(IJCNS) International Journal of Computer and Network Security, 29

Vol. 1, No. 3, December 2009

Handwritten Tamil Character Recognition Using

SVM
Prof. Dr.J.Venkatesh1, C. Sureshkumar2
1
Assistant Professor – Systems and Production, Anna University Coimbatore, Tamilnadu, India.
venkijv@rediffmail.com
2
Assistant Professor- CSE, J.K.K.M College of Technology, Tamilnadu, India.
ck_sureshkumar@yahoo.co.in

Abstract: Hand written Tamil Character recognition refers to recognition it is easier to recognize English alphabets and
the process of conversion of handwritten Tamil character into numerals than Tamil characters. Tamil has the longest
Unicode Tamil character. The scanned image is segmented into unbroken literary tradition amongst the Dravidian
paragraphs using spatial space detection technique, paragraphs
languages. Tamil is inherited from Brahmi script. The
into lines using vertical histogram, lines into words using
horizontal histogram, and words into character image glyphs
earliest available text is the Tolkaappiyam, a work
using horizontal histogram. Each image glyph is subjected to describing the language of the classical period. There are
feature extraction procedure, which extracts the features such several other famous works in Tamil like Kambar
as character height, character width, number of horizontal Ramayana and Silapathigaram but few supports in Tamil
lines(long and short), number of vertical lines(long and short), which speaks about the greatness of the language. For
horizontally oriented curves, the vertically oriented curves, example, Thirukural is translated into other languages due
number of circles, number of slope lines, image centroid and
to its richness in content. It is a collection of two sentence
special dots. The extracted features considered for recognition
are given to Support Vector Machine (SVM) where the poems efficiently conveying things in a hidden language
characters are classified using supervised learning algorithm. called Slaydai in Tamil. Tamil has 12 vowels and 18
These classes are mapped onto Unicode for recognition. Then consonants. These are combined with each other to yield
the text is reconstructed using Unicode fonts. This character 216 composite characters and 1 special character (aayutha
recognition finds applications in document analysis where the ezhuthu) counting to a total of (12+18+216+1) 247
handwritten document can be converted to editable printed characters.
document. This approach can be extended to recognition and
reproduction of hand written documents in South Indian
languages. 1.2 Vowels
Tamil vowels are called uyireluttu (uyir – life, eluttu –
Keywords: Character recognition, Unicode, Support letter). The vowels are classified into short (kuril) and long
Vector Machines (SVM).
(five of each type) and two diphthongs, /ai/ and /auk/, and
three "shortened" (kuril) vowels. The long (nedil) vowels
1. Introduction are about twice as long as the short vowels. The diphthongs
are usually pronounced about 1.5 times as long as the short
Tamil is an ancient language with a rich literary tradition vowels, though most grammatical texts place them with the
and Ancient India was popular in several fields such as long vowels.
medicine, astronomy and business. Ancient people recorded
their knowledge in various fields in palm leaves. The 1.3 Consonants
handwritten text written in palm leaves decayed over a Tamil consonants are known as meyyeluttu (mey - body,
period of time. It is very difficult to preserve them in the eluttu - letters). The consonants are classified into three
same form. This paper proposes a new approach for categories with six in each category: vallinam - hard,
converting handwritten Tamil script using unicode. The mellinam - soft or Nasal, and itayinam - medium. Unlike
style of writing and the font were different compared to most Indian languages, Tamil does not distinguish aspirated
present day scripts. Lot of software tools is available only to and unaspirated consonants. In addition, the voicing of
read present day printed Tamil text with better recognition plosives is governed by strict rules in centamiḻ. Plosives are
and accuracy. unvoiced if they occur word-initially or doubled. Elsewhere
they are voiced, with a few becoming fricatives
1.1 Tamil Language intervocalically. Nasals and approximants are always
voiced. As commonplace in languages of India, Tamil is
Tamil is a South Indian language spoken widely in characterised by its use of more than one type of coronal
Tamilnadu in India. Handwritten character recognition is a consonants. Retroflex consonants include the retroflex
approximant , which among the Dravidian languages is also
difficult problem due to the great variations of writing
found in Malayalam (example Kozhikode), disappeared
styles, different size and orientation angle of the characters.
from Kannada in pronunciation at around 1000 AD (the
Among different branches of handwritten character
30 (IJCNS) International Journal of Computer and Network Security,
Vol. 1, No. 3, December 2009

dedicated letter is still found in Unicode), and was never size of the input image is as specified by the user and can be
present in Telugu. Dental and alveolar consonants also of any length but is inherently restricted by the scope of the
contrast with each other, a typically Dravidian trait not vision and by the scanner software length.
found in the neighboring Indo-Aryan languages.
2.3 Preprocessing
1.4 Tamil Unicode
This is the first step in the processing of scanned image. The
scanned image is pre processed for noise removal. The
The Unicode Standard is the Universal Character encoding resultant image is checked for skewing. There are
scheme for written characters and text. It defines the uniform possibilities of image getting skewed with either left or right
way of encoding multilingual text that enables the orientation. Here the image is first brightened and binarized.
exchange of text data internationally and creates the The function for skew detection checks for an angle of
foundation of global software. The Tamil Unicode range is orientation between ±15 degrees and if detected then a
U+0B80 to U+0BFF [3].The Unicode characters are simple image rotation is carried out till the lines match with
comprised of 2 bytes in nature. For example, the Unicode the true horizontal axis, which produces a skew corrected
for the character is 0B85; the Unicode for the image.
character is 0BAE+0BC0. The Unicode is designed for
various other Tamil characters.

2. Tamil character recognition functional block

diagram
Figure 2. Histograms for skewed and skew corrected images
The schematic block diagram of handwritten Tamil
Character Recognition system consists of various stages as
shown in figure. They are Scanning phase, Preprocessing,
Segmentation, Feature Extraction, Classification, Unicode
mapping and recognition and output verification.
Figure 3. Original Texts

Scan Document

Preprocessing

Figure 4. Character Segmentation

Segmentation
2.4 Segmentation
Feature Extraction
After pre-processing, the noise free image is passed to
Classification (SVM) the segmentation phase, where the image is decomposed into
individual characters. Fig.4 shows the image and various
Unicode Mapping steps in segmentation.
Algorithm for segmentation:
Recognized Text (1) The binarized image is checked for inter line spaces.
(2) If inter line spaces are detected then the image is
segmented into sets of paragraphs across the interline gap.
Figure 1. Handwritten Character Recognition System (3)The lines in the paragraphs are scanned for horizontal
space intersection with respect to the background. Histogram
2.1 Character Recognition Functions—Phase I
of the image is used to detect the width of the horizontal
lines. Then the lines are scanned vertically for vertical space
This phase includes the scanning, preprocessing,
segmentation and feature extraction. intersection.
Here histograms are used to detect the width of the words.
2.2 Scanning Then the words are decomposed into characters using
character width computation.
A properly printed document is chosen for scanning. It is
placed over the scanner. A scanner software is invoked which 2.5 Feature extraction
scans the document. The document is sent to a program that
saves it in preferably TIF, JPG or GIF format, so that the image The next phase to segmentation is feature extraction where
of the document can be individual image glyph is considered and extracted for
obtained when needed. This is the first step in OCR. The features. Each character glyph is defined by the following
 (IJCNS) International Journal of Computer and Network Security, 31
Vol. 1, No. 3, December 2009

attributes: (1) Height of the character. (2) Width of the

character. (3) Numbers of horizontal lines present—short
and long. (4) Numbers of vertical lines present—short and
long. (5) Numbers of circles present. (6) Numbers of
horizontally oriented arcs. (7) Numbers of vertically Figure 5. Classification using SVM
oriented arcs. (8) Centroid of the image. (9) Position of the
various features. (10) Pixels in the various regions.
3. Character Recognition Functions Phase II Mapping Solution

The second phase of the Character Recognition functions

consists of classification and Unicode mapping and Margin
recognition strategies.

3.1 Classification Input Space Input Space

The various classification methods are as follows:

3.1.1 A typical rule based Classifier

The height of the character and the width of the character,

various distance metrics are chosen as the candidates for Figure 6. The SVM Classification Algorithm
classification when conflict occurs. Similarly, the
classification rules are written for other characters. This 3.2.1 Classification SVM Type-1
method is a generic one since it extracts the shape of the
characters and need not be trained. When a new glyph is For this type of SVM, training involves the minimization of
given to this classifier block it extracts the features and the error function:
compares the features as per the rules and then recognizes
the character and labels it.
--- (1)
3.1.2 Support Vector Machine based classifier
subject to the constraints:
The architecture chosen for classification is Support Vector
machines, which in turn involves training and testing the
use of Support Vector Machine (SVM) classifiers has gained --- (2)
immense popularity in recent years. SVMs have achieved
excellent recognition results in various pattern recognition Where C is the capacity constant, w is the vector of
applications [1]. Also in off-line character recognition they Coefficients, b a constant and ξi are parameters for
have been shown to be comparable or even superior to the handling nonseparable data (inputs). The index i label the N
standard techniques like Bayesian classifiers or multilayer training cases [13]. Note that y±1 represents the class labels
perceptrons. SVMs are discriminative classifiers based on and xi is the independent variables. The kernel φ is used to
vapnik’s structural risk minimization principle. They can transform data from the input (independent) to the feature
implement flexible decision boundaries in high dimensional space. It should be noted that the larger the C, the more the
feature spaces. The implicit regularization of the classifier’s error is penalized. Thus, C should be chosen with care to
complexity avoids over fitting and mostly this leads to good avoid over fitting.
generalizations. Some more properties are commonly seen
as reasons for the success of SVMs in real-world problems. 3.2.2 Classification SVM Type-2
[11] The optimality of the training result is guaranteed.
In contrast to Classification SVM Type-1, the Classification
3.2 Classification using SVM SVM Type 2 model minimizes the error function:

Support Vector Machine (SVM) is a classifier which

performs classification tasks by constructing hyper planes in
a multidimensional space [12] .It supports both --- (3)
classification and regression tasks. It is classified into two
types namely 1. Classification SVM type-1(also known as C subject to the constraints:
-SVM classification). 2. Classification SVM type-2 (also
known as nu - SVM classification).
--- (4)
32 (IJCNS) International Journal of Computer and Network Security,
Vol. 1, No. 3, December 2009

3.3 Kernel Function 4. Unicode Mapping

s
There are number of kernels that can be used in Support The Unicode standard reflects the basic principle which
Vector Machine models. These include 1. Linear, emphasizes that each character code has a width of 16 bits.
2.Polynomial, 3.Radial basis function (RBF), 4.Sigmoid. Unicode text is simple to parse and process and Unicode
characters have well defined semantics [3] [7]. Hence
Unicode is chosen as the encoding scheme for the current
f (x)=0
work. After classification the characters are recognized and
a mapping table is created in which the unicodes for the
X2 corresponding characters are mapped.

X1

Figure 7. Sample SVM Classification

SVM consists of a learning module (svm_learn) and a

classification module (svm_classify)

Figure 10. Sample Tamil Letters for Unicode Characters

5. Character recognition

The scanned image is passed through various blocks of

functions and finally compared with the recognition details
from the mapping table [6] from which corresponding
unicodes are accessed and printed using standard Unicode
fonts so that the Character Recognition is achieved.

Figure 8. Sample identified character 6. Conclusion

Character Recognition is aimed at recognizing handwritten

Tamil document. The input document is read preprocessed,
feature extracted and recognized and the recognized text is
displayed in a picture box. The Tamil Character
Recognition is implemented using a Java Neural Network. A
complete tool bar is also provided for training, recognizing
and editing options. Tamil is an ancient language.
Maintaining and getting the contents from and to the books
is very difficult. Character Recognition eliminates the
difficulty by making the data available in handwritten
format. In a way Character Recognition provides a paperless
environment. Character Recognition provides knowledge
exchange by easier means. If a knowledge base of rich
Tamil contents is created, it can be accessed by people of
varying categories with ease and comfort.

Figure 9. Sample Character Encoding

 (IJCNS) International Journal of Computer and Network Security, 33
Vol. 1, No. 3, December 2009

Authors Profile
References
Prof. Dr. J. Venkatesh received a MBA
[1] C. Papageorgiou and T. Poggio, “A trainable system
degree in 1997 and a PhD in System
for object detection,” International Journal of Information in 2008, both from the University
Computer Vision, vol. 38, no. 1, pp. 15–33, 2000. of Bharathiar, Tamilnadu, India. He is
[2] G. Guodong, S. Li, and C. Kapluk, “character working as an Assistant Professor at Anna
recognition by support vector machines,” in Proc. University Coimbatore, Tamilnadu, India,
IEEE International Conference on Automatic Face and specialised in the field of Systems and
Gesture Recognition, 2000, pp. 196–201. Production. He published many papers on
[3] K. Jonsson, J. Matas, J. Kittler, and Y. Li, “Learning computer vision applied to automation,
Support Vectors for face verification and motion analysis, image matching, image classification and view-
based object recognition and management oriented empirical and
recognition,” in Proc. IEEE Int Conf on
conceptual papers in leading journals and magazines. His present
AutomaticFace and Gesture Recognition, 2000. research focuses on statistical learning and its application to
[4] B. Heisele, P. Ho, and T. Poggio, “Face recognition computer vision and image understanding and problem
with support vector machines: global versus recognition.
component-based approach,” in ICCV, 2001, pp. 688–
694. Mr. C. Sureshkumar received a ME degree in Computer Science
[5] A. Mohan, C. Papageorgiou, and T. Poggio, in 2006, from the University of Anna, Tamilnadu, India. He is a
“Example-based object detection in images by part-time PhD research scholar in the Department of Computer
Components,” IEEE Trans. Patt. Anal. Mach. Intell., Science and Engineering, Anna University Coimbatore. His main
vol. 23, pp. 349–361, 2001. interests in research are Hand written Tamil Character recognition
[6] B. Scholkopf, P. Simard, A. Smola, and V. Vapnik, refers to the process of conversion of handwritten Tamil character
into printed Tamil character.
“Prior knowledge in support vector kernels,” in
Advances in Neural Inf. Proc. Systems, vol. 10. MIT
Press, 1998, pp. 640–646.
[7] O. Chapelle, P. Haffner, and V. Vapnik, “SVMs for
histogram-based image classification,” IEEE
Transactions on Neural Networks, special issue on
Support Vectors, 1999.
[8] F. Jing, M. Li, H. Zhang, and B. Zhang, “Support
Vector Machines for region-based image retrieval,” in
Proc. IEEE International Conference on Multimedia
and Expo, 2003.
[9] S. Belongie, C. Fowlkes, F. Chung, and J. Malik,
“Spectral partitioning with indefinite kernels using the
nystr¨om extention,” in ECCV, part III, Copenhagen,
Denmark, may 2002,
[10] C. Wallraven, B. Caputo, and A. Graf, “Recognition
with local features: the kernel recipe,” in Proceedings
of the International Conference on Computer Vision,
vol. I, 2003, p. 257ff.
[11] T. Evgeniou, M. Pontil, and T. Poggio,
“Regularization networks and support vector
machines,” Advances in Computational Mathematics,
vol. 13, pp. 1–50, 2000.
[12] V. Vapnik, The nature of statistical learning Theory.
John Wiley and sons, New York, 1995.
[13] D. M. J. Tax and R. Duin, “Uniform object generation
for optimizing one-class classifiers,” Journal of
Machine Learning Research, Special Issue on Kernel
methods, no. 2, pp. 155–173, 2002.