International Journal of Image Processing (IJIP) Volume (2) Issue (3)

Editor in Chief Professor Hu, Yu-Chen

International Journal of Image
Processing (IJIP)

Book: 2008 Volume 2, Issue 3
Publishing Date: 30-06-2008
Proceedings
ISSN (Online): 1985-2304

This work is subjected to copyright. All rights are reserved whether the whole or
part of the material is concerned, specifically the rights of translation, reprinting,
re-use of illusions, recitation, broadcasting, reproduction on microfilms or in any
other way, and storage in data banks. Duplication of this publication of parts
thereof is permitted only under the provision of the copyright law 1965, in its
current version, and permission of use must always be obtained from CSC
Publishers. Violations are liable to prosecution under the copyright law.

IJIP Journal is a part of CSC Publishers
http://www.cscjournals.org

©IJIP Journal
Published in Malaysia

Typesetting: Camera-ready by author, data conversation by CSC Publishing
Services – CSC Journals, Malaysia

CSC Publishers

Table of Contents

Volume 2, Issue 3, May/June 2008.

Pages
1 – 11 MAHII: Machine And Human Interactive Interface
G. Sahoo , Bhupesh Kumar Singh.

13 - 17
Segmentation of Handwritten Text in Gurmukhi Script
Rajiv K. Sharma, Amardeep Singh

18 – 28 Object-Oriented Image Processing Of An High Resolution Satellite
Imagery With Perspectives For Urban Growth, Planning And
Development
Afroz Shaik Mohammed1, Shaik Rusthum2

A Simple Segmentation Approach for Unconstrained Cursive
29 - 35 Handwritten Words in Conjunction with the Neural Network.
Amjad Rehman Khan, Zulkifli Mohammad

International Journal of Image Processing (IJIP) Volume (2) : Issue (3)

G. Sahoo & Bhupesh Kumar Singh

MAHII: Machine And Human Interactive Interface

G. Sahoo drgsahoo@yahoo.com
Department of Computer Science & Engineering
Birla Institute of Technology, Mesra,
Ranchi: 835 215, India

Bhupesh Kumar Singh kumar.bhupesh04@gmail.com
Department of Computer Science & Engineering
Lingaya’s Institute of Mgt. & Technology,
Faridabad: 121002, India

Abstract

Free-style sketching is more natural than drawing a sketch using mouse and
palette based tool. A number of sketching systems have been developed but not
much attention is paid towards friendliness of the system. We present here an
efficient Sketching system MAHII for sketching the strokes from multiple
domains. The proposed paper demonstrates how a sketching system can be
made user friendly and user interactive too. We also present the architecture of
our system that solves many flaws encountered in the previously built systems.
The interface should be such that it minimizes the gap between a human being
and a machine. Our system provides a robust architecture that proves to be
efficient. It also allows the user to sketch the diagram freely. And at the same
time it edits the sketches as well as recognizes the sketches efficiently. This
paper represents more robust and natural environment to the user. The MAHII
provides the user an interface that is built in accordance with a novice user.
MAHII provides free style sketching, recognizes the sketch produced by the user
with accurate results.

Keywords: MAHII, sketch, interface, stroke, multi-domain, Heuristics.

1. INTRODUCTION
Sketching and drawing are being used in various aspects of life. In the recent years, many
sketching systems have been developed, but there is always a gap between how a user draws a
sketch and how a machine interprets the sketch. A little has been thought about the development
of interactive human-machine interface. However, for some specific purpose many pen-based
sketching systems and sketch recognition systems have been developed in the literature. But
these systems suffer from the basic problem of naturalness of the sketching environment [2]. The
sketching interface must be such that it allows the user to draw sketches freely and naturally. It
must know when the system should take input from the user, edit the sketch as well as when it
should recognize the sketch.

For sketching and recognizing the user drawn sketches, we propose here a system known as
Machine And Human Interactive Interface (MAHII). The proposed system gives a solution to the
problems that have been encountered in previously built sketching interfaces. Further, our

International Journal of Image Processing, Volume (2) : Issue (3) 1


system works on multi-domain still maintaining the efficiency and naturalness of the system. We
present the interface architecture of our system whose framework has already been explained in
our earlier work [1]. This system MAHII is built so that it proves to be useful to a computer-expert
and even to a novice person. It contains some of the known commands of computer as well as
normal behavioral commands that a layman can use in sketching the diagrams. The System
MAHII also gives the user more liberty to draw sketches freely and naturally. Moreover, the user
need not to learn every capability of the system but still he can draw the sketch and get accurate
results of recognition.

The paper is organized as follows. In the next section we discuss the design issues of the
sketching systems followed by a brief overview of the proposed system. Section 4 and 5 take
care of the presenting the abstract view and layered architecture of the proposed system. Section
6 describes the interface commands followed by experiments and user feedback in the section 7.
The comparison study has been discussed in section 8 followed by the conclusion and future
work in the following sections.

2. DESIGN ISSUES
Various design issues have been predicted from previously sketch-based systems. We arrive at
the following design issues that can help in overcoming these issues successfully.

i) The system should display recognition result only when the user completes
sketching.
ii) The system should have some editing layers so that an editing layer can perform
editing after every sketch input
iii) The recognition should take place continuously and simultaneously but the result can
be displayed only when the waiting time of the user is more than average waiting
time between strokes.
iv) The recognition process should be restricted to a single domain until automatic
domain detection becomes feasible.

This can be achieved by the use of some interfacing mechanism, which is our main concern in
this work. The primary goal of the interfacing mechanism is to connect the domain class
description to the recognition engine only when its few corresponding elements are recognized by
the recognition engine through Heuristic engine using a database containing shapes. With or
without giving more stress to the recognition efficiency, one should go for describing a system
that is more natural having user-friendly environment. Further, the system should posses a
powerful and robust architecture and can be more useful for the user.

3. THE PROPOSED SYSTEM
In this section we give the overview of our proposed system called Machine And Human
Interactive Interface (MAHII). By defining stroke as a set of points from pen-up, we categories
strokes that are drawn or sketched as:
1. Single stroke input
2. Multi strokes input
And then we say that
i) The system MAHII recognizes both single stroke input and multi-strokes input as well.
ii) Further, it provides a new style of interface that takes care of human- machine interaction
as well as free-style sketch recognition.
iii) Even by considering a single window for both input and output in our designing process
we still maintain a low complexity regarding the architecture of the system. For this we
provide a three-layered architecture that we will explain in subsequent sections.
In our construction, we consider two modes – sketch mode and edit mode. MAHII provides the
user more friendly environment because it does not require the user to select modes –sketch



mode or editing mode. When the user draws sketches, the system shows him the edited images
simultaneously based on which he/she can modify the respective sketch. So, he/she does not
need to switch between sketch mode and edit mode [4], [13]. User can draw whenever he wishes
to draw. MAHII can work in single domain as well as multi-domains. Even the user is not required
to select the domain which he wants to draw in. For this, we provide domain class descriptions for
each domain and an interface mechanism through which the system links to the particular domain
automatically. MAHII automates all the functionalities. MAHII does not use specialized tools like
palette based tools etc. and provides features for novice as well as expert users. So, it fulfils the
demands of all types of users.

Again it can be pointed out here that some work has been done on Human machine interaction
[5] but in solitude; similarly many pen-based sketching systems have been developed but without
paying much attention towards human machine interaction. However, the proposed system
explores the user aspects of a recognition based sketching system that efficiently recognizes the
naturally drawn sketches and displays the result on the same window. MAHII reduces the
complexity and increases the efficiency of the sketching system and gives the result with
accuracy. It changes itself according to the user environment.

4. ABSTRACT VIEW OF THE SYSTEM FRAMEWORK
The system framework (Figure 1) that has been discussed in [1] gives the information about the
working of the system. Now, this framework is implemented as the architecture of the system and
the corresponding interface is named as MAHII. The architecture is explained below.

FIGURE 1: System framework

User provides input to the system by drawing sketches on the MAHII. This interface gets the
sketch input from the user and sends it to the input refining block where input is refined by
removing noise from the sketch input and then it is sent to the recognition engine as well as to the
MAHII window. The edited or refined sketch is sent back to MAHII window because it displays the
user the edited version of the sketch drawn at previous time instant. If it is not according to the
user’s requirement, user can change it by using simple commands that are of common behavior.
We explain such commands in a later section. When the edited input is sent to MAHII, it is also
sent to the recognition engine also. After recognition, the output of the recognition is sent to the
MAHII window. The flexibility and ease use of this system reflects here that user needs not to
know when he should change the sketch mode, edit mode or recognition mode. The system
takes care of these things by itself.

If a user draws a stroke say in 2 seconds and next stroke after 4 seconds then he draws the third
stroke after 5 seconds and so on, the system takes the average of the first few time gaps
between the consecutive strokes. Thus, it calculates the average of time difference then sets the



systems’ recognition time greater than the calculated average waiting time. The system does this
for every user. So, if a user is faster in sketching, the system provides him results in less time and
for a novice user, it displays the result according to the time taken by the user to draw next
stroke. Thus, the system changes itself according to the user. It reduces the problem when the
system should or should not recognize sketches.

The interface, which we represent in the framework that is between domain class description and
recognition engine, helps the system to work in a multi-domain environment while it works
actually over a single domain. In comparison to the earlier multi-domain sketch systems [4], [12],
[13] where the user had to select explicitly the domain that he wants to use, MAHII does not
impose any such requirement. The recognition engine actually requires the domain descriptions.
Moreover, the database contains knowledge about recognized shapes and their particular
domain. So, when the system needs to select a particular domain, it identifies few basic shapes
or elements that are parts of a particular domain followed by the class description with the
recognition engine. By using this modular approach we not only increase the system efficiency,
naturalness of the interface but also reduce the complexity of the system and enable the system
to work in a multi-domain environment. For an example, let us consider three domains as one
domain for mechanical engineering, one domain for electronic circuit designing and one domain
for dancing choreography. Initially, the user draws the input in a particular domain and gets the
edited and refined sketch at the next instant of time. If he draws 3-4 basic shapes, the system
relates them to the particular domain and gets to know that it belongs to, for instance,
choreography domain. It gets this basic information, of course, from the corresponding database.
Then, it interfaces the choreography domain with the recognition engine.

5. LAYERED ARCHITECTURE OF MAHII
The MAHII interface uses modular approach. Basically the architecture of MAHII consists of three
layers namely
i) Input layer or sketching layer or presentation layer
ii) Editing layer or middle layer or refining layer
iii) Core layer or main processing layer or recognition layer
and are depicted in Figure 2.

FIGURE 2: MAHII architecture

5.1 Input layer
The input layer is the window that is visible to the user for drawing the sketches. The user can
draw frequently the sketches and these are used as input to the system. The user even may not
have the knowledge of the functionality of the computer. As the user draws the symbols or
sketches, he gets the edited image as well. The user need not wait for the system as and when to
draw a sketch on a pen-based system.
This window after taking the input from the user not only displays the smooth sketch but also
displays the recognition layer’s result. We consider many issues and tackle various problems
encountered in the earlier sketch interfaces.

The sketch interface MAHII consists of a single window. However, the editing and recognition
takes place in the backend. The window represents simple, congestion-free screen to the user.



Though it contains menus for the computer known user still it contains many commands that a
novice user can use.

We now describe some of the advantages of MAHII over existing system like SkRUI. The system
SkRUI which was developed earlier contains two separate windows [4]. User draws in one
window and gets the result in the other window. So, user switches explicitly between the two.
MAHII makes this task easier. It uses a single window for all purposes and it does not have any
synchronization problem. Further, many interfaces provide menu based tools like edit, undo,
redo, cut, copy, paste. We make these tools available in our system in addition to some extra
natural commands that can easily be used by novice.

In SkRUI, user cannot draw new items when it is in edit mode whereas MAHII does editing by
itself and allows the user to draw the new sketches as well as to modify the previously drawn
items. For this purpose, it uses buffers for temporary storage of previously drawn sketches and
whenever the user wants to copy a sketch, he just needs to redraw the sketch over the existing
one. The MAHII interface shows large buttons (menu based) so that user can easily identify them
and the button is labeled with the pictures of what it does.

5.2 Editing Layer
This layer performs the editing of the sketch input. This layer includes the functioning of image
filtering, image segmentation and image editing. It provides input to the recognition layer. It gets
the input from the input layer and refines it by removing the noise from the sketch and increases
the contrast of faded image/sketch. Any unclosed or unconnected Figures or basic shapes are
connected by the system itself [6]. For example, if user wants to draw a circle, but incomplete, the
editing layer does this for the user (Figure 3).

FIGURE 3: Working principle of editing layer

This layer takes input strokes one by one from the input layer as soon as the user draws the
strokes on the MAHII interface. If the user draws a stroke at t time instant, the editing layer gets
the input at say (t+∆t) time instant and edits during (t+2∆t) and it returns the result at (t+3∆t) time
instant to both the input layer and recognition layer (Figure 4). The recognition layer does its job
during 2∆t units of time. The final output of recognition layer gets displayed at (t+5∆t). Here ∆t
represents a very small delay of the order of microseconds.

FIGURE 4: Process of editing at different time instants



Previously developed systems have a drawback that system should wait until the user finish
sketching to display editing or recognition results. But since both of these activities take place
simultaneously at the backend so, neither user nor the system should wait for sketching or
responding respectively.

Here, an alternation that we want to provide to the user to know about the mode is that one can
use pen-based editing, that is, when the cursor is in pen-mode, then the user can draw sketches
and when it is in cursor mode it means the editing mode. Thus, sketching and editing use distinct
pen motions.

5.3 Recognition layer
This layer is the core of our proposed system. This is the only layer that is given the power of
recognition (the output) and named as recognition layer. It gets the input from the editing layer as
mentioned above and uses heuristic engine to perform semantic checking, and searches for the
relevant details from the system database. Finally, it shows the result to the MAHII interface or
input window.
For processing, the system uses the domain class descriptions through an interface. The
heuristic engine is being used here for structure matching and semantic checking. We can say
that the recognition layer introduces the concept of domain independence. It recognizes the basic
elements belonging to a particular domain using the Heuristic engine and database of samples of
shapes and then identifies to which domain those elements belong to and then selects a
particular domain class description out of many and connects it to the recognition engine. Thus, it
makes the architecture as a disconnected type of architecture. When a new domain is referred to,
the previous domain interface gets destroyed. It also helps the domain descriptor to define
separate domain class descriptions and different domains never mingle with each other. This also
reduces the complexity of domain description plus interprets the results with accuracy.

The recognition layer recognizes the output and displays the result to the input-output layer of the
interface (MAHII). The recognition layer provides the result in real time because, it calculates the
average waiting time for each user and then responds accordingly. It does not wait for editing or
recognition by user, it starts working as soon as it gets the input.

FIGURE 5: Abstract layered view

Based on the input-output, we represent a simple view of layered architecture of MAHII interface.
The Figure 5 reflects the simple view of a three layered architecture through which we can
estimate the real time responses. We use a single window for getting input from the user as well
as for showing the output of editing and recognition of the sketch. This is an efficient approach
since the user uses a single interface there is no need to switch between the sketch mode and
the edit mode.

6. INTERFACE COMMANDS
Our system promises to be a user-friendly interface by indulging itself according to the user’s
choice. It can respond the way the user behaves. For a computer-literate user, it provides menu-



based buttons and tools. At the same time, it provides normal or generally used commands [7]
that a novice can use it as he/she draws the sketch on the paper with a pen.

If a user draws a sketch on a paper and if it is not drawn according to the user’s choice then
he/she simply cuts it using a X symbol (Figure 6(a)). In the similar way, this command deletes the
figure. If a user wants to copy the same figure he/she needs to redraw it over the figure itself
(Figure 6(b)). This creates a copy in the buffer of that window area wherever the user leaves the
cursor or with a click the sketch can be pasted. For the same purpose, the user can use the
command C, say, for copy and P for paste (Figure 6(c)).

(a)

(b)

(c)
FIGURE 6: Demonstration of (a) Delete command, (b) Copy command, (c) Paste command

If user wants to cut a sketch and paste it somewhere else he needs to delete the figure after copy
and paste. If user wants to redo, he can simply do it by using the command like R or r over the
window area which will redo the last undo (Figure 7) part.

FIGURE 7: Redo command



To undo, the user needs to use the symbol U or u. This undoes the thing what the user has done
just before (Figure 8).

FIGURE 8: Undo command

To rub a portion of the sketch, a zigzag command can be used (Figure 9).

FIGURE 9: Rubbing a portion of a sketch

7. EXPERIMENTS AND USER FEEDBACK
We have provided the system to be used by any type of users. There are two groups of people to
use the system. Some of them are good enough in computer skills and some have never any
interaction with the computer before. Our basic aim is to get the idea of how the system MAHII is
interactive and natural to the users. There are no instructions or demonstration given how MAHII
basically works; instead it provides amazing results of much satisfactory and fulfilling the target of
natural or user-friendly environment.

The experiment was done as follows. There were 12 users who were asked to draw sketches of
various domains. There were 6 novice users and 6 were computer known people. Novice users
found our system simple and easy to operate. They draw sketches of electronic circuits and
shortest-path graphs without explicitly switching between the domains. Our system was
successful in finding the specific domain itself. This provided the novice users enough flexibility.
Even MAHII provided self-editing feature that the users appreciated. Since people did not know
any command that MAHII provides, so they used normal behavioral commands e.g. to delete a
figure they just crossed over it. So, MAHII proves to be successful as natural, simple and user
friendly.

Then we tried to find out how it encourages the computer known people. They were also asked to
draw sketches from various domains. Our system proved successful for them as well. They
sketched strokes and they found it easy to use the menu based commands since the menu
based commands were based on buttons that were labeled with the corresponding images rather
than the command name. MAHII was successful for expert users as well in all respects.



Table 1 shows our efficiency measurement in terms of ease of use.

USER Efficiency in Terms of
Ease of Use
Novice 93%

Expert 94%

Table 1: Efficiency of MAHII for various users

8. RELATED WORK AND COMPARISON STUDY
In this section we compare the efficiency and effectiveness of the proposed system withy other
existing systems defined for sketching. In Table 2 we have given the recognition rate of MAHII for
different inputs, and domains.
Comparison with previous work shows that MAHII improves the sketching, editing and
recognition. This shows its capability better than the earlier systems.

Input Type Static 97%
Dynamic 95%
Domain Single Domain 96%
Multi Domain 87%

Table 2: Recognition rate

Various sketch systems like discussed as in paper [11] have shown the efficiency (6.5 / 7) of their
system to 92.8%. CALI [10] gives recognition efficiency up to 95.3% for an overall recognition
rate. The recognition rate for static inputs has been improved to 95.1%, but for dynamic input, it is
limited up to 70% only. Bayesian approach [9] that gives the accuracy rate till 94% is based on
the probability. Whereas using the Fuzzy Logic approach the efficiency is up to 92% for a single
domain system and not much work has been done to increase the efficiency of multi-domain
sketching system [8]. However, MAHII gives 87% efficiency for multi-domain sketching.

Further, for the purpose of various comparison aspects we consider here the existing systems
CAD, SkRUI, SketchRead, FIRST and depict the corresponding graphs in Figures 10 and 11.

FIGURE 10: The comparison graph respect to sketching, editing and recognition



(a)

(b)

FIGURE 11: Efficiency curves respect to easy of drawing for (a) Expert user, (b) Novice user

From the above Table and Figures we can say that the system MAHII has a capability in sketching,
editing and recognition over the existing systems. It provides better results with a higher rate of
usefulness: its automatic editing feature makes it a perfect sketching system.

9. CONCLUSION AND FUTURE WORK
MAHII interface presented here is a new system that exhibits the powerful yet more natural
environment for the user. It takes care about the every kind of user and the people of various
domains. Users need not to learn computer to use this system. It reduces the complexity still
showing more freedom to the user in drawing the sketches.

MAHII satisfies the requirements of novice user and it can further be enhanced to provide more
real time responding and friendly environment. It may use large amount of buffer area, but
satisfies the demand of multi domain recognition without the knowledge of user. This work can be
carried out later.

10. REFERENCES:
1. G. Sahoo, Bhupesh Kumar Singh, “A New Approach to Sketch Recognition using
Heuristic”, International Journal of Computer Science and Network Security, Vol. 8 No. 2,
2008., PP. 102-108.,
2. Mankoff, J. Hudson, S.E., and Abowed. G. D., “Providing integrated toolkit-level support
for ambiguity in recognition-based interfaces”. In proceedings of CHI 2000 conference on
Human Factors in Computing Systems, 368-375.



3. Newman. M.W., Lin. J., Hong. J. I., and Landay. J. A., “DENIM: An Informal web site
design tool inspired by observations of practice.” Human-Computer Interaction 18(3):
259-324
4. Alvarado C., “Sketch Recognition And User Interfaces: Guidelines for Design and
Development”, AAAI Fall Symposium on Pen-Based Interaction, 2004.
5. Landay. J.A., and Myres, “Interactive Sketching for the early stages of user interface
design”. In proceedings of CHI 95, 43-50. B.A.1995
6. Mahoney. J. V., and Fromherz. M. P. J., “Three main concerns in Sketch recognition and
an approach to addressing them”. AAAI Spring Symposium on Sketch Understanding,
2002.
7. Rubine D. “Specifying gesture by example”. SIGGRAPH 1991, Las Vegas, pp.329-327
8. Fonseca. M., Jorge. J., “Using Fuzzy Logic to recognize geometric shapes interactively”.
Proceedings of the 9th IEEE conference on Fuzzy systems, pp. 291-296, 2000.
9. Shilman. M., Pasula. H., Russell. S., and Newton. R. “Statistical visual language models
for ink parsing.” Proceedings of AAAI spring Symposium on Sketch Understanding,
Stanford University, pp. 126-132, March 2002.
10. Fonseca. M.’ Pimetal. C., “CALI: an online scribble recognizer for calligraphic interface”.
Proceedings of AAAI Spring Symposium on sketch Understanding, Stanford University,
PP.51-58, March 2002.
11. Wolin. A., Smith. D., and Alvarado. C., “A Pen—based Tool for efficient labeling of 2D
sketching”, EUROGRAPHICS workshop on sketch based Interfaces and Modeling
(2007).
12. Alvarado. C. and Oltmans. M., Davis. R.’ “A Framework for multi-domain sketch
recognition”. Proceedings of AAAI Spring Symposium on sketch Understandings,
Stanford University, pp.1-8, March 2002.
13. Alvarado. C.’ “Multi-domain sketch understanding”’ PhD thesis, Department of Electrical
Engineering and computer science, Massachusetts Institute of Technology, September
2004.


Rajiv K. Sharma & Dr. Amardeep Singh

Segmentation of Handwritten Text in Gurmukhi Script

Rajiv K. Sharma rajiv.patiala@gmail.com
Sr. Lecturer, SMCA
Thapar University,
Patiala, 147002
Punjab, India

Dr. Amardeep Singh amardeep_dhiman@yahoo.com
Reader, UCoE,
Punjabi University,
Patiala, 147004
Punjab, India

Abstract

Character segmentation is an important preprocessing step for text recognition.
The size and shape of characters generally play an important role in the process
of segmentation. But for any optical character recognition (OCR) system, the
presence of touching characters in textual as well handwritten documents further
decreases correct segmentation as well as recognition rate drastically. Because
one can not control the size and shape of characters in handwritten documents
so the segmentation process for the handwritten document is too difficult. We
tried to segment handwritten text by proposing some algorithms, which were
implemented and have shown encouraging results. Algorithms have been
proposed to segment the touching characters. These algorithms have shown a
reasonable improvement in segmenting the touching handwritten characters in
Gurmukhi script.

Keywords: Character Segmentation, Middle Zone, Upper Zone, Lower Zone, Touching Characters.
Handwritten, OCR

1. INTRODUCTION
In optical character recognition (OCR), a perfect segmentation of characters is required before
individual characters are recognized. Therefore segmentation techniques are to apply to word
images before actually putting those images to reorganization process. The simplest way to
segment the characters is to use inter – character gap as a segmentation point[1]. However, this
technique results in partial failure if the text to be segmented contains touching characters. The
situation becomes grim if text consists of handwritten characters. The motivation behind this
paper is that to find out a reasonable solution to segment handwritten touching characters in
Gurmukhi script. Gurmukhi script is one of the popular scripts used to write Punjabi language
which is one of popular spoken language of northern India. Because our work is related with
segmentation of Gurumukhi script, so it is better to discuss some characteristics of the said script
so that the reader can have a better idea of the work.

International Journal of Image Processing, Volume (2): Issue (3) 12


2. CHARACTERISTICS OF GURMUKHI SCRIPT
Gurmukhi script alphabet consists of 41 consonants and 12 vowels[2] as shown in FIGURE 2.
Besides these, some characters in the form of half characters are present in the feet of
characters. Writing style is from left to right. In Gurmukhi, There is no concept of upper or
lowercase characters. A line of Gurmukhi script can be partitioned into three horizontal zones
namely, upper zone, middle zone and lower zone. Consonants are generally present in the
middle zone. These zones are shown in FIGURE 1. The upper and lower zones may contain
parts of vowel modifiers and diacritical markers.

FIGURE 1 : a) Upper zone from line number 1 to 2, b) Middle Zone from
line number 3 to 4, c) lower zone from line number 4 to 5
In Gurmukhi Script, most of the characters, as shown in FIGURE 2, contain a horizontal line at
the upper of the middle zone. This line is called the headline. The characters in a word are
connected through the headline along with some symbols as i, I, A etc. The headline helps in the
recognition of script line positions and character segmentation. The segmentation problem for
Gurmukhi script is entirely different from scripts of other common languages such as English,
Chinese, and Urdu[3] etc. In Roman script, windows enclosing each character composing a word
do not share the same pixel values in horizontal direction. But in Gurmukhi script, as shown in
FIGURE 1, two or more characters/symbols of same word may share the same pixel values in
horizontal direction.
This adds to the complication of segmentation problem in Gurmukhi script. Because of these
differences in the physical structure of Gurmukhi characters from those of Roman, Chinese,

Consonants (Vianjans)

FIGURE 2 a) : Consonants (Vianjans)



Vowels and Vowel diacritics (Laga Matra)

FIGURE 2 b) : Vowels and Vowel diacritics (Laga Matra)

Other symbols

FIGURE 2 c) : Other symbols

Japanese and Arabic scripts, the existing algorithms for character segmentation of these scripts
does not work efficiently for handwritten Gurmukhi script.

3. PREPROCESSING
Preprocessing is applied on the input binary document so that the effect of spurious noise can be
minimized in the subsequent processing stages. In the present study, both salt and peeper noise
have been removed using standard algorithm [4]. It is supposed that height and width of document
can be known easily. The image is saved in the form of an array. For that purpose a 2-D array
with number of rows equal to height of the document and number of columns equal to width of the
document is created. Calculate the maximum intensity of pixels in the document using any
standard function available in the tool used for the implementation, it is getRGB() method
available in java. Scan every pixel of document and compare its intensity with the maximum
intensity. If the intensity is equal to maximum intensity, store one in the array at that location, and
if it is not equal store zero in the array.

4. PROPOSED PROCEDURES TO SEGMENT LINE, WORD and CHARACTER
Line Detection
The following procedure is implemented to find the location of lines in the document.
i. Create an array of size equal to height of the document and with two columns.



ii. Start from the first row and count the number of 1’s in that row. If it is zero, move to next
row. And if it is not zero, that is the starting location of that line. Store that location in the
array.
iii. Check consecutive rows until we get 0.The before we get zero is the ending location of that
line. Store that value in the array.
iv. Also calculate the location of maximum intensity in each line and store it in the second
column before that line. It would be used as the starting position of characters.
v. Repeat step (ii) to (iv) for the whole document.
Word Detection
The following procedure is implemented to find location of words in each line.
i. Create a 2-D array.
ii. For each line move from 0th pixel up to width.
iii. Calculate the number of one’s in first column from the starting location of line to the ending
position of line.
iv. If number of 1’s are not zero, that is the starting location of word. Save that location in that
array. Keep on moving to the right until we get no one in any column. The column with 0
1’s is the ending location of the word. Store that location in array too.
v. Repeat this until we reach the width.
vi. And repeat step (ii) to (v) for each line.
Character Detection
The following procedure is implemented to find the location of character in each word.
i. Create a 3-d array. Its first index will represent line number. Second index will represent
word number and third index will contain the location of character. This array will be
created dynamically.
ii. Repeat the step (iii) to (iv) for each line and each word detected so far.
iii. Move from starting position of the word to the ending position of the word.
iv. Start from the starting position of line and move downwards to the ending position. Count
the number of one’s in that column leaving the location of line with maximum intensity. If it
is not zero, that is the starting position of character. Move to right until we get column with
no ones. that will be the ending location of character.
This process will generate the location of characters.
The above approach was put to number of documents; the image of one such scanned document
is given below.

FIGURE 3: Scanned Image of a Document
The result of the scanned document after processing is given below.



FIGURE 4: Processed Document

The main objective of the work was to segment the lines, words and to segment the touching
characters present in handwritten document in Gurmukhi script. We obtained the following table
after putting handwritten Gurmukhi documents for segmentation. The results are summarized as
in following tables:

Document No of Lines Correctly Inaccurate segmentation Accuracy
Detected
Doc1 5 4 1 80%
Doc2 8 7 1 87.5%
Doc3 10 8 2 80%
Doc4 13 11 2 84.61

TABLE 1: ACCURACY for Line Segmentation

Document No of Words Correctly Inaccurate segmentation Accuracy
Detected
Doc1 38 32 6 84.21%
Doc2 56 49 7 87.5%
Doc3 95 79 16 83.15%
Doc4 110 90 20 81.81

TABLE 2: ACCURACY for Word Segmentation

Document No of Characters Correctly Inaccurate segmentation Accuracy
Detected
Doc1 79 71 8 89.8%
Doc2 168 145 23 86.30%
Doc3 224 175 49 78.12%
Doc4 289 232 57 80.27

TABLE 3: ACCURACY for Character Segmentation

5. CONCLUSION AND FUTURE WORK
This work was carried out to detect lines present in scanned document in handwritten Gurumukhi
script. So firstly we are to find out the lines present in the document then to find words present in
each line detected at the first step. Using the detected words it is to segment characters present
in each word. Therefore using line detection algorithm (the first approach) lines were detected.
Mostly we found the correct lines, but some were not detected correctly. The reason behind this
may be the writing style of Gurumukhi script as shown in FIGURE 1. So the words presents in the



lower zone were considered as a different line. The correctly detected lines were further put to
word detection algorithm. Here the results were good, but sometimes when the words were not
joined properly then that was detected as a different word. The locations of the detected words
were used to segment the characters. At few point segmentation was good but at few point it was
not upto the expectations. This may be because of the similarity in the shapes of few characters.
All these issues can be dealt in the future for handwritten documents written in 2-dimensional
script like Gurumukhi by making few changes to proposed work.

6. REFERENCES
1. Y. Lu. “Machine Printed Character Segmentation – an Overview”. Pattern Recognition,
vol. 29(1): 67-80, 1995

2. M. K. Jindal, G. S. Lehal, and R. K. Sharma. “Segmentation Problems and Solutions in
Printed Degraded Gurmukhi Script”. IJSP, Vol 2(4),2005:ISSN 1304-4494.

3. G. S .Lehal and Chandan Singh. “Text segmentation of machine printed Gurmukhi
script”. Document Recognition and Retrieval VIII, Proceedings SPIE, USA, vol. 4307:
223-231, 2001.

4. Serban, Rajjan and Raymund. “Proposed Heuristic Procedures to Preprocesses
Character Pattern using Line Adjacency Graphs”. Pattern recognition, vol. 29(6): 951-
975, 1996.

5. Veena Bansal and R.M.K. Sinha. “Segmentation of touching and Fused Devanagari
characters, ". Pattern recognition, vol. 35: 875-893, 2002.

6. R. G. Casy and E. Lecolinet. “A survey of methods and strategies in character
segmentation”. IEEE PAMI, Vol. 18:690 – 706,1996.

7. U. Pal and Sagarika Datta. “Segmentation of Bangla Unconstrained Handwritten Text”.
Proceedings of the Seventh International Conference on Document Analysis and
Recognition (ICDAR ), 2003.

8. U. Pal, S. Sinha and B. B. Chaudhuri. “Multi-Script Line identification from Indian
Documents”, Proceedings of the Seventh International Conference on Document
Analysis and Recognition (ICDAR) 2003.

9. Rajean Plamondon, Sargur N. Srihari. “On – Line and Off – Line Handwritting
Recognition: A Comprehensive Survey”, IEEE Transaction on Pattern Analysis and
Machine Intelligence, Vol 22(1). Janurary, 2000.

10. Giovanni Seni and Edward Cohen. “ External word segmentation of off – line handwritten
text lines”. Pattern Recognition, Vol. 27(1): 41-52, 1994.


Afroz Shaik Mohammed & Dr Shaik Rusthum

OBJECT-ORIENTED IMAGE PROCESSING OF AN HIGH RESOLUTION
SATELLITE IMAGERY WITH PERSPECTIVES FOR
URBAN GROWTH, PLANNING AND DEVELOPMENT

Afroz Shaik Mohammed1* Email:smafroz@yahoo.com
Deccan college of Engg. and Technology,
(Affiliated to Osmania University)
Dar us Salam, Near Nampally, Hyderabad-500 001, (A.P), India.
Mobile: 9959732140, Landline: 914023535891

Dr Shaik Rusthum2 srgisace_2k7@rediffmail.com
Professor & Principal,
VIF College of Engg. & Technology, Gandipet, Hyderabad.
Mobile: 9848530370

Abstract

The management of urban areas by urban planners relies on detailed and updated knowledge of
their nature and distribution. Manual photo-interpretation of aerial photographs is efficient, but is
time consuming. Image segmentation and object-oriented classifications provide a tool to
automatically delineate and label urban areas. Here single pixels are not classified but objects
created in multi-resolution segmentation process, which allows use of, spectral responses but
also texture, context and information from other object layers. This paper presents a
methodology allowing to derive meaningful area-wide spatial information for city development
and management from high resolution imagery. Finally, the urban land cover classification is
used to compute a spatial distribution of built-up densities within the city and to map
homogeneous zones or structures of urban morphology.

Key words: Object oriented, Classification, Segmentation, Spatial information,
Accuracy assessment, Urban morphology

1. INTRODUCTION

Human land use decisions on the environment are influenced by socioeconomic factors which
can be represented by spatially distributed data. The accelerating urban sprawl, often
characterized by a scattered growth, has rarely been well planned, thus provoking concerns over
the degradation of our environmental and ecological health[2]. Up-to-date and area-wide
information management in highly dynamic urban settings is a critical endeavor for their future
development. Thematic assessments of urban sprawl involve procedures of monitoring and
mapping, which require robust methods and techniques[3]. Conventional survey and mapping
methods cannot deliver the necessary information in a timely and cost-effective mode. Limited
spatial information within the built-up zone hinders urban management and planning. Especially
in growing and altering cities lack of up-to-date data is apparent. The challenge of classifying
urban land cover from high resolution remote sensing data arises from the spectral and spatial
heterogeneity of such imagery. There to the high dissimilarity of functions like industrial or

International Journal of Image Processing, volume (2) issue (3) 18


residential areas as well as parks or agricultural regions causes problems in terms of an indirect
inferring of land use [7,8].

2. STUDY AREA, METHODOLOGY AND RESULTS

2.1 Study Area

The study site, Vijayawada city, known as the political capital of the State, located in the
south-east of India is the third largest city of Andhra Pradesh state. Vijayawada is located on the
banks of the sacred Krishna River and is bounded by the Indrakiladri Hills on the West and the
Budemeru River on the North.

The other details of Vijayawada city are given in Table 1.

Table 1: Details of Vijayawada city

1)State Andhra Pradesh 6)Time zone IST (UTC+5:30)

2)District Krishna 7)Population 10,35,536

3)Coordinates 16.30° N 80.37° E 8)Density of 17,854 /km²
Population
4)Area 58 km² 9)Postal code 5200xx

5)Elevation 125 m 10)Telephone code +91 866

2.2 Data

High resolution multispectral IRS P-6 LISS-3(Band 2,3,4 &5) images were taken. This satellite
carries three sensors (LISS-III, AWiFS & LISS-IV) with 5.8m, 23.5m & 56m resolutions and
fore-aft stereo capability. The payload is designed to cater to applications in cartography, terrain
modeling, cadastral mapping etc., These images were supplied by NRSA, Hyderabad,
India.(http://www.nrsa.gov.in)

Global Positioning System (GPS) receiver has been used for ground truth data that records the
coordinates for the polygons of homogeneous areas, and also it records the coordinates that will
be used for geometric correction. The GPS is in existence since the launch of the first satellite in
the US Navigation System with Time and Ranging (NAVSTER) system on February 22, 1978,
and the availability of a full constellation of satellites since 1994. The US NAVSTAR GPS
consists of a constellation of 24 satellites orbiting the Earth, broadcasting data that allows a GPS
receiver to calculate its spatial position (Erdas imagine, 2001).



Ground truth data is used for use in image classification and validation. The user in the field
identifies a homogeneous area of identifiable land cover or use on the ground and records its
location using the GPS receiver. These locations can then be plotted over an image to either train
a supervised classifier or to test the validity of a classification.

2.3 Methodology

Here, the description about the land cover types and their distributions of the study area is given.
Except this, the remote sensing images, ground truth used in this study are described in detail and
also the data preprocessing before conducting the classification is described. Methodology to
perform this research is given in flow chart 1.

Data

Remote Field data Recording of GPS
IRS P-6 sensing (Ground points and GCP’s and
LISS-III images truth) land cover type
knowledge in study area

Object oriented
Unsupervised Supervised image analysis
classification classification

Cross classified Cross classified Image Classify image
image with image with segmentation with NN
ground truth ground truth
(eCognition) classifier
for error matrix

Classified Cross classified
Accuracy Accuracy image in image with
assessment assessment eCognition ground truth
(Error (Error
matrix) matrix)

Accuracy
assessment
(Error Final land
Final land
matrix) cover
cover
image image

Study urban morphology in the
study area (Growth in the study
area)

END
Flow chart 1: Methodology



2.4 Results

2.4.1 Unsupervised and Supervised classification

The basic premise for unsupervised classification is that spectral values within a given land cover
type should be close together in the measurement space, whereas spectral data in different
classes should be comparatively well separated (Lillesand, 2001). Unsupervised classification is
fast and has the ability to analyze the image spectral statistics completely and systematically,
thus unsupervised classification can give useful indication of detectable classes for supervised
classification (Mather, 1987).

Supervised classification result of the study area (Vijayawada city) with different land cover
types is presented in Plate 1.

Plate 1: Supervised classification result of Study area(Vijayawada city) from IRS P-6
LISS-III Imagery

2.4.2 Object oriented image analysis

Using the object oriented image analysis approach to classify the image is performed in
eCognition. Object oriented processing of image information is the main feature of eCognition.
The first step in eCognition is always to extract image object primitives by grouping pixels. The
image objects will become building blocks for subsequent classifications and each object will be



treated as a whole in the classification. Multi-resolution segmentation is a basic procedure in
eCognition for object oriented image analysis. The segmentation rule is to create image objects
as large as possible and at the same time as small as necessary. After segmentation, a great
variety of information can be derived from each object for classifying the image. In comparison
to a single pixel, an image object offers substantially more information.

2.4.3 Comparison of segmentation results with different scale parameters in the study
area

Plate 2 is the original image of the study area. Plates 3, 4 and 5 show the effect of segmentation
results using different segmentation parameters. Except scale difference, the other parameters
that influence the segmentation result are color, shape, smoothness and compactness but these
are kept constant. Plate 3 is the segmentation result with a scale parameter 5.Comparing this
segmentation result with the original image, it is found that neighbor pixels are grouped into
pixel clusters-objects, and because of the low value of scale parameter, there are too many small
objects. Plate 4 is the segmentation result with scale parameter 10. It is found by comparing it
with Plate 3 that higher scale parameter value generates larger objects. Plate 5 is the
segmentation result with scale parameter 20. By visual comparison, a scale parameter of 10 is
selected because the segmentation result fits the information class extraction best. Based on these
parameters, segmentation process is performed.

2.4.4 Image classification

Classification is the process of connecting the land cover classes with the image objects. After
the process of classification, each image object is assigned to a certain (or no) class. In
eCognition, the classification process is an iterative process. The classification result can be
improved by editing the result: defining unclassified objects with the correct classes, correcting
wrongly classified objects with the correct classes, etc.

Plate 2: Original image of the study area (Vijayawada city)



Plate 3: Segmentation result with scale parameter 5, color 0.8, shape 0.2, smoothness
0.9, & compactness 0.1





2.4.5 Accuracy assessment

Accuracy assessment values were generated in eCognition by creating a test area and training
mask (TTA) as shown in table 2. The TTA mask contained 52 “Urban,” “Vegetation,” and
“rocky” objects and 25 “water” objects. These objects, representing actual land cover were
compared against the classified identity of these objects. The “water” class was very accurately
classified, and was therefore limited to 25 testing objects in order to reduce its inflationary effect
on the accuracy statistics.

Table 2: Error matrix and Accuracy statistics
Reference data*
Classification U.A W V R Total
Data
U.A 12766 0 0 951 13717

W 5168 59897 0 0 65065

V 0 0 17600 0 17600

R 0 0 1180 30336 31516
Total 17934 59897 18780 31287 127898

*U.A-Urban Area, W-Water, V-Vegetation, R-Rocky



Producer’s accuracy can be calculated using the formula:

PA (class I) = aii ki

Producer’s accuracy (%):
Urban area=12766/17934=71.18
Water=59897/59897=100
Vegetation=17600/18780=93.7
Rocky=30336/31287=96.9
User’s accuracy can be calculated using the formula:

UA (class I) = aii ik

User’s accuracy(%):
Urban area=12766/13717=93
Water=59897/65065=92
Vegetation=17600/17600=100
Rocky=30336/31516=96.2

Over all accuracy can be calculated using the formula:
OA= kk ik 1/n kk
Over all accuracy= (12766+59897+17600+30336)/127898=94.2%

Kappa Statistics can be computed as:
2
K=N ii- i+* +i) N - i+* +i)

Where
n=no. of the rows in the matrix
ii=the no. of observations in row i and column i (on the major diagonal)
i+=total of observations in row i
+i= total of observations in column i
N=total no. of observations included in matrix
Therefore
Kappa Statistics:
K=

=0.91

An overall accuracy of 0.942 and a Kappa Index of Agreement (KIA) of 0.91 are fairly
reasonable and good accuracy levels. However, it is felt that there is still much misclassification
that can be improved upon. It is hoped that this can be improved by exploiting some class related
features and topological relationships.

Histogram for this classification is given in figure 1.



Statistics of the classified image are given in table 3.

Figure 1: Histogram of the classified image

Table 3: Statistics of classification result

Land cover lasses Pixel number Pixel no. P(%) Area(Sq.Km)

1)Urban 152746 10.54 34.6

2)Vegetation(Forestry) 50326 3.47 11.4

3)Others(Water, Rocky 46795 3.23 10.6
area etc..)

From the Histogram of this classification it is clear that out of the 58 sq.kms of the study area
the urban area covers 34.6 sq.km which includes residential, commercial, industrial, traffic and
transportation, public utility etc., the vegetation (trees, plants, shrubs etc.,) covers 11.4 sq.km and
water, rocky area etc., covers 10.6 sq.km.

3. CONCLUSIONS, RECOMMENDATIONS AND PROPOSALS

3.1 PLANNING EFFORTS

The way with which the city is growing and developing due to the migration of population from
rural areas for employment and other opportunities, it has been proposed that the ultimate land



use structure of the Vijayawada urban area in the coming 20 years should be around 130 sq.km.
The residential area is proposed to cover about 48% followed by transport and recreation uses.
The land use pattern for the coming 20 years should definitely be far more balanced compared to
the prevailing situation if the authorities concerned look in to the following recommendations
and proposals.

3.2 RECOMMENDATIONS AND PROPOSALS

• The proposals aim at municipal performance improvement of environmental
infrastructure and aims at socio-economic development.
• The proposals for municipal reforms are aimed at enhancing the efficiency, effectiveness
and service delivery with accountability.
• The reform proposals should include privatization of advertisement tax collection,
revenue improvement, town development, operation and maintenance of critical
infrastructure investment.
• The environmental infrastructure proposals aim at improvement of infrastructure in the
prioritized poor settlements as per poverty and infrastructure deficiency matrices and
linked infrastructure for poor settlements.
• These include rehabilitation of existing infrastructure provision of water supply, roads,
drains, sanitation and street lighting based on community prioritization and construction
of drains to improve the living environment.
• The social development proposals aim at addressing the socio-economic needs identified
and prioritized through participatory micro planning process.
• These proposals cover areas of health, education, livelihood, vulnerability and
strengthening of SHGs (Self help groups), with focus on gender issues.
• This leads to the reduction of poverty and improvement in living conditions of the people
in the poor settlements.

4. REFERENCES

1. “Erdas Imagine, 8.5”, User Guide, 2001
2. T.M. Lillesand, and R.W. Kiefer. “Remote Sensing and Image Interpretation”.
4th ed, John Wiley and Sons, inc. USA, 2001, ISBN: 0471255157, 2001
3. Mather, M.Paul. “Computer Processing of Remotely-Sensed Images”. St
Edmundsbury Press Ltd., Bury St Edmunds, Suffolk, Wiley and Sons, ISBN: 0-
471-90648-4,1987
4. “ECognition user guide”, Concept and Methods, 2001
5. U.C. Benz, P. Hofmann, G. Willhauck, I. Lingenfelder, M. Heynen. “Multi-
resolution, object-oriented fuzzy analysis of remote sensing data for GIS - ready
information”. ISPRS Journal of Photogrammetry and Remote Sensing, 58: 239-
258, 2004
6. N. Goward, F. Huemmrich, and H. Waring. “Visible-near infrared Spectral
reflectance of landscape components in western Oregon”. Remote Sensing
environment, 47:190-203, 1994



7. D. Klimešová, E. Ocelíková. “Spatial data in land management and local
government”. Proceedings of the 4th Conference of the European Federation for
Information Technology in Agriculture. Debreceen, 363-368, 2003
8. S.P.S. Kushwaha, S.K. Subramanian, G.CH. Chennaiah, J. Ramana Murthy,
S.V.C. Kameswara Rao, A. Perumal, and G. behera. “International Remote
Sensing and GIS methods for sustainable rural development”. International
Journal of Remote Sensing, 17: 3055-3069, 1996
9. J.C. Price. “How unique are spectral signatures?”. Remote Sensing of
Environment, 49: 181- 186, 1994


Amjad Rehman Khan & Dzulkilfli Muhammad

A Simple Segmentation Approach for Unconstrained Cursive
Handwritten Words in Conjunction with the Neural Network.

Amjad Rehman Khan amjadbzu2003@yahoo.com
PhD researcher
Department of Computer Graphics and Multimedia
University Technology Malaysia
Skudai, 81310, Malaysia

Dr. Zulkifli Mohammad dzullkilfli@utm.my
Associate Professor
Department of Computer Graphics and Multimedia
University Technology Malaysia
Skudai, 81310, Malaysia

Abstract

This paper presents a new, simple and fast approach for character segmentation
of unconstrained handwritten words. The developed segmentation algorithm
over-segments in some cases due to the inherent nature of the cursive words.
However the over segmentation is minimum. To increase the efficiency of the
algorithm an Artificial Neural Network is trained with significant amount of valid
segmentation points for cursive words manually. Trained neural network extracts
incorrect segmented points efficiently with high speed. For fair comparison
benchmark database IAM is used. The experimental results are encouraging.

Keywords: Image analysis, Segmentation, Neural Network, Preprocessing, Pattern matching.

1. INTRODUCTION
An extensive research has been done in the field of handwriting recognition in the last few
decades [1]. It seems that the research has been reached to its maturity for the recognition of
isolated characters recognition, hand printed words recognition, automatic address processing
and bank check reading (holistic approaches) [2-4]. In contrast for the analytical approaches
where the word is segmented into its component characters, recognition results for unconstrained
handwriting is still low due to the poorly segmented words. Segmentation errors mislead classifier
during character recognition [5-7]. In fact segmentation problem has persisted for nearly as long
as handwriting recognition problem itself. In literature, segmentation algorithms for unconstrained
handwritten words can be generalized into two categories.[7] External segmentation and Holistic
segmentation. In the former category letter boundaries are determined prior to recognition while
in the latter, segmentation and recognition are carried out at the same time and the final character
boundaries are determined dynamically by semantic analysis and classification performance.[8-
10].
Higher the segmentation accuracy, the more beneficial it is to the recognition rates [11]. Hence
the segmentation is the backbone of the recognition process and is still active research topic.
Researchers have acknowledged the important role that segmentation plays in handwriting
recognition process [7, 12-13]. That is why more innovative, accurate and fast methods need to
be employed and compared to the work of other researchers using benchmark databases.



In most of the existing segmentation algorithms, human writing is evaluated empirically to deduce
rules [15]. Sometimes the rules derived are satisfactory but there is no guarantee for their
optimum results in all style of writing. Because human writing varies from person to person and
even for the same depending on mood, speed, environment etc. On the other hand researchers
have employed artificial neural networks, hidden Morkov models, statistical classifiers etc to
extract rules based on numerical data [16-21, 36-37]
This research attempts to integrate, rule based segmentation approach and intelligent method for
the character segmentation of unconstrained handwritten words.
A simple but efficient, rule based segmentation algorithm is presented that performs character
segmentation very well with high speed but some characters are over-segmented. Therefore an
ANN is integrated with the proposed approach as artificial neural networks have been
successfully used in the field of pattern recognition [14, 20-23]. To verify the segmentation points
marked by proposed algorithm, an artificial neural network is trained with correct and incorrect
segmentation points for the words images taken from benchmark database [24].
The rest of the paper is organized in four sections. Section 2 presents proposed segmentation
algorithm along with segmentation results. In section 3, neural based experimentation is
performed and results are discussed. Finally, conclusion and future work is drawn in section 4.

2. PROPOSED SEGMENTATION APPROACH
In this section segmentation algorithm and preprocessing steps are presented. Artificial neural
network is trained manually for the correct and incorrect segmentation points obtained from the
proposed segmentation technique. MATLAB 7.0 is used for all experiments performed on system
of 1.6 GHz processor and 1 GB DDR RAM.

2.1 Preprocessing and proposed segmentation algorithm
The original grey scaled image is binarized using Otsu algorithm by selecting automatically a
threshold value for a given image [22]. If required, following binarization, slant correction is
performed [23]. Finally, image is converted to skeleton format to allow users verity of writing
device, pen tilt and to suppress extra data. The proposed segmentation algorithm is explained in
the figure 1.

Step 1. Take word image from database.
Step 2. Perform pre-processing.
Step 3. Calculate sum of foreground pixels (white pixels) for each column. Save those columns
as candidate segment column (CSC) for which sum is 0 or 1 only.
Step 4. By previous step, we have more candidate segmentation columns than actual required.
Hence threshold (approximate character width) is selected empirically from candidate
segment columns to come out with actual segment columns.

FIGURE 1: Proposed Segmentation Algorithm

Due to the simplicity of the proposed segmentation technique, it is very fast and performs well in
most of the cases. For few characters such as m, n, u, v and w over segmentation occurs and
this technique fails to find accurate character boundaries. Segmentation results by the proposed
segmentation technique are shown in figure 2.



FIGURE 2: Segmentation Results by Proposed Segmentation Approach.

It can be seen from the results that segmentation is good except for few characters, where over
segmentation occurs. Therefore it is required to integrate this technique with some intelligent
method to increase its performance. In this regard a trained neural network is employed.
It is mention worthy that over segmentation is minimum and occurs for few characters only.
Hence it lessened burden of the classifier used and therefore processing speed increased.

2.2 Handwriting database
For the fair comparison, patterns are selected from IAM V3.0 benchmark database [24]. A few
samples for segmentation, training and testing of the ANN are shown in figure 3. The reason for
selecting this database is that, it is freely available for researchers to comparing their results.

FIGURE 3: Samples of Word Images from IAM Database

3. EXPERIMENT AND RESULTS

3.1 Training Artificial Neural Network.
A simple program in MATLAB 7.0 is developed to detect co-ordinates of all segmentation points
given birth by the proposed segmentation technique for each pattern. These segmentation points
are divided into correct and incorrect categories manually and stored in a training file. Data is
preprocessed prior to use for ANN training.
For training, ANN with standard back propagation algorithm is used. A number of experiments
with different structures, weights, epochs, momentum and learning rate are performed to enable
ANN to distinguish between correct and incorrect segmentation points. The ANN trained with
25072 training patterns (segmentation points) taken from 2678 words. The optimal structure of
ANN thus found contained 235 to 310 inputs, 25 to 38 hidden units and one output (correct or
incorrect segmentation point) with 300 epochs. Learning rate and momentum was set to 0.2 and
0.6 respectively. MATLAB 7.0 is used for implementation. Trained neural network operates in
figure 4.

FIGURE 4: Incorrect Segmentation Points are rejected by Trained Neural Network.



3.2 Performance of the neural rule-based Segmentation Technique.
After training, the testing phase occurs. For testing phase 2936 samples are selected from the
IAM database. These new patterns are segmented by the proposed algorithm. Segmentation
points thus obtained are fed to the train ANN for their classification into correct and incorrect
categories. Finally, correct are left and incorrect are rejected by trained ANN as shown in figure 4.
Segmentation results for test set are presented in table 1.

Correct segmentation rate. 2678/2936
% Correct segmentation rate. 91.21 %
% Miss segmentation rate. 5.38 %
% Over segmentation rate. 3.20 %

TABLE 1: Segmentation Rates

3.3 Analysis and discussion of results
The neuro rule-based segmentation algorithm achieved recognition rate of 91.21% for valid
identification of 2936 segmentation points pattern. Two problems are found during the analysis of
the results. Firstly, noisy characters, so some additional preprocessing is done before training
ANN. Secondly, touched/ overlapped characters. This type of problem is very hard to deal with.
When two characters are tight together, ligatures can’t be found and therefore they can’t be
segmented. Hence overall correct segmentation results decreased.
It is very hard to compare segmentation results with the other researcher because segmentation
is intermediate process of recognition. In addition to that many researchers report recognition
rates only. Moreover different researcher used different database and report results under some
constrains.
Segmentation results are reported in literature is presented in table 2 for fair comparison.

Author Segmentation Segmentation Database used Comments
method rate
Tappert et al [25] Feature based + Rule 81.08% CEDAR number of words
based not mentioned
Srihari [26] ANN 83% Handwritten zip No alphabetic
codes
Han and Sethi [27] Heuristic algorithm 85.7 % Latin handwritten Only 50 mail
Words on 50 real envelopes are
mail envelopes taken.
Lee et al [28] ANN 90% Printed latin Printed
alphanumeric alphanumeric
characters characters used
Eastwood et al ANN 75.9 % Cursive latin 100,000 training
[29] handwritten from pattern used
CEDAR database
Blumenstein and ANN + conventional 81.21 % 2568 words from
Verma [30] method CEDAR
Yanikoglu and Linear Programming 97 % 750 words No bench mark
Sandon [31] database used
Nicchiotti and Rule-based 86.9 % CEDAR 850 words used
Scagliola. [32] only.
Verma and Gader Feature based + 91 CEDAR words number not
[33] ANN mentioned

Blumenstein and Feature based+ ANN 78.85% CEDAR words number not
Verma [34] mentioned
Verma[35] Feature based + ANN 84.87 CEDAR 300 words only
Cheng et al [36] Feature based + ANN 95.27 CEDAR 317 words only
Cheng et al [37] Enhanced feature 84.19 CEDAR 317 words only
based+ ANN



TABLE 2: Comparison of Segmentation Results in the Literature.

4. CONCLUSION & FUTURE WORK
Segmentation is the important step of analytical approaches employed to handwritten word
recognition. Hence it is the base of most modern approaches. It is admitted fact that no
segmentation method can directly locate character location accurately without an intelligent
method. In this paper, proposed segmentation algorithm is integrated with neural network using
standard back propagation. Initial experiments exhibit very encouraging results with segmentation
accuracy up to 91.21%. Speed is another important factor, overlooked in many past researches.
Due to the minimum over segmentation, neural network is least burdened and therefore speed is
optimum. This paper has briefly described one stage on our progress towards final goal of
unconstrained cursive words recognition with higher recognition rate and speed.

ACKONOLOGEMENT
This research work is fully supported by the Ministry of Science, Technology and Innovation
Malaysia under grant vote 79258. The author is also thankful to colleague researcher Fajri
Kurniawan for his help in experimentation.

5. REFERENCES

1. H.Bunke. “Recognition of cursive roman handwriting, past, present and future”. In
Proceeding of 7th International Conference on Document Analysis and Recognition, 448-
461, 2003.

2. C.Y.Suen, R.Legault, C. Nadal, M.Cheriet and L.Lam. “Building a new generation of
handwriting recognition systems”. Pattern Recognition Letters, 14: 305-315, 1993.

3. S.W. Lee. “Multilayer cluster neural network for totally unconstrained handwritten
numerical recognition”. Neural Networks, 8: 783-792, 1995

4. H. I. Avi-Itzhak, T. A. Diep and H.Garland. “High accuracy optical character recognition
using neural networks”. IEEE Trans. Pattern Analysis and Machine Intelligence, 18: 648-
652, 1996.

5. S.B. Cho. “Neural networks classifiers for recognition totally unconstrained handwritten
numericals”. IEEE Trans. On Neural Networks, 8.

6. S.W. Lee. “Off-line recognition of totally unconstrained handwritten numericals using
multilayer cluster neural network”. IEEE Trans Pattern Analysis and Machine Intelligence,
18: 648-652, 1996.

7. R.G. Casey, E. Lecolinet. “A survey of methods and strategies in character
segmentation”. IEEE Trans. Pattern Analysis and Machine Intelligence, 18: 690-706,
1996.

8. R. Farag. “Word–level recognition of cursive script”. IEEE Trans. Computing 28: 172-175,
1979.

9. J. M. Bertille, M. E. Yacoubi. “Global cursive postal code recognition using hidden Marko
models”. In Proceeding of the First European Conference Postal Technology, France,
129-138, 1993.

10. J. Wang, J. Jean. “Segmentation of merged characters by neural networks and shortest
path” Pattern Recognition 27(5): 649-658, 1994.



11. C. K. Cheng, M. Blumenstein. "The neural-based segmentation of cursive words using
enhanced heuristics". In Proceedings of Eighth International Conference on Document
Analysis and Recognition (ICDAR'05), 650-654, 2005.

12. S.N. Srihari. “Recognition of handwritten and machine printed text for postal address
interpretation”. Pattern recognition letters, 14: 291-302, 1993.

13. M. Gilloux. “Research into the new generation of character and mailing address
recognition systems at the French post office research center”. Pattern Recognition
Letters.14: 267-276 1993.

14. M. Blumenstein, B. K. Verma. “An artificial neural network based segmentation algorithm
for off-line handwriting recognition”. In proceedings of International Conference on
Computational Intelligence and Multimedia Applications (ICCIMA’ 98), Gippsland,
Australia, 1997.

15. X. Xiao, G. Leedham. “Knowledge–based English cursive script segmentation”. Pattern
recognition letters 21: 945-954, 2000.

16. M. Giloux. “Hidden Markov Models in Handwriting Recognition”. Fundamentals in
Handwriting Recognition, S.Impedovo ed., NATO ASI Series F: Computer and System
Science, 24: Spinger Verlang, 1994.

17. M. EI. Yacoub, M. Gilloux and J. M. Bertille. “A statistical approach for phrase location
and recognition with in a text Line”. An application to Street Name Recognition, IEEE
Trans. Pattern Analysis and Machine Intelligence, 24(2): 172-188, 2002.

18. A. Khotanzad, J. Lu. “Shape and texture recognition by a neural network”. Artificial Neural
Networks in Pattern Recognition, Elsevier Science Publishers B.V., Amsterdam,
Netherlands, 109-131, 1991.

19. B. Zheng, W. Qian and L.Clarke. “Multistage neural network for pattern recognition in
mammogram screening”. IEEE ICNN, Orlando, 3437-3448, 1994.

20. A.D. Kulkarni. “Artifical neural networks for image understanding”. Van Nostrand
Reinhold, New York. 154-270, 1994.

21. K. Han, I. K. Sethi. “Handwriting signature retrieval and Identification”. Pattern
Recognition Letters, 17, 83-90, 1996.

22. N. Otsu. “A threshold selection method from gray level histograms”. IEEE transactions on
systems, Man and Cybernetics, 9(1): 62-66, 1979.

23. S. Knerr, E. Augustin. “A neural network–hidden markov model hybrid for cursive word
recognition”. In Proceedings of International Conference on Pattern Recognition,
Brisbane, 2: 1518-1520, 1998.

24. U. Marti, H. Bunke. “The IAM database: An English sentence database for off-line
handwriting recognition”. International Journal of Document Analysis and Recognition,
15: 65-90, 2002.

25. C. C. Tappert., C. Y. Suen and T. Wakahara. “The state of the art in on-line handwriting
recognition”. IEEE Trans. Pattern Analysis. Machine. Intelligence. 12: 787-808, (1990)



26. S. N. Srihari. “Recognition of handwritten and machine-printed text for postal address
Interpretation”. Pattern Recognition Letters 291-302, 1993.

27. K. Han, I. K. Sethi. “Off-line cursive handwriting segmentation”. ICDAR 95, Montreal,
Canada, 894-897, 1995

28. S-W. Lee, D-J. Lee and H-S Park. “A new methodology for gray–scale character
segmentation and recognition”. IEEE Transaction on Pattern Analysis and Machine
Intelligence, 1045-1051, 1996.

29. B. Eastwood, A. Jennings and A. Harvey. “A feature based neural network segmenter for
handwritten words”. ICCIMA’97 Australia, 286-290. 1997

30. M. Blumenstein, B. Verma. “A segmentation algorithm used in conjunction with artificial
neural networks for the recognition of real-word postal addresses”. In Proceeding of
International Conference on Computational Intelligence and Multimedia Applications
(ICCIMA’97), Gold Coast, Australia. 155-160, 1997.

31. B. Yanikoglu, P.A.Sandon. “Segmentation of off-line cursive handwriting using linear
programming”. Pattern Recognition, 31: 1825-1833, 1998.

32. G. Nicchiotti, C.Scagliola. “Generalized projections: a tool for cursive handwriting
normalisation”. In Proceedings of 5th International Conference on Document Analysis
and Recognition, Bangalore, 729-733, 1999.

33. B. Verma, P. Gader. "Fusion of multiple handwritten word recognition techniques". Neural
Networks for Signal Processing X, 2000. In Proceedings of the IEEE Signal Processing
Society Workshop, 2: 926-934, 2000.

34. M. Blumenstein, B. Verma. "Analysis of segmentation performance on the CEDAR
benchmark database". In Proceedings of Sixth International Conference on Document
Analysis and Recognition (ICDAR'01), 1142, 2001.

35. B. Verma. "A contour character extraction approach in conjunction with a neural
confidence fusion technique for the segmentation of handwriting recognition”. In
Proceedings of the 9th International Conference on Neural Information Processing. 5: 18-
22, 2002.

36. C. K. Cheng., X. Y. Liu., M. Blumenstein and V. Muthukkumarasamy. “Enhancing neural
confidence-based segmentation for cursive handwriting recognition”. In Proceeding of 5th
International Conference on Simulated Evolution and Learning (SEAL '04), Busan, Korea,
SWA-8, 2004.

37. C. K. Cheng, M. Blumenstein. “Improving the segmentation of cursive handwritten words
using ligature detection and neural validation”. In Proceedings of the 4th Asia Pacific
International Symposium on Information Technology (APIS 2005), Gold Coast, Australia,
56-59, 2005.


International Journal of Image Processing (IJIP) Volume (2) Issue (3)

More Related Content

Viewers also liked (9)

Similar to International Journal of Image Processing (IJIP) Volume (2) Issue (3) (20)

International Journal of Image Processing (IJIP) Volume (2) Issue (3)