolar PWM. Because
of these three attractive features of the unipolar PWM, we (7)
develop unipolar PWM in this paper.
whereVrefis the reference voltage. In the inverter, the
output voltage is forced to be equal to Vrefby appropriate
switching.Considering continuous current mode operation of
the inverter and, selecting the e and e as state variables, the
system equations in terms of the state variables x1 and x2 can
be rewritten as follows
(8)
Therefore,

(9)
The SMC for single phase unipolar inverters and the non-
linear load is shown in Fig 3 and Fig 4.The values of the
components are choose based on the IEC 62040-3 standard .
Fig. 2 (a) Inverter output voltage. (b) Bipolar PWM.
(c) Unipolar PWM.
The ac output voltage of the inverter can be written as:
Vout = VDC m Sin (2fo t) (5)
whereVout, VDC, m, and forepresent the output voltage,

the input voltage, modulation factor, and output frequency, Fig. 3 Existing Controller for single phase inverters

Fig. 4 Nonlinear load according to IEC62040-3 standard.

Fig. 6 Total Harmonic Distortion for 6-KVA nonlinear load.
V. Simulation Result
VI. Proposed System
The existing control method has been simulated by
Experimental results show that fixed frequency sliding
SimulinkToolbox in MATLAB for an inverter whosemain
mode control results in variations in dynamic responses.
characteristics are mentioned in Table II, in which fs and fr are
It leads to significant overshoot, long settling time and
switching frequency and cut-off frequency, correspondingly.
obnoxious oscillations before the output of interest settles
Controller parameters of the simulated inverter are listed
down at the desired value. So in order to have a better settling
in Table III. V is selected considering the electronic
time and to reduce oscillations a new control method called
circuitslimitation. The simulation results is shown in fig 5
Sliding Mode-PI Controller is introduced.
and fig 6.
VII. PI Current Regulator Design
Table II Simulated Inverter Characteristics
Conventional PI control theory is widely used in designing

control systems for industrial applications. This is due to its
simplicity and low cost. PI control deals predominantlywith
linear systems having constant parameters and being
regulated at fixed operating points. Unfortunatelyin practical
situation, however, inherent nonlinearities are unavoidable in
unipolar inverters. When these nonlinearities or factors are
introduced, the linear constant coefficient approximation will
Table III Controller Parametersa not always be satisfactory.
In this work we do not intend to linearize the plant since the

load is inherently nonlinear. So conventional pole-placement
method is not suitable in designing a PIcontroller of the
non-linear load. For a certain operating point trialand-error
method is used to find the optimum controller parameters.
PWM signals are used to produce a controlled voltage to
feed the single phase inverter.The duty cycle K of the PWM
is proportional to controller output C(t). The controller is
expressed as
e(t) = i(t) - i * (t) (10)

(11)
Fig.5 Simulation result.Output voltage and
current at 6-kVA nonlinear load.
The controller parameters Kp, and Ki, are only optimumat

Fixed Frequency Sliding Mode - PI Control for Single Phase Unipolar Inverters
a certain operating point in the sense of fast settlingtime, IX. Simulation Results of Proposed Work
small overshoot and small torque ripples. When the operating
point changes, Kp and Ki, should also change accordingly to
achieve the optimum performance.
VIII. Sliding Mode - PI Control Design
Controller parameters are only suitable for a certain

operating point. When the operating point changes,
performance deteriorates. It also shows poor capability
of tolerating disturbances. Therefore, we conclude that PI
control is unacceptable for high performance non-linear
loads. Fig. 8 Simulation result.Output voltage and current at
6- kVA nonlinear load.
The sliding mode control, however, has the disadvantages
such as the chattering and steady state error, which is
inferior to PI control.To integrate the advantages of the two
schemes above, the sliding mode-PI control scheme can be
designed as shown in Fig.7. When inverter is operating far
away from the stable point,the controller is sliding mode
controller, in order to acquire good dynamic characteristic.
On the other hand, when the inverter is operating near to the
stable point, the controller will switch to PI controller, in
order to acquiregood precision.
Fig. 9 Total Harmonic Distortion for 6-KVA nonlinear load.
X. Conclusion
In this work, a sliding controller design with the

proportional-integral (PI) control form is presented. First, a
theoretical analysis of the considered sliding PI controller and
Fig. 7 Existing Controller for Single Phase Inverters. 1.7% at maximum linear and nonlinear load, respectively.
Table IV Comparison of Inverter Characteristics: Published and This Work

The simulation and experimental results show that the load References
regulation is about 1% at the steady state as well. With [1] K. Zhang, Y. Kang, J. Xiong, and J. Chen, Deadbeat control of PWM
this compensator, the load regulation was measured which inverter with repetitive disturbance prediction, in Proc. 14thAnnu.
has been below 0.2%. Table IV summarizes the inverter Appl.Power Electron. Conf. Expo., 1999, pp. 10261031.
characteristics and compares them with previous works. As [2] N. M. Abdel-Rahim and J. E. Quaicoe, Analysis and design of a
multiple feedback loop control strategy for single-phase voltage-
can be seen from Table IV, the proposed inverter shows an
source ups inverters, IEEE Trans. Power Electron., vol. 11, no. 4, pp.
impressive performance that is better than or comparable 532541, Jul. 1996.
to that of the previous works. Although THD of output [3] P. A. Dahono and E. Taryana, A new control method for single-
voltage in [9] is better than this work, its switching frequency phasePWM inverters to realize zero steady-state error and fast
response, inProc. 15th Int. Conf. Power Electron. Drive Syst., 2003,
is relatively higher and it is variable from 20 to 40 kHz.
pp. 888892.
Moreover, in spite of high switching frequency, it has been
[4] Y. T.Woo and Y. C. Kim, A digital control of a single-phase ups
used a large output capacitor. Cut-off frequency of the output inverterfor robust ac-voltage tracking, in Proc. 30th Annu. Conf.
filter is about 520 Hz in [9], while for this work is about 2800 IEEE Ind.Electron. Soc., Busan, Korea, Nov. 26, 2004, pp. 1623
1628.
Hz.
[5] A. A. Ahmad, A. Abrishamifar, and S. Elahian, Fixed frequency
slidingmode controller for the buck converter, in Proc. 2nd Power
Electron.,Drive Syst. Technol. Conf., Tehran, Iran, Feb. 2011, pp. 557
561.
[6] W. Guo, S. Duan, Y. Kang, and J. Chen, A new digital multiple

feedback control strategy for single-phase voltage-source PWM
inverters, in Proc.Int. Conf. Elect. Electron. Technol., Aug. 2001, vol.
2, pp. 809813.
[7] A. M. Salamah, S. J. Finney, and B. W. Williams, Single-phase

voltage source inverter with a bidirectional buckboost stage for
harmonic injectionand distributed generation, IEEE Trans. Power
Electron., vol. 24, no. 2, pp. 376387, Feb. 2009.
[8] AdibAbrishamifar, Ahmad Ale Ahmad, and Mustafa Mohamadian,

Fixed Switching Frequency Sliding Mode Control for Single-Phase
Unipolar Inverters, IEEE Trans. on power electronics, vol 27,no.2,
May2012.
[9] Yang Haiqing, Sanjib K. Panda, Liang Yung Chii, Performance

Comparison of Sliding Mode Control with PI Control for Fourquadrant
Operation of Switched Reluctance Motors, IEEE Trans. Power
Electron., vol. 14, pp. 381-387, May 1999.

ISSN: 2249 - 6297 Vol. 2 No. 2, 2013, pp. 31-37
Retinal Blood Vessels Segmentation Using the Curvlet Transform

Academic Director, 2Head of the Department,
1
Department of EEE, Travancore Engineering College, Oyoor, Kollam District, Kerala, India
E-mail: wiselin16@yahoo.com
(Received on 10 June 2013 and accepted on 18 July 2013)
Abstract - Retinal image having very vital information. It plays About 10% of all diabetic patients have diabetic
important roles in finding of some diseases in early stages, such retinopathy, which is the primary cause of blindness
as diabetes, and cardiovascular disease. In this proposed system in the Western World. Since this type of blindness can
a new algorithm used to detect the blood vessels effectively from be prevented with treatment at an early stage, the who
the retinal image. The initial image enhancement is carried advises yearly ocular screening of patients Other indirect
out by using Adaptive Histogram Equalization, followed by applications include automatic generation of retinal maps
for the treatment of age-related macular degeneration,
the curvelet Transforms are applied to the equalized image
extraction of characteristic points of the retinal vasculature
and the curvelet coefficients are obtained. The modifications
for temporal or multimodal image registration, retinal image
to the Curvelet transform coefficients are carried out by
mosaic synthesis, identification of the optic disc position,
suppressing all the coefficients of one band. This combined
and localization of the fovea. Furthermore, the network of
effect of the equalization and the Curvelet Transforms provides
retinal vessels is distinctive enough to each individual and
a better enhancement to the image. This enhanced image is can be used for biometric identification, although it has not
used for the extraction of blood vessels. Afterward, eliminate yet been extensively explored [2].
the ridges not belonging to the vessels tree by morphological
operators by reconstruction while trying to preserve the thin
The three most important structures of the human retina
are Vessels, fovea, and optical disk, these structures
vessels unchanged. In order to increase the efficiency of the
mostly used for several applications such as retinal image
morphological operators by reconstruction, they were applied
registration, illumination correction, as well as pathology
using multi-structure elements and local adaptive thresholding
detection inside the retina. Detection of these important
method along with connected components analysis (CCA)
structures manually is time consuming and depends on the
indicates the remained ridges belonging to vessels. expertise of the user. As compare to all three vessels play
vital role for apply many application. The segmentation of
Keywords: Blood vessel segmentation, connected component
blood vessels from fundus photographs can be difficult for
analysis, curvelet transform, multistructure elements
a number of reasons. The two most influential factors that
morphology, retinal image
make the segmentation difficult are the improper retinal
I. Introduction image contrast and the uneven background illumination.
The uneven illumination is from the acquisition process.
Digital fundus imaging in plays an important role in The different vessels have different contrast with
medical diagnosis of primary levels of diabetes and blood background. In other words, arteries have higher contrast
pressure as well as cardiovascular disease[1] .Some of than veins. Thick vessels also display a higher contrast with
the main clinical objectives reported in the literature for the background than do thin ones. Existing paper also has
retinal vessel segmentation are the implementation of a deficiency of missing some thin blood vessels because of
screening programs for diabetic retinopathy, evaluation of the simple thresholding method. There is a tradeoff between
the retinopathy of prematurity, cardiovascular diseases, and removing more false edges and preserving more pixels of
small vessels.
computer-assisted laser surgery.

M.C John Wiselin and A. Prabin
In this paper, existing simple threshold is replaced II. Curvelet Transform

by local adaptive threshold method; this is new proper
A. Curvelet transform
approach threshold method. A method based on using
curvelet transform is proposed to enhance and prepare Candes et al.[3] introduced a new multi-scale transform,
the retinal image for better vessel detection. In the past Curvelet transform. It can efficiently represent the edges
decade, introduced the curvelet transform, a new multi scale along curves than the traditional transforms and also
transform. The second generation of curvelet transform, allows an optimal sparse representation of objects with
is faster and simpler than the first version. Therefore, the C2 singularities the best m-term approximation, fm for a
second generation of curvelet transform, discrete curvelet smooth object, f with C2 discontinuities along a generic
smooth curve using Curvelet approximation, fm achieves
n
transform (DCT), and modified the DCT coefficients by a
suitable nonlinear function are used. One way to increase
the image contrast is to enhance the image ridges, which (1)
play an important role in enhancing image contrast. In order
to simultaneously enhance the weak edges and eliminate the B. Digital Curvelet Transform(DCT)
noise, the modifying function parameters are defined based There are two methods to implement the second
on some statistic features of fast DCT (FDCT) coefficients. generation DCT: Wrapping method and Unequis paced Fast
Mathematical morphology using multi structure elements Fourier Transform (USFFT) method. Both methods are
are applied to obtain the image ridges. Then, morphological described detail in [3].
opening by reconstruction helps to remove the detected For the same accuracy, the wrapping method is faster and
ridges not belonging to the vessel tree while preserving easier to implement than USFFT method. Here, we use the
the thin vessel edges. The morphological opening by wrapping method in this paper. The architecture of Fast
reconstruction benefits from using multistructure elements, Discrete Curvelet Transform (FDCT) [4] via wrapping is
which helps to improve the performance of this step. There is as follows:
a restriction on size of structure elements (SEs) concerning
the blood vessels diameter. Therefore, the remaining false 1. Apply the 2D FFT and obtain Fourier samples
edges will be removed by means of connected components n n
f [n1 , n2 ], n1 , n2 < (2)
analysis (CCA) along with length filtering. In order to act 2 2
locally, image is decomposed to several tiles and CCA, and 2. For each scale and angle, form the product
length filtering is individually applied to each tile. Modified
(3)
CCA is proposed to predict the length of the blood vessels
dynamically. A result shows a promising performance Here, U j ,l [n1 , n2 ] is the discrete localizing window as
in segmentation of the blood vessels. The quantitative said in the reference [5]
performance results of both segmentation and enhancement
steps show that our method effectively detects the blood 3. Wrap this product around the origin and obtain
vessels with better accuracy. fj ,l [n1 , n2 ] = W (U j ,l , f )[n 1 , n2 ] (4)
The rest of the paper is organized as follows. In Where the range for n1 is now 0 n1 < L 1, j jand 0
Section II, curvelet transform is briefly described. Then, n2 < L 2, j j; L 1, j ~ 2 j and L L 2, j ~ 2 2 are
multistructure elements morphology and morphology constants.
operators by reconstruction elaborated in Section III.
Thereafter, the proposed method is explained in Section IV 4. Apply the inverse 2D FFT to each fj ,l hence collecting
the experimental results are illustrated in Section V. Section the discrete coefficients
VI concludes the works described in this paper.

III. Mathematical Morphology

g S = (f rec ) ( g S ) (6)
A. Multistructure Elements Morphology
(frec ) ( g S )
gS = (7)
Since the size, shape, and direction of the SE determines
the final result of morphological processing, the choosing Therefore, morphological opening by reconstruction in
of SE is a key factor in morphological image processing. its first step eliminates bright features smaller than the SE.
Single and symmetrical SEs are normally selected in order in the next step, it dilates iteratively to restore the contours
to perform the morphological processing; such SEs are of components that have not been completely removed by
successful in detecting ordinary, simple, and straight edges opening and it is performed by considering the original
of an image. While the edges complexities increase, their image as the reference. In a similar manner, closing by
success in detection of such complex edges decrease. In order reconstruction is accomplished in case of dark features.
to deal with such edges and achieve a reliable precision, we Therefore, as a valuable result, producing new edges, edge
need more advanced SEs. The basis of the multistructure drift and deforming the contours and edges, which often
elements morphology theory relies on gathering several occur by applying conventional morphological opening and
SEs in one square window. In other words, decomposing an closing will not appear by applying opening and closing by
SE produces the si . reconstruction.
Therefore, such SE is capable of detecting different edges IV. Proposed Method

with different directions, efficiently.
A. Image Selection
Let {I ( m, n) | m, n Z } be a digital image; an SE in
The images used here for the experimentation are
(2N +1) (2N + 1) square window is defined as follows:
obtained from the well known DRIVE database [8] . The
Si ={I (m + m0 , n + n0 ,i =i | N m0 , n0 N } (5) gray scale version of these images is used. This grayscale
coming from the green channel of the colored image

0
where i = 0, 1, . . . , 4N 1, = 180 4N , and i is the because blood vessels in the green channel image of the
direction angle of Si . With regard to the required directional original colored retinal image have the highest contrast. The
resolution , the size of the SE windowis determined. Si for blue channel tends to be empty and the red channel tends to
0
= 15 and the 7 7 SE built from integration of all Si of be saturated.
all directions.
B. Retinal Image Enhancement using Adaptive Histogram
B. Morphological Operators by Reconstruction Equalization
Both morphology closing and opening leave the features Adaptive Histogram Equalization is an enhancement
larger than SE unchanged. However, the main drawback technique capable of improving an images Local Contrast.
of conventional opening and closing is that they do not It differs from ordinary histogram equalization in the
preserve edge information perfectly. Bangham et al. [6] respect that adaptive method computes several histograms
introduced a new operator to address this defect, called each corresponding to distinct section of the image and uses
M- and N-sieves. The new operator emphasizes only on them to redistribute the lightness values of image. It acts as
the size of the features, but rejects the shape completely. a good tool for the enhancement of the edges. Comparison
Morphological operators by reconstruction are designed of the input and the enhanced image is shown in Figure
to address this problem and consider both shape and size (fig). 1
of the features [7]. Opening by reconstruction and closing
by reconstruction are denoted by g S and g S ,
respectively, and defined as

1. Applying FDCT via wrapping method, we get a set of

scales S j , each scale consists of a set of directional
bands Di containing coefficients.
2.. For each directional band in each scale Dji, do the

following:
a) Calculate the noise standard deviation ij ;
b) Determine the value of m.
(a)
3. Multiply each coefficient individually by corresponding
y.
4. Reconstruct the enhanced image using modified

curvelet Coefficients.
(b)
Fig.2 FDCT Image
D. Edge Detection Using Multistructure Elements

Morphology
(c)
In order to perform edge detection using multistructure
Fig.1 (a) Original Image (b) Gray Scale image
elements morphology, the earlier SE of morphological edge
(c) Adaptive Histogram Equalized image
detector should be replaced by new introduced SE and
C. Vessels Contrast Enhancement Using FDCT follow the following algorithm.
1) Produce the proposed SEs Si with regard to the required

The curvelet transform is well adapted to represent the
directional resolution.
images containing edges; it is a good candidate for edge
enhancement. Curvelet coefficients can be modified to 2) Apply the selected edge detector function F on the
enhance the edges in an image, which improves the image original image using the achieved SEs in 1 and get the
contrast. The aim of enhancement step is enhancing the sub edge image F(I)i .
thin vessels having low contrast to detect better in the edge 3) Put the F(I)i obtained in 2) in the following equation to
detection step. Here contrast enhancement is carried out by achieve the whole of detected edges:
means of Adaptive Histogram Equalization. It is a technique
M 1
by which image brightness changes sharply.
F ( I ) i F ( I )i (8)
i =0
Consequently, our proposed method to enhance the
retinal image consists of following steps. where F(I) is the total edge image, M = 180/ is the
number of Si and i is the assigned weight to each of
subedge image.

In order to have the same effect of each F(I)i , the assigned F. Length filtering with local adaptive threshold method
weights can be defined as i = 1/M, or they can be calculated
In order to obtain a clear final result without presence of
by other methods [9] as well. Also, if any information about
pixels that do not belong to vessel tree, we use length filtering
the processed image exists, the weights can be assigned
with the aim of removing the small pixel blocks. In this
according to the degree of importance of information that
case, the concept of modified CCA is used where connected
may exist in each of F(I)i . we calculate the weights by
components pixels which are identified above a specific
F ( I )i (9) threshold and labeled using eight connected neighborhood
i =

M 1 and are considered as a single object. Modified CCA is used
i 0
F ( I )i
to predict the length of the blood vessels dynamically here
E. False-Edges Removal Using Morphological Operators the threshold value is automatically calculated. Considering
by Reconstruction the entire image in CCA and length filtering with simple
threshold leads to inferior results. A kind of adaptive CCA,
In the result image of edge detection step, there are
that is consider images in separate tiles and apply CCA and
edges not belonging to blood vessels but that arise from
length filtering to each tile, individually. By this means,
uneven background illumination. A simple method to
there is no large range of gray levels in each block, and a
eliminate these undesired objects is using morphological
proper threshold can be chosen which separates the false
opening by reconstruction. Opening by reconstruction
edges from vessel edges efficiently. After applying modified
includes two steps: conventional morphological opening
CCA, all the small length blood vessels are identified.
and reconstruction by dilation. In order to improve
Finally, all of the results are integrated in a single image as
the performance of the morphological opening by
the final blood vessel detection result.
reconstruction, the opening using multi structure elements
is performed.
Since the multi structure elements are highly sensitive

to edges in all directions, it helps to accurately eliminate
the false edges. The SE used in this step is the same as in
the edge detection step. The only difference is in assigned
weight. Here, instead of assigning weights to each F(I)i ,
the maximum F(I)i is selected to construct the F(I). This
method allows us to eliminate the weak false edges and
prevent them from participating in construction of F(I).
Fig.4.a) Input Retinal image
(a) (b)
Fig.3 (a) Edge Detection image Using Multistructure Elements

Morphology (b) Edges Removed image Using Morphological Operators
b) Extracted blood vessels by simple threshold method

M.C John Wiselin and A. Prabin
Still There is a tradeoff between removing more false V. Experimental Results

edges and preserving more pixels of small vessels only We implemented our proposed method using MATLAB
because of simple threshold instead of that we use the local 7.1. We take DRIVE database image as an input image. As
adaptive threshold method. it selects an individual threshold mentioned earlier, the green channel is the best choice to
for each pixel based on the range of intensity values in its be processed. In the next step, the fundus region disk was
local neighborhood
Therefore, the algorithm is as follows.
1) Partition image into tiles of N N pixels with 50%

interpolation to avoid windowing effect.
2) Apply the described thresholding algorithm to each

part individually and obtain the desired threshold of
each tile.
3) Apply CCA to each tile with considering only the pixels

whose gray levels are more than the corresponding
threshold.
4) Apply length filtering to each tile individually and

retain the components having length larger than the
elements morphology was capable of detecting the
corresponding threshold.
5) Gather all the results in one image.
(a)
Fig. 6 Results of proposed method. (a),(b) are related to images 16 and18

of DRIVE database, respectively. From top to bottom they are the green
channel, result of enhancement using FDCT, result of edge detection and
last row are result of morphology operators by reconstruction and length
(b) filtering,respectively.
Fig.5 a) Extracted blood vessels, b)Extracted blood vessels
using local adaptive threshold

produced. Afterward, the FDCT was applied to the selected [5] E. J. Cands and D. L. Donoho, 1999 Curvelets A Surprisingly
Effective Nonadaptive Representation For Objects with Edges,
channel. This co-efficient is modified. hence the image
Curves and Surfaces.
vessels. Then, the edges of image were detected by modified
[6] J. A. Bangham, R. Harvey, P. D. Ling, and R. V. Aldridge,
top hat using the multistructure elements morphology. We Morphological scale-space preserving transforms in many
used a 7 7 SE with = 150. During edge detecting some dimensions, J. Electron. Imag., vol. 5, no. 3, pp. 283299, Jul.
1996.
of the undesired objects also found. This false edges are
removed by the morphological opening by reconstruction. [7] S. Mukhoopadhyay and B. Chanda, Multiscale morphological
segmentation of gray-scale images, IEEE Trans. Image Process.,
Finally the image is decomposed to several blocks and vol. 12, no. 5, pp. 533549, May 2003
length filtering, and CCA are applied locally in order to
[8] Available at: http://www.isi.uu.nl/Research/Databases/DRIVE/.
better remove the remained false edges,The threshold value
{9] Y. Ma,M. Yang, and L. Li, A kind of omni-directional multi-angle
is find by local adaptive threshold method. structuring elements adaptive morphological filters, J. Chin. Inst.
Commun., vol. 25, no. 9, pp. 8692, 2004.
VI. Conclusion
[10] Ch. Wu, G. Agam, and P. Stanchev, A general framework for
vesselsegmentation in retinal images, in Proc. IEEE Int.Symp.
In this paper, a new most efficient method for the
Comput. Intel.Rob. Autom., Jun.2007, pp. 3742.
retinal vessel segmentation has been presented; all sections
[11] E. J. Candès and D. L.Donoho, CurveletsA surprisingly
are upgraded with efficient technique such that adaptive effective nonadaptive representation for objects with edges, in
histogram in image enhancement with curvlet transform, Curves and Surfaces. Nashville, TN: Vanderbilt Univ. Press, 1999,
pp. 123143.
Modified top hat transformation is used as an edge detector
in edge detection and local adaptive threshold technique [12] Y. Zhao, W. Gui, and Zh. Chen, Edge detection based on multi-
structure elements morphology, Proc. 6th World Congr. Intell.
in threshold method. All these modifications are combined Control Autom., pp. 97959798, 2006.
and moulted as a most efficient one. And it has concluded
[13] S. Mukhoopadhyay and B. Chanda, Multiscale morphological
that proposed technique is giving much better results than segmentation of gray-scale images, IEEE Trans. Image Process.,
existing ones. vol. 12, no. 5, pp. 533549, May 2003.
[14] Zh. B. Zhao, J. Sh. Yung, O. Gao, and Y. H. Kong, Wavelet image
Hence, our future work is modifying this threshold denoising method based on noise standard deviation estimation,
method with a more proper approach to increase accuracy in Proc. Int. Conf. Wavelet Anal. Pattern Recog., 2007, pp. 1910
1914.
of the method and deal with the problem of presence some
[15] J. Starck, F. Murtagh, E. J. Candès, and D. L. Donoho, Gray and
severe lesions in retinal images.
color image contrast enhancement by the curvelet transform, IEEE
Trans. Image Process., vol. 12, no. 6, pp. 706717, Jun. 2003.
References
[1] S. Dua, N. Kandiraju, and H.W. Thompson, Design and
implementation of a unique blood-vessel detection algorithm towards
early diagnosis of diabetic retinopathy, in Proc. IEEE Int. Conf. in
Inf. Technol., Coding Comput., 2005, pp. 2631.
[2] A.M.Mendonca and A. Campilho, Segmentation of retinal blood

vessels by combining the detection of centerlines and morphological
reconstruction, IEEE Trans. Med. Imag., vol. 25, no. 9, pp. 1200
1213, Sep. 2006.
[3] E. Candès, L. Demanet, D. Donoho, and L. Ying, Fast discrete

curvelet transforms, Multiscale Model. Simul., vol. 5, no. 3, pp. 861
899, 2006.
[4] Stephen N.Pizer, E. Philip Amburn, John.D.Austin, Robert

Cromatie, Ari Geselowitz, Trey Greer, Brat Ter Haar Romeny,
John.B.Zimmerman and Karel Zuiderveld ,Retinal Image Analysis
using Curvelet Transform and Multi structure Element Morphology by
Reconstruction. Mohammad Saleh Miri and Ali Mahloojifar.

ISSN: 2249 - 6297 Vol. 2 No. 2, 2013, pp. 38-44
Hybrid Method for Retinal Image Segmentation and

Identifying True Vessels
A.Prabin1, J.Veerappan2, M.C. John Wiselin3, S.Binu Sathiya4 and Parvathy R Nair5
1
Assistant Professor, Immanuel Arasar JJ College of Engineering, Nattalam, Tamil Nadu, India
2
Professor, Sethu Institute of Technology, Virudhunagar, Tamil Nadu, India
3
Professor, Travancore Engineering College, Oyoor, Kerala, India
4
Assistant Professor, NI University, Thuckalay, Tamil Nadu, India
5
Cape Institute of Technology, Levinchipuram, Tamil Nadu, India
E-mail:wiselin16@yahoo.com
(Received on 20 July 2013 and accepted on 10 August 2013)
Abstract - As digital imaging and computing power increasingly

I. Introduction
develop, so too does the potential to use these technologies in
ophthalmology. Image processing, analysis and computer vision DIABETIC retinopathy, hypertension, glaucoma, and
techniques are increasing in prominence in all fields of medical macular degeneration are now-a-days some of the most
science, and are especially pertinent to modern ophthalmology, common causes of visual impairment and blindness. Early
as it is heavily dependent on visually oriented signs. We describe diagnosis and appropriate referral for treatment of these
a novel technique that utilizes the global information of the diseases can prevent visual loss. Usually, more than 80%
segmented vascular structure to correctly identify true vessels of global visual impairment is avoidable and in the case of
in a retinal image. The model segmented vascular structure as a diabetes by up to 98%. All of these diseases can be detected
vessel segment graph and transform the problem of identifying through a direct and regular ophthalmologic examination
true vessels to that of finding an optimal forest in the graph. of the risk population. However, population growth,
An objective function to score forests is designed based on aging, physical inactivity and rising levels of obesity are
directional information. Our proposed solution employs contributing factors to increase of them, which causes
candidate generation and expert knowledge to prune the search the number of ophthalmologists needed for evaluation by
space. Each vessel is tracked individually by repeatedly finding direct examination is a limiting factor. So, a system for
the next vessel point with a scoring function that considers the automatic recognition of the characteristic patterns of these
pixel intensity and orientation in the vicinity of the current pathological cases would provide a great benefit.
point in the image. Bifurcations and crossovers are detected
Regarding this aspect, optic disc (OD) segmentation is
using some intensity profile. Tracking for the same vessel then
a key process in many algorithms designed for the lesions,
continues along the most likely path. The importance of our
and the identification of other fundus features. Change in
proposed work disambiguate between vessels at bifurcations
the shape, color or depth of the OD is an indicator of various
and crossovers, we need to figure out if linking a vessel segment
ophthalmic pathologies, especially glaucoma. Although the
to one vessel will lead to an adjacent vessel being wrongly
OD has well defined features and characteristics, localizing
identified .
the OD automatically and in a robust manner is not a
Keywords: Principal Component Analysis, Watershed straightforward process, since the appearance of the OD may
Transformation, Optic Disc segmentation, Region vary significantly due to retinal diseases and the size varies
discrimination, Circular approximation from one person to another. Detection is critical due to the
geometric relationship that exists between the vasculature
and the position of the OD in the retina. Locating the OD

Hybrid Method for Retinal Image Segmentation and Identifying True Vessels
is quite important for many reasons, since it may be easily

confounded with large exudative lesions by image analysis
techniques.
OD detection remains a problematic task due to the

discontinuities along the boundary where blood vessels
cross, dramatic hue changes within the OD boundary,
with the most extreme being the intra-disc hemorrhage.
OD segmentation is difficult since some parts of the disc
boundary are not well defined and some parts are partly
obscured by the blood vessels in the retinal image, which
makes the disc shape more complicated. General purpose Fig. 2 Fundus Photography
algorithms often fail to segment the OD due to fuzzy

Fundus photography is performed by a fundus
boundaries, inconsistent image contrast or missing edge
camera, which basically consists of a specialized low
features. Disc boundary detection should therefore be aimed
powermicroscopewith an attachedcamera. A fundus
to correctly segment the OD by detecting the boundary
camera provides an upright, magnified view of the fundus.
between the retina and the nerve head.
A typical camera views 30 to 50 of retinal area, with a
magnification of 2.5x, and allows some modification of this
relationship through zoom or auxiliary lenses from 15,
which provides 5x magnification, to 140 with a wide angle
lens, which minifies the image by half. The OD location
helps to avoid false positives in the detection of exudates
associated with diabetic retinopathy, since both of them are
spots with similar intensity. The OD margin can be used for
establishing standard and concentric areas in which retinal
vessel diameter measurements are performed by calculating
some important diagnostic indexes for hypertensive
retinopathy, such as central retinal artery equivalent (CRAE)
Fig. 1 Afundus photograph showing the macula as a spot to the left and central retinal vein equivalent (CRVE). Moreover, the
center, or even the border, of the OD also serves as initial
Themaculaormaculalutea(fromLatin macula, point for vessel tracking algorithms due to the fact that all
spot + lutea, yellow) is an oval-shaped highly retinal vessels are originated from there.
pigmented yellow spot near the center of theretina of
the human eye. It has a diameter of around 6mmand is
often histologically defined as having two or more layers
of ganglion cells. Near its center is thefovea, a small pit
that contains the largest concentration ofcone cellsin the
eye and is responsible for central, high resolution vision.
The macula also contains thepara fovea and peri fovea.
Because the macula is yellow in colour it absorbs excess
blue and ultraviolet light that enter the eye, and acts as a
natural sun block (analogous to sunglasses) for this area of
Fig.3 Probing of an image with a structuring element
the retina. The optic disc is the area on the right where (white and grey pixels have zero and non-zero values, respectively).
blood vessels converge. The grey, more diffuse spot in the
centre is a shadowartefact.

Morphological techniques probe an image with a fundus image in order to obtain a gray image in which the
small shape or template called astructuring element. The different structures of the retina such as vessels and OD
structuring element is positioned at all possible locations are differentiated more clearly to get a accurate detection
in the image and it is compared with the corresponding of the OD. Next a variant of the watershed transformation,
neighbourhood of pixels. Some operations test whether the stochastic transformation, followed to a stratified
the element fits within the neighbourhood, while others transformation are implemented on a region of the original
test whether it hits or intersects the neighbourhood. A image. The algorithm is fully automatic, so process is
morphological operation on a binary image creates a new speeded up and also the user intervention is avoided. The
binary image in which the pixel has a non-zero value only goal of this proposed method is to make easier the early
if the test is successful at that location in the input image. detection of diseases related to the fundus and reduces the
Thestructuring elementis a small binary image, i.e. a consultation time. Stochastic Watershed Transformation
small matrix of pixels, each with a value of zero or one. is used here. It is a segmentation technique for gray-scale
The matrix dimensions specify thesizeof the structuring images.
element. The pattern of ones and zeros specifies theshapeof
This algorithm is a powerful segmentation tool
the structuring element. Anoriginof the structuring element
whenever the minima of the image represent the objects
is usually one of its pixels, although generally the origin can
of interest and the maxima are the separation boundaries
be outside the structuring element. Morphological operators
between objects. Due to this the input image of this method
often take abinary image and astructuring element as
is usually a gradient image. The watershed transformation
input and combine them using a set operator (intersection,
produces a segmentation which can be viewed as a set of
union, inclusion, complement). They process objects in the
closed contours of segmented regions. This transformation
input image based on characteristics of its shape, which
uses random markers to build a probability density function
are encoded in the structuring element. The mathematical
of contours, which is then segmented by volumic watershed
details are explained inMathematical Morphology.
for defining the most significant regions.
II. Related Work
III. Preliminaries
A KNN regressor is utilized to predict the distance in
The first step of the pre processing stage consists of PCA
pixels in the image to the object of interest at any given
to transform the input image to grey scale. This technique
location in the image based on a set of features measured at
combines the most significant information of the three
that location. The method combines cues measured directly
components RGB in a single image so that it is a more
in the image with cues derived from a segmentation of the
appropriate input to the segmentation method. Principal
retinal vasculature. The location of the fovea is estimated as
Component Analysis (PCA). The central idea of PCA is
the point of lowest matched filter response within a search
to reduce the dimensionality of a data set consisting of a
area determined by the optic disc location. Second, optic
number of interrelated variables, while retaining as much
disc segmentation is performed. Based on the detected
as possible of the variation present in the data set. This
optic disc location, a fast hybrid level-set algorithm which
is achieved by transforming to a new set of variables, the
combines the region information and edge gradient to
principal components (PCs), which are uncorrelated, and
drive the curve evolution is used to segment the optic disc
ordered. This technique maximizes the separation of the
boundary.
different objects that compose a image so that the structures
The optic disc location algorithm is designed based on of the retina are better appreciated. Inpainting technique is
the characteristics of the different channels in digital colour the next step used. Retinal vessels are originated from the
fundus images. The method is used to detect the optic disc OD therefore there are numerous vessels crossing its border
contour based on mathematical morphology along with which makes its discrimination difficult. So vessel removal
principal component analysis. In the past KNN regressor of the enhanced image is implemented by the inpainting
is used mainly focused on locating the optic disc center. technique.
Principal component analysis is applied on the RGB

The main aim is to obtain the OD-boundary more The method is used to detect the optic disc contour
precisely and to reduce the existing borders within the OD based on mathematical morphology along with principal
which increase the risk of sub-segmentation. Inpainting component analysis. In the past KNN regressor is used
algorithms are used in diverse applications, from the mainly focused on locating the optic disc center. Principal
restoration of damaged photographs to the removal or component analysis is applied on the RGB fundus image
replacement of selected objects. In OD- Segmentation inorder to obtain a gray image in which the different
Stochastic Watershed Transformation is used. It is a structures of the retina such as vessels and OD are
segmentation technique for gray-scale images. This differentiated more clearly to get a accurate detection of
algorithm is a powerful segmentation tool whenever the the OD. Next a variant of the watershed transformation,
minima of the image represent the objects of interest and the the stochastic transformation, followed to a stratified
maxima are the separation boundaries between objects. Due transformation are implemented on a region of the original
to this the input image of this method is usually a gradient image. The algorithm is fully automatic, so process is
image. speeded up and also the user intervention is avoided.
The goal of this proposed method is to make easier the
The watershed transformation produces a segmentation
early detection of diseases related to the fundus and
which can be viewed as a set of closed contours of
reduces the consultation time. Histogram equalizationis a
segmented regions. This transformation uses random
method inimage processing of contrast adjustment using
markers to build a probability density function of contours,
the images histogram. This method usually increases
which is then segmented by volumic watershed for defining
the globalcontrast of many images, especially when the
the most significant regions. Region Discrimination is the
usable data of the image is represented by close contrast
discrimination between the significant and non significant
values. Through this adjustment, theintensities can be
regions is based on the average intensity of the region. The
better distributed on the histogram. This allows for areas
value of each region will be equal to:
of lower local contrast to gain a higher contrast. Histogram
equalization accomplishes this by effectively spreading out
the most frequent intensity values. The method is useful
in images with backgrounds and foregrounds that are both
Being N is the number of pixels of the corresponding
bright or both dark.
region . The regions belonging to the optic disc will be
light regions around darker regions therefore the residue In histogram equalization we are trying to maximize the
of a close-hole operator is calculated on to obtain the image contrast by applying a gray level transform which
regions. Afterwards, a threshold is applied on the resulting tries to flatten the resulting histogram. It turns out that the
image to select the valid regions. gray level transform that we are seeking is simply a scaled
version of the original images cumulative histogram.
In Post Processing step, the region of interest has been
That is, the gray level transform T is given by T[i] = (G-1)
obtained, the result must be fitted to eliminate false contours,
c(i), where G is the number of gray levels and c(i) is the
which are detected due to the blood vessels that pass through
normalized cumulative histogram of the original image.
the OD. The technique was performed to remove most of
them, as previously mentioned, however some irregularities When we want to specify a non-flat resulting histogram,
can still be appreciated in the final region contour. In this we can use the following steps:
work, the OD-contour has been estimated as a circle in the
1. Specify the desired histogram g(z)
same way that in although a elliptical shape could also have
been chosen. The main reason for fitting the OD by a circle 2. Obtain the transform which would equalize the
is because is algorithm will later be used to establish a zone specified histogram, Tg, and its inverse Tg-1
of the retina.
3. Get the transform which would histogram equalize
the original image, s=T[i]

4. Apply the inverse transform Tg-1on the equalized B. Principal Compound Analysis
image, that is z=Tg-1[s]
The central idea of PCA is to reduce the dimensionality
Histogram equalization often produces unrealistic of a data set consisting of a number of interrelated variables,
effects in photographs; however it is very useful for scientific while retaining as much as possible of the variation present
images likethermal,satelliteorx-rayimages, often the same in the data set. This is achieved by transforming to a new
class of images that user would applyfalse-colourto. Also set of variables, the principal components(PCs), which are
histogram equalization can produce undesirable effects uncorrelated, and ordered. This technique maximizes the
(like visibleimage gradient) when applied to images with separation of the different objects that compose a image so
lowcolour depth. that the structures of the retina are better appreciated.
Watershed Transformation is the transformation used C. Inpainting

here. A watershed is a basin-like landform defined by
Retinal vessels are originated from the OD therefore there
highpoints and ridgelines that descend into lower elevations
are numerous vessels crossing its border which makes its
and stream valleys. Inimage processing, different watershed
discrimination difficult. So vessel removal of the enhanced
lines may be computed. In graphs, some may be defined on
image is implemented by the inpainting technique. The
the nodes, on the edges, or hybrid lines on both nodes and
main aim is to obtain the OD-boundary more precisely and
edges. Watersheds may also be defined in the continuous
to reduce the existing borders within the OD which increase
domain.
the risk of sub-segmentation. Inpainting algorithms are used
There are also many differentalgorithmsto computer in diverse applications, from the restoration of damaged
watersheds. For asegmentation purpose, the gradient photographs to the removal or replacement of selected
magnitude (i.e., the length of the gradient vectors) is objects.
interpreted as elevation information.
D. OD- Segmentation
Different approaches may be employed to use the
1. Stochastic Watershed Transformation
watershed principle forimage segmentation.
It is a segmentation technique for gray-scale images.
Local minima of the gradient of the image may be
This algorithm is a powerful segmentation tool whenever
chosen as markers, in this case an over-segmentation is
the minima of the image represent the objects of interest
produced and a second step involves region merging.
and the maxima are the separation boundaries between

Marker based watershed transformation make objects. Due to this the input image of this method is
use of specific marker positions which have been usually a gradient image. The watershed transformation
either explicitly defined by the user or determined produces a segmentation which can be viewed as a set of
automatically with morphological operators or other closed contours of segmented regions. This transformation
ways. uses random markers to build a probability density function
of contours, which is then segmented by volumic watershed
IV. Optic Disc Segmentation
for defining the most significant regions.
A. Pre- processing
E. Region Discrimination
The first step of the pre processing stage consists of PCA
The discrimination between the significant and non
to transform the input image to grey scale. This technique
significant regions is based on the average intensity of the
combines the most significant information of the three
region. The value of each region will be equal to:
components RGB in a single image so that it is a more
appropriate input to the segmentation method.

Being N is the number of pixels of the corresponding D. Morphological Post Processing Image
region . The regions belonging to the optic disc will be
Short segments between two junctions are not necessary.
light regions around darker regions therefore the residue
Hence we propose to use morphological post processing
of a close-hole operator is calculated on to obtain the
image.
regions. Afterwards, a threshold is applied on the resulting
image to select the valid regions.
F. Post Processing
Once the region of interest has been obtained, the

result must be fitted to eliminate false contours, which are
detected due to the blood vessels that pass through the OD.
The technique was performed to remove most of them, as
previously mentioned, however some irregularities can still
be appreciated in the final region contour. In this work,
the OD-contour has been estimated as a circle in the same
way that in although a elliptical shape could also have been Fig. 4(a) Input Image
chosen. The main reason for fitting the OD by a circle is
because is algorithm will later be used to establish a zone of
the retina.
V. Experimental Result
This frame work consisting overall project layout of

getting original input fundus image and enhancement image
as well as inpainting of vessel mask .The final interface
gives the optic disc segmentation.
A. Original Fundus Image
Variability between fundus images in color, intensity,

size, presence of artefacts, etc., makes each state-of-the- Fig.4 (b) Enhanced Image
art method uses a different input image. The validation of

the method has been carried out using different optic disk
images.
B. Enhanced Image
Enhancement analysis maximizes the separation of the

different objects that compose a image so that the structures
of the retina are better appreciated. In addition, it is much
less sensitive to the existing variability in a fundus image
regarding colour, intensity, etc.
C. Segmented Image
The discrimination between the significant and non Fig. 4(c) Segmented Image
significant regions is based on the average intensity of
the reg. the final segmentation, providing robustness and
reliability.

References
[1] A. Hoover and M. Goldbaum, Locating the optic nerve in a retinal
image using the fuzzy convergence of the blood vessels, IEEE
Trans., vol. 22, no.8, pp. 951-958, Aug 2003.
[2] D.Pascolini and S.P. Mariotti, Global estimates of visual

impairment 2010, Br.J.Ophthalmol., pp. 614-621,2011
[3] I.T.Jolliffe, Principal component analysis, 2nd ed. springer, 2002
[4] H.Li and O.Chutatape, Automated feature extraction in color

retinal images by a model based approach, IEEE Trans. Biomed.
Eng., vol.51, no.2, pp.246-254,Feb 2004
[5] J.C.Russ, Image processing handbooks, 5th ed. Boca Raton, FL.CRC
Press,2007
Fig. 4 (d) Result of segmented true vessels
[6] J.Serra, Image analysis and Mathematical morphology, London,
U.K, 2008, vol.1
VI. Conclusion
[7] M. Foracchina, E. Grisan, and A.Ruggeri Detection of optic disc in
The algorithm has been validated on five different public retinal images by means of geometrical model of vessel structure,
IEEE Trans. Med.Image., oct 2004., Med. Image., vol. 23, no.10,
databases obtaining promising results and improving the pp. 1189-1195
results of other methods of the literature. The final goal of
[8] M,Park, J.S.Jin,andS.Luo, Locating the optic disc in retinal
the proposed method is to make easier the early detection images, in proc. IEEE int. conf. comput. Graph., 2006 pp. 141-145
of diseases related to the fundus. Its main advantage is the [9] M. Niemeijer, M.D.A. Moff, Fast detection of the optic disc and
full automation of the algorithm since it does not require fovea in color fundus photographs, med. Image, 2003., vol.13,
any intervention by clinicians, which releases necessary no.6, pp. 859-870
resources (specialists) and reduces the consultation time, [10] T. Walter, J.C. Klein, A contribution of image processing to the
diagnosis of diabetic retinopathy- detection of exudates in color
hence its use in primary care is facilitated. As for future fundus images of the human retina, IEEE trans. Med. Image., vol
lines, the optic disc will also be detected with the goal 21, no. 10,oct 2012
of measuring the cup-to-disc ratio. A high C/D ratio will [11] T. Walter, J.C. Klein, A contribution of image processing to the
indicate that a fundus is suspicious of glaucoma. diagnosis of diabetic retinopathy- detection of exudates in color
fundus images of the human retina, IEEE trans. Med. Image., vol
21, no. 10,oct 2012.

AJES Vol.2 No.2 (July-Dec 2013)

Загружено:

Сведения о документе

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

AJES Vol.2 No.2 (July-Dec 2013)

Загружено:

Авторское право:

Доступные форматы

Asian Journal of Electrical Sciences

Volume 2 Number 2 July - December 2013

Sl. No. Title Page No.

2. Conditional Power Monitoring of Domestic Loads through Human Machine Interface 08

3. Vehicle Braking Performance Improvement by SVM Based Road Surface Detection 14

4. Approaching Vehicle Identification for the Visually Challenged 22

6. Retinal Blood Vessels Segmentation Using the Curvlet Transform 31

Information Science (IS) and Its Increasing Interaction with

Abstract Information Science plays an important role in the field

1 AJES Vol.2 No.2 July - December 2013

Fig. 1 Main task of information science

A.Information Science: Two Foci Web Technology and Systems;

AJES Vol.2 No.2 July - December 2013 2

Fig. 2 Role and uses of Manual Information Sciences

Fig. 3 New opportunities through Information Science

3 AJES Vol.2 No.2 July - December 2013

Fig. 4 From IS to IST nomenclature through Electronic Science

AJES Vol.2 No.2 July - December 2013 4

Fig. 5 Changing Character of IS

Design and development of information retrieval Mobile Information Services.

5 AJES Vol.2 No.2 July - December 2013

E-Resource management-indirectly related with

AJES Vol.2 No.2 July - December 2013 6

VII. Conclusion [7] Mitra, C.R. University-Industry interaction with reference to

[6] Lahiri, Ramansu. Research in library science in India (195095):

7 AJES Vol.2 No.2 July - December 2013

Conditional Power Monitoring of Domestic Loads through

AJES Vol.2 No.2 July - December 2013 8

Fig. 1 Schematic Representation

Fig. 2 Voltage transducer schematics

9 AJES Vol.2 No.2 July - December 2013

Fig. 3 System Module under test

Fig. 4 Software interface panel

AJES Vol.2 No.2 July - December 2013 10

Fig. 5 Test Pane in the Graphical System Design

Fig. 6 Sensed current formula

Table I Parametric Records For Different Loads

11 AJES Vol.2 No.2 July - December 2013

Fig. 7 System Result Identification

Fig. 8 RMS voltage waveform

AJES Vol.2 No.2 July - December 2013 12

V. Discussions obtained. Capacitor bank design and voltage regulation are

13 AJES Vol.2 No.2 July - December 2013

Vehicle Braking Performance Improvement By

AJES Vol.2 No.2 July - December 2013 14

15 AJES Vol.2 No.2 July - December 2013

C. Classification of Smooth Roads 3. Sand

AJES Vol.2 No.2 July - December 2013 16

Whenever there is a transition from one surface to

F. Coefficient of friction values for different roads

Table I Coefficient of Friction

Fig. 5 Rough road

Fig. 6 Grass road

Fig. 2 Asphalt road

Fig. 3 Rough Asphalt Road

17 AJES Vol.2 No.2 July - December 2013

Fig. 10 Intensity histogram for cement

AJES Vol.2 No.2 July - December 2013 18

Fig. 11 Intensity histogram for Grass

Fig. 12 Intensity histogram for Sand

Fig. 13 Margins for an SVM trained with samples

19 AJES Vol.2 No.2 July - December 2013

Fig. 14 Flow chart for the proposed Algorithm

The stopping distance of the ABS has been reduced