SVM CLASSIFIER FOR THE PREDICTION OF ERA OF AN EPIGRAPHICAL SCRIPT

International Journal of Peer to Peer Networks (IJP2P) Vol.2, No.2, April 2011
DOI : 10.5121/ijp2p.2011.2202 12
SVM CLASSIFIER FOR THE PREDICTION OF ERA OF
AN EPIGRAPHICAL SCRIPT
Soumya A and G Hemantha Kumar
Department of Computer Science & Engineering, R V College of Engineering,
Karnataka, India
soumyaa@rvce.edu.in
Department of Studies in Computer Science, University of Mysore, India
ghk2007@yahoo.com
ABSTRACT
Inscriptions are the main source for reconstructing the history and culture of ancient civilizations. The
scripts of modern Indian languages have evolved over centuries and finally transformed to the present
form. Modern readers find difficulty in interpreting a script of olden days. The characters have changed
over time. Hence for reading ancient scripts the period has to be determined, so as to have knowledge of
which character set of ancient days is to be employed for automatic reading. Prediction of the era of a
given ancient script is a follow-on member of an ancient script recognition system and can be used as a
component of the OCR system for ancient scripts. This knowledge can be used by archaeologists and
historians for further explorations. In this paper we demonstrate period identification of various ancient
Kannada scripts using SVM classifier. A system is proposed for prediction of the era and it is being done
by examining a few characters in Kannada script of various periods referred to as test characters. These
test characters are sampled from the script automatically and matched with the characters available for
different periods using machine intelligence.. This classifier is tested on quite number samples of
Kannada epigraphical document images belonging to nine different periods.
KEYWORDS
Inscription, Epigraphy, Paleography, Support Vector Machine (SVM), Optical Character Recognition
(OCR)
1. INTRODUCTION
Inscriptions have remarkable importance to mankind. The information culled out from the
inscriptions provides us the knowledge of history, culture, astronomy, medicine, management,
political, religious, social, economic, tax, administrative, educational conditions and also
languages and scripts that prevailed during yester years[1][2][3][4] . Epigraphy is the science
of identifying the inscriptions on rocks, pillars, temple walls, copper plates and other writing
materials. It is a primary tool of archaeology when dealing with literate cultures. Many
inscriptions do not contain enough historical details to fix their authorship conclusively. For
example, from the inscriptions with the name Rajaraja, it is not always clear whether Rajaraja
the First or the Second or the Third is intended. To assign dates to such inscriptions and to
identify the kings, palaeography is the main tool. Paleography is the study of ancient
handwriting and the practice of deciphering and reading historical manuscripts. The
paleographer must have the knowledge of: first, the language of the text and second, the
historical usages of various styles of handwriting, common writing customs, and scribal/notarial
abbreviations.

13
Scripts denote the writing systems employed by the languages. Any language can be written in
any script. Having or not having 'own script' is neither a status nor any impediment for
language. Three important varieties of scripts that were prevalent in ancient India were: Indus
valley script, Brahmi Script and Kharosti script. The scripts of modern Indian languages have
evolved from one of these scripts over the centuries. The evolution of the script is dependent on
many factors: the writing material, (Stone, Copper, Palm leaf, Paper etc), writing tools, modes
of writing and the background of the scribes. In India, prior to invention of writing / printing
papers, Palmyra leaves and birch leaves were used for writing purposes. As they could not be
long lasting, engraving on rocks, pillars and plates made of copper/ gold / silver came into
practice. In India currently there are 13 scripts and 23 official languages for communicating at
state level. Apart from these, there are many languages & dialects used by number of people.
The Kannada script has been used to write in Kannada language. Kannada which belongs to
highly acclaimed Dravidian language family is the official language of the State of Karnataka
and is one of the most enriched languages in India with is long historical heritage. Its earliest
written records dates back to about the third century B.C. and the modern script has evolved
gradually from the ancient script known as Brahmi. Kannada script has been evolved in its
present form, undergoing several twins and turns. In fact, the Indian linguists have demarcated
the whole of this evolutionary process in to four broad phases.
• Poorvada Halegannada or Pre-ancient Kannada - The first written record in the
Kannada language is found in Emperor Ashoka`s Brahmagiri decree, which dated back
to as early as 230 BC.
• Halegannada or Ancient Kannada- marks the second phase, covering a time period of
the 9th to 14th centuries CE.
• Nadugannada or Middle Kannada- third phase started from the 14th century and
continued till 18th century CE.
• Hosagannada or Modern Kannada – the Kannada works that were produced at the
end of the 19th
century and also much later are categorized as Hosagannada or Modern
Kannada.
Kannada script has traveled a long way from the earlier Brahmi model as indicated in Figure 1.
The fonts have evolved over the centuries and it has undergone many changes till now. It has
undergone a number of changes during the regimes of Shatavahana, Kadamba, Ganga,
Rashtrakuta, Chalukya, Hoysala,Vijayanagara and Wodeyar dynasties, and transformed finally
into a shape which we are using today [2]. This poses a challenge in understanding these
ancient scripts. The stages of evolution, though generally understood by epigraphists, are not
known to the educated layman of that province. There was also a certain amount of regional
variation. Thus even for the experts, it is difficult to assign dates to many inscriptions, whether
complete or fragmentary. The expert epigraphists, who decipher these scripts and translate them
into the regional languages, are expected to become extinct in near future and also the
importance of inscriptions to mankind is remarkable. Hence there is a dire need for automation
of classification and recognition of ancient epigraphical scripts. This work helps us in
integrating the new technological concepts with underlying cultural features of the scripts,
thereby minimizing the divide between “Man and the History”. The age of the document
reflects to which era the script belonged to and the particular dynasty in which it
existed.
In automation of deciphering epigraphical scripts, the characters present in the script are to be
identified and recognized. The characters have evolved over years and transformed to the shape
we are using today. Every era has its own character set, which is to be determined before

International Journal of Peer to Peer Networks (IJP2P) V
recognition. Hence there is a need for dating of inscriptions to a particular period, which is the
initial step towards decipherment.
The ancient scripts are engraved on rocks, copper, silver
inscriptions are captured and later subjected to digitization process. The script of document
images that has been digitized is
addresses off line character recognition.
practical interest in the extraction of discriminating and invariant information from a
handwritten specimen. One of the major difficulties in off
the great variations observed in different samples of scripts from the same writer over time or
from different scriptors. There is no perfect mathematical model that can describe such extreme
variations and hence it is very difficult
different writing styles.
Figure 1: Changes in character shape since 3
2. LITERATURE REVIEW
High accuracy OCR systems are reported for English with excellent performance in presence of
printing variations and document degradation.
as there are only 26 letters and each letter is quite distinct from others compared to recognition
of Indian language characters. For Indian and many other
not yet able to successfully recognize printed / handwritten document images of varying scripts,
quality, size, style and font. Compared to European languages, Indian languages pose many
initial step towards decipherment.
scripts are engraved on rocks, copper, silver or gold plates. The images of these
that has been digitized is to identified and later recognized. Hence the present problem
recognition. Handwriting identification and recognition are of great
handwritten specimen. One of the major difficulties in off-line word recognition originates from
e great variations observed in different samples of scripts from the same writer over time or
very difficult to find the characteristic features that are invariant with
Changes in character shape since 3rd
century B.C to
18th
century A.D
printing variations and document degradation. Recognizing English characters is much simpler
For Indian and many other oriental languages- OCR systems are
14
The images of these
Hence the present problem
Handwriting identification and recognition are of great
line word recognition originates from
e great variations observed in different samples of scripts from the same writer over time or
ic features that are invariant with
English characters is much simpler
OCR systems are

15
additional challenges.Many researchers have been working on script recognition for more than
three decades but there are very few tools to identify these scripts. Indian languages are
characterized with the properties:
(i) Large number of vowels, consonants, and conjuncts.
(ii) Have a base character along with vowels attached, forming single character called the
compound character.
(iii) Most scripts spread over several zones.
(iv) Lack of standard test databases of the Indian languages.
There are many conventional OCR systems available for present day Kannada Script, but very
few work on ancient Kannada epigraphical scripts are reported. Ancient Script recognition
poses additional challenges.Some of researchers are making special efforts to make this
knowledge more easily available.Nevertheless, it remains to be one of the most challenging
problems in pattern recognition.
Anasuyadevi [2000] has worked on recognition of ancient Indian scripts [5]. She has proposed
[2003] a fuzzy neural network approach for the recognition of Brahmi characters.
A hybrid neural network architecture for age identification of ancient Kannada Scripts was
proposed by K Harish Kashyap, Bansilal, P Arun Koushik [2003]. After pre-processing the
characters, the work is implemented in two phases. The first phase which identifies the base
character incorporates an Artificial Neural Network (ANN). ANN is trained by Back
propagation algorithm to identify the present day base character corresponding to input
character. In the second phase, for identification of age pertaining to the base character, a
Probabilistic Neural Network (PNN) - a Bayesian classifier is used taking the advantage that no
training is involved prior to classification [6].
The research work by Srikanta Murthy K [2005] for transforming epigraphical objects into
machine recognizable form, involves the preprocessing techniques - for removal of noises,
segmentation of lines and characters, thinning and classification of the epigraphical document
belonging to different period [16]. Two preprocessing techniques for removal of noise have
been proposed – the first algorithm is based on a rectangle fitting wherein the height of the
character to be retained is assumed to be greater than the noisy pixel. The second algorithm
employs a template to obtain the minimum majority of white pixels. Segmentation of lines and
characters are carried out using- a Partial Eight Direction Based Line Segmentation (PEBLS)
algorithm wherein horizontal Projection profile is applied to identify the base and
supplementary reference lines. The second approach is based on Nearest Neighbor Clustering
(NNC), could be used even when the document is skewed. Three thinning algorithms - two-step
algorithm, fully parallel thinning algorithm and rotation invariant 4- step algorithm have been
designed. Classification of the epigraphical document belonging to different period is carried
out using - a method based on texture features, a method based on invariant moments which are
invariant to rotation, translation and scaling, and for accurate estimation of the period, a neural
network based approach is adopted [17][18][19].
3. PROPOSED METHODOLOGY
Age identification of ancient scripts is important, which enables to know which character set is
to be employed for automatic reading of age old scripts. It is a problem characterized under the

International Journal of Peer to Peer Networks (IJP2P) V
genre of pattern recognition and image analysis
classification of epigraphical document images
The Age prediction classifier involves the following steps:
I. Image Acquisition: The pr
the inputs. These objects may be camera grabbed or scanned if the inscriptions are
engraved on a plate. The photographs of inscriptions on rock or pillars may not result in
good objects. Hence estempages are produced [Dani A.H, 1986; Sirc
later scanned to get the object image of the inscription.
II. The input image of the inscription may be degraded due to the presence of the broken
characters, erased characters, touching
irrelevant symbols engraved by the scribes and so on
between the lines and characters of epigraphical document and the skew could
complicate the process of deciphering the script.
preprocessed for removal of noise
characters.
III. Appropriate Feature extraction method is devised
information contained in a pattern so that the
by a formal procedure.
The features extracted have the following properties:
• Easy to extract, which reduces the complexity of the program.
• Distinct, which eases the
• Independent of font type and size, which is a big
Figure 2. Age Identification S
There are totally14 features extracted from the character of which 4 of them are for the whole
image as listed below:
genre of pattern recognition and image analysis. Figure 2 illustrates the steps towards
classification of epigraphical document images to their respective era.
The Age prediction classifier involves the following steps:
The present work proposes to take the objects of the inscriptions as
estempages are produced [Dani A.H, 1986; Sircar D.C, 1965] and
to get the object image of the inscription.
The input image of the inscription may be degraded due to the presence of the broken
characters, erased characters, touching characters, distortion due to fossils settled,
irrelevant symbols engraved by the scribes and so on. The non uniform spacing
complicate the process of deciphering the script. Hence the input document image is
removal of noise and skew correction, followed by segmentation of
Feature extraction method is devised for measuring the relevant shape
information contained in a pattern so that the task of classifying the pattern is made easy
The features extracted have the following properties:
• Easy to extract, which reduces the complexity of the program.
• Distinct, which eases the classification process.
of font type and size, which is a big advantage.
Figure 2. Age Identification System
14 features extracted from the character of which 4 of them are for the whole
16
. Figure 2 illustrates the steps towards
the inscriptions as
ar D.C, 1965] and
The input image of the inscription may be degraded due to the presence of the broken
characters, distortion due to fossils settled,
The non uniform spacing
the input document image is
segmentation of
for measuring the relevant shape
task of classifying the pattern is made easy
14 features extracted from the character of which 4 of them are for the whole

17
1) Height / Width
2) Number of black pixels / number of white pixels image
3) Number of horizontal transitions
4) Number of vertical transitions
In addition, the image is divided into four regions as shown in Figure 3 and the following
features are extracted from these regions:
5) Black Pixels in Region 1/White Pixels in Region 1
9) Black Pixels in Region 1/Black Pixels in Region 2
IV. The characters have evolved gradually over a course of time, leading to varying
character sets from time to time. Hence the era to which an epigraphical script belongs
to has been predicted so as to further enable automatic recognition of the epigraphical
characters. For the classification of the script to their respective period, a Support
Vector Machine (SVM) Classifier using linear kernel function is devised. The
principle of SVM is to map the input data onto a higher dimensional feature space and
determine a separating hyper plane with maximum margin between two classes in the
feature space. We use SVM as a multiclass-classifier system which is trained on the
distinct/unique characters in the Kannada script of various periods. Each vector is given
to SVM which classifies it into a class in the vector space. The important criteria for
deciding the final character are: the number of points belonging to a particular class and
the minimum distance from a class. It represents the degree to which a particular
snapshot resembles a character. We search for the unique (test) characters identified
aprior, of various periods in the given ancient script. The number of matches of these
unique characters of different era, with the characters of input document image is
accounted. The period of input epigraphical document image of ancient Kannada script
is the one with maximum number of matches of characters of input ancient script with
unique (test) characters. Thus the input document image is classified to the period of
any of these dynasties: Ashoka dynasty , Shatavahana dynasty , Kadamba dynasty
,Badami Chalukya dynasty , Rashtrakuta dynasty, Kalyan Chalukya dynasty, Hoyasala
dynasty, Vijayanagar dynasty or Mysore Wodeyar dynasty.

18
.
Figure 3. Dividing the image to 4 regions and extracting features
V. Display the era of the Script
Thus the input epigraphical document image is classified to the period of any of the
nine dynasties that prevailed in ancient Karnataka: Ashoka dynasty , Shatavahana
dynasty , Kadamba dynasty ,Badami Chalukya dynasty , Rashtrakuta dynasty, Kalyan
Chalukya dynasty, Hoyasala dynasty, Vijayanagar dynasty or Mysore Wodeyar dynasty
4. IMPLEMENTATION AND EXPERIMENTAL RESULTS
We have implemented the Age identification System on Window’s Xp Operating System, using
MATLAB version 7.8.0.347(R2009a) on a PENTIUM IV 2.6 GHz processor. The
functionalities of Image processing toolbox is used for developing GUI, preprocessing and
segmenting input document image.
In this section we provide sample experimental results out of several experiments conducted.
We have tested the proposed method on more than 50 samples of ancient Kannada Epigraphical
documents. Figure 4 to Figure 6 illustrates the snapshots of GUI developed, the output of
segmentation and feature extraction, and output display of the classifier respectively. Figure 7 to
Figure 9 depicts sample results of Classification and Display of the period for the given input
epigraphical script.

19
Figure 4. GUI of the Period prediction System for ancient Kannada Script
Figure 5. Snapshot of Segmentation and Feature extraction for input epigraphical document
Figure 6. Snapshot of classification of input epigraphical document into respective period and
corresponding dynasty

20
Input epigraphical document Output image with details pertaining
to the period predicted
Figure 7. Brahmi Script of Ashoka Dynasty being dated
Figure 8. A Sample Kannada Script of Hoysala Dynasty being dated
Figure 9. A Sample Kannada Script of Wodeyar Dynasty being dated

21
5. CONCLUSION & FUTURE ENHANCEMENT
Epigraphical sources reflect the importance during regime of several rulers. Script
identification is a key step that arises in document image analysis, especially for the documents
which belong to ancient days. The study of Epigraphical scripts is vital in knowing the civilized
past and hence classification of character belonging to various periods is imperative before
using the character bank of the particular period. The proposed system for the prediction of the
era is being done by examining a few characters in Kannada script of various periods referred to
as test characters. These test characters are sampled from the script automatically and matched
with the characters available for different periods using machine intelligence. The proposed
system here has various modules like Preprocessing and Segmentation (involves binarization,
Noise removal, Skew removal & Segmentation), feature extraction and finally classification of
input epigraphical documents into their respective eras using Support Vector Machine. This
classifier is tested on more than 50 samples of Kannada epigraphical document images
belonging to nine different periods. To sum up, the research issue taken here is to produce a
computer perceivable image from a raw epigraphical script which are the inscriptions on rocks
or pillars or plate and then predict the era of the ancient script. Prediction of period of ancient
scripts is the first step in automatically deciphering epigraphical scripts. Automatic period
identification for a given document image, of a script facilitates the selection of the script
specific OCR in an environment where scripts of various periods are given as input. The work
can be further extended to facilitate the computer dating of inscriptions of various other scripts
that were prevalent in ancient India and other non Indian regions, during the regime of various
rulers.
REFERENCES
1. D Dayalan Computer Application in Indian Epigraphy, Bharatiya Kala Prakashan publication
(2005).
2. A.V.Narasimha Murthy - ‘Kannada Lipiya Ugama Mattu Vikasa’, Kannada Adhyayana
Samsthe, Mysore University, Mysore, (1968).
3. Dr M G Manjunath, G K Devarajaswamy – ‘Kannada Lipi Vikasa’, Jagadhguru Sri
Madhvacharya Trust, Sri RagavendraSwamy Matt, Mantralaya
4. Dr. Devarakonda Reddy : Lipiya Huttu Mattu Belavanige — Origin and Evolution of Script,
Published by Kannada Pustaka Pradhikara (Kannada Book Authority), Bangalore
5. Anasuya Devi H.K, (2002), “Automated Recognition of Ancient Indian Scripts”, Proceedings of
National workshop on Computer Vision, Graphics and Image processing, WVGIP, pp 216-219.
6. K Harish Kashyap, Bansilal, P Arun Koushik – ‘Hybrid Neural Network Architecture for Age
Identification of Ancient Kannnada Scripts’, 2003
7. Haidar Almohri. John S. Gray and Hisham Alnajjar – ‘A Real-time DSP-Based Optical
Character Recognition System for Isolated Arabic characters using the TI TMS320C6416T’,
Proceedings of the IAJC-IJME International Conference, 2008
8. Works on OCR for Indian Languages [online] http:// Indira Gandhi National Centre for the Arts
(IGNCA's) Southern Regional Centre, Bangalore
9. Gopal Datt Joshi, Saurabh Garg, and Jayanthi Sivaswamy – ‘Script Identification from Indian
Documents’, H. Bunke and A.L. Spitz (Eds.): DAS 2006, LNCS 3872, pp. 255–267, 2006. c
Springer-Verlag Berlin Heidelberg 2006
10. Manish Kumar – ‘Degraded Text recognition of Gurumukhi Script’, Ph.D thesis, Thapur
University, Patiala (Punjab), March 2008
11. Naveen Garg – ‘Handwritten Gurumukhi Character Recognition using Neural networks’, M.E
thesis, Thapur University, Patiala, June 2009
12. T V Ashwin and P S Sastry – ‘ A Font and size-independent OCR system for printed Kannada
documents using Support Vector Machines’, Sadhana, Vol 27, Part 1, February 2002

22
13. S Pletschacher, J Hu and A Antonacopoulos- ‘ A New framework for Recognition of Heavily
Degraded Characters in Historical Typewritten Documents Based on Semi-Supervised
Clustering’, 10th
International Conference on Document Analysis and Recognition, 2009
14. Sheikh Faisal Rashid, Faisal Shafait and Thomas M. Breuel – ‘Connected Component level
Multiscript Identification from Ancient Document Images’ , Copyright 2010 by the author(s)
15. P. Subashini, M. Krishnaveni and N. Sridevi – ‘ Period Prediction System for Tamil
Epigraphical Scripts Based on Support Vector Machine’, Information Systems for Indian
Languages - Communications in Computer and Information Science, Volume 139, Part 1, 23-30,
2011.
16. Srikanta Murthy.K – ‘Transformation Of Epigraphical Objects Into Machine Recognizable
Image Patterns’, Ph.D Thesis, University Of Mysore, Mysore, December 2005
17. Srikanta Murthy.K - ‘Prediction of Era of Epigraphical scripts based on texture features-A
statistical approach’, Journal of Analysis and Computation (JAC), New Delhi, 2006
18. Srikanta Murthy.K “Invariant Moments Based Feature Extraction for the Classification of
Epigraphical Characters”, Journal of Statistics and Applications, 2005
19. Srikanta Murthy.K , “Neural Network classifier for the prediction of an Epigraphical script”,
Journal of Computer Society of India (CSI), 2004
20. Srikanta Murthy.K “Prediction of Era of Epigraphical scripts based on texture features-A
statistical approach”, National conference on convergence of linguistics, E-Governance & I.T.
for making kannada- A Tech-savvy Language, Shimoga 2006
Authors
Soumya A
Soumya A, Assistant Professor, Department of Computer Science and
Engineering, R.V College of Engineering, Bangalore. She has obtained
M.S degree in Computer Cognition Technology, University of Mysore,
Mysore and B.E in Computer Science and Engineering, Bangalore
University, Bangalore. Her areas of research are Artificial Intelligence,
Soft Computing, Pattern Recognition and Image Processing. She has
guided several Under- graduate Projects and 7 Post Graduate projects.
Dr G Hemantha Kumar
Dr G Hemantha Kumar, Professor & Chairman, Department of studies in
Computer Science, University of Mysore, Mysore. Serving as Course -
Coordinator of Chinese B.Tech Programme, University of Mysore,
Mysore. He was awarded Ph.D. for his thesis titled : "On Automation of
Text Production From Pitman Shorthand Notes" from University of
Mysore, Mysore. His areas of research are: Image Processing, Pattern
Recognition, Numerical Techniques, Bio-Metric. He has to his credits 31
publications in International / National Journals and 46 publications in
International / National Conferences/Workshops. He has guided several
Ph.D candidates and is presently guiding 5 Ph.D candidates.

SVM CLASSIFIER FOR THE PREDICTION OF ERA OF AN EPIGRAPHICAL SCRIPT

More Related Content

Similar to SVM CLASSIFIER FOR THE PREDICTION OF ERA OF AN EPIGRAPHICAL SCRIPT (20)

More from ijp2p (20)

Recently uploaded (20)

SVM CLASSIFIER FOR THE PREDICTION OF ERA OF AN EPIGRAPHICAL SCRIPT