A Simple and Effective method for recognition of
Kannada characters based on Freeman Chain Code
ABSTRACT
With the recent emergence and widespread application of multimedia
technologies, there is increasing demand to create a paperless environment
in our daily life. In transformation from the traditional paper-based society
to a truly paperless electronic information society, document image
processing in general and optical character recognition (OCR) in particular
will play an important role. Machine reading of optically scanned text is
usually called optical Character recognition [S Basavraj Patil et. al, 2002].
It is observed from the literature survey that much work has been
done on recognition of characters of many languages like English, Chinese,
Japanese, and Korean. The survey also reports on the research work on
recognition of many Indian languages like Tamil, Bangla, Telugu, and
Malayalam. It is also observed that the work on recognition of Kannada
characters is still an open research problem. Efforts are being made to
develop OCR for Kannada language.
In this paper, an attempt has been made to develop a simple and
effective method for the recognition of Kannada characters. The present
method is based on template matching to recognize basic printed Kannada
characters and uses freeman chain code based on contour tracing method.
The contour of character gives both inner & outer border of a character using
connected component analysis. The proposed method obtains the contour of
the character and then generates a freeman chain code for contoured
character. The obtained chain code is unique for printed character.
Experimental results show the relatively high accuracy of the method when
it is tested on all size characters.
Keywords: OCR, Freeman Chain Code (FCC) , Contour Tracing, Connected Components.
1. INTRODUCTION
OCR is sub field of pattern recognition which recognizes the printed or
handwritten character in the documents. Due to impact and developments in
the information technology, nowadays more emphasis is given in Karnataka
to use Kannada at all levels and hence use of Kannada in computer is also
necessary[10]
The Kannada script
The script has forty-nine characters in its alpha-syllabary and is phonemic.
The Kannada character set is almost identical to that of other Indian
languages. The number of written symbols, however, is far more than the 49
characters in the alpha-syllabary, because different characters can be
combined to form compound characters (ottaksharas). Each written symbol
in the Kannada script corresponds with one syllable, as opposed to one
phoneme in languages like English. The Kannada writing system is an
abugida, with consonants appearing with an inherent vowel.
The characters are classified into three categories: swaras (vowels),
vyanjanas (consonants) and yogavaahas (part vowel, part consonants).
The name given for a pure, true letter is akshara, akshara or Varna. Each
letter has its own form (ākāra) and sound (shabda); providing the visible
and audible representations, respectively. Kannada is written from left to
right. Kannada alphabet (aksharamale or varnamale) now consists of 49
letters.
Figure 1. Vowels in Kannada.
Figure 2. Consonants in Kannada.
2.LITERATURE SURVEY
Many researchers have been working on English, Arabic and Farsi character
recognition. The following are some of the works that were reported
A.Montazer, N.Jafari have proposed a method of identifying the Persian
character. The features that are extracted are used to identify multi font
numerals. The features extracted from the size normalized and thinned
binary array, form the input for recognition process. The identification
system is designed so as to optimize the structural parameters in the
justification function, in order to enhance the recognition rate.[18]
M S Khorsheed has proposed a system to recognize cursive Arabic
typewritten text. The system is built using the Hidden Markov Toolkit which
is a portable toolkit for speech recognition system. The proposed system
decomposes the page into text lines and then extracts a set of simple
statistical features from overlapped windows running through each text line.
The feature vector sequence is injected to the global model for training and
recognition purposes. A data corpus which includes Arabic text from two
computer–generated fonts is used to assess the performance of the
proposed system.[19]
Some of the works on Indian Languages that have been reported are as
follows:
Negi Atul, Chakravarthy Bhagavathi, Krishna B have proposed an OCR
system for telgu(A south Indian Language) characters based on Fringe
distance measurement. Template of all the characters to be recognized by
the system is formed. A type of similarity measure
is performed between
the test character and the templates. The template yielding the highest
similarity score is declared as the class of the character.[15]
B.M.SAGAR, Dr SHOBHA G, Dr RAMAKANTH KUMAR Phave proposed a
method which uses character ASCII value, character name, character BMP
image, character width, character length and total number of ON pixel in the
image as the feature vector. [6]
R SANJEEV KUNTE and R D SUDHAKER SAMUEL have proposed a system to
Identify basic symbols in printed Kannada text. HU’s invariant moments and
Zernike moments have been used to extract the features of printed Kannada
characters. [10]
ASHWIN and SASTRY have proposed an OCR system in which the character
image is split into three basic zones. Each zone is divided into a number of
circular tracks and sectors and feature like the number of ON pixels in each
annular region is used for classification using Support Vector Machine(SVM)
[1]
3. PREPROCESSING:
Fig3
The input to the image looks like the one above. This character is
preprocessed to remove any disturbances .Once the input has been given
the image will be segmented into characters. This phase has three steps
a. Line segmentation
b. Word segmentation
c. Character segmentation
a. Line Segmentation:
The process of identifying lines in the obtained image is called line
segmentation. Following are the steps for line segmentation.
1. Scan the obtained image horizontally to find the first ON pixel and store y
coordinate as y1.
2. continue scanning the image as long as you find a ON pixel.
3. When you encounter OFF pixel remember the coordinate y as y2.
4. y1 to y2 is the required line.
5. Repeat the above steps till the end of the image.
b. Word Segmentation:
Since there is a distance between a word and another word we use word
segmentation.
1. Scan the obtained image vertically to find the first ON pixel and store x
coordinate as x1.
2. Continue scanning the image as long as you find a ON pixel.
3. The distance between the characters is less than the distance between
two words. This information is used to segment the line into words.
4. The above step is repeated until all the line segments are segmented into
words.
c. Character Segmentation:
1. Scan the obtained image vertically for the identified word to find the first
ON pixel and store x coordinate as x1.
2. continue scanning the image as long as you find a ON pixel.
3. When you encounter OFF pixel store the coordinate x as x2.
4. x1 to x2 is the required character.
5. Repeat the above steps for all recognized words.
Contour Tracing/Analysis
Contour tracing is also known as border following or boundary following.
Contour tracing is a technique that is applied to digital images in order to
extract the boundary of an object or a pattern such as a character. The
boundary of a given pattern P is
Fig4(a) 4-neighbors (4-connected); 4(b) 8-neighbors (8-connected).
For the obtained characters we use contour tracing.
4.FEATURE EXTRACTION:
Once the contour of the image is obatined we apply freeman chain-code.
Fig 5 :Eight chain-code directions.
The coordinates of the boundary pixels are obtained first , based on these
coordinates the chain code of the character image is found.
The normalized chain code is obtained by transforming it to a two
dimensional matrix. The first row of this matrix contains the value of the
chain code, and the second row contains the frequency of occurrence of that
value.[3] . For example, if the chain code of a given character is:
88883331111225833312 then it can be converted into the following form of
a
2 × 9 matrix:
831258312
434211311
we remove all values whose frequencies are
1. For instance, in the above
example, the chain code will be reduced to:
83123
43423
8312
4642
The process of removing the less-frequent digits can be continued. For
instance in our test, the frequencies less than or equal to five were deleted.
Again in the resulted chain code the frequency of each remained digit is
summed. Then to transform the chain code matrix to a normalized chain
code with length of 10, the relative frequency of each digit is computed
using:
Where
is the normalized frequency and
is
the frequency of each digit in the chain code respectively. In the above
example we will obtain:
8
3
1
2
2.22 3.33 2.22 1.11
Then the normalized frequency would be rounded of to nearest decimal
which in turn would be concatenated to generate the length 10 chain code:
8833311222
5. Proposed Method
1.The input image consists of the characters ,this image after preprocessing
line,word,character segmentation processes are done.
2. Once the characters are segmented, the contour of the image is obtained.
After obtaining the contour, the coordinates of the Boundary pixels are
obtained which is sent to obtain the Freeman chain code.
3. The obtained chain code is then normalized.
4. This normalized chain code is compared with the existing database. If
match occurs then respective character is identified. Else error message is
generated.
6. EXPERIMENTAL RESULTS
Matlab was used to implement the proposed system and database of all
characters along with normalized chain code is used. After getting the
normalized chain code, the codes were compared with existing database to
recognize the character which gives the 98% accuracy.
Table I : RESULTS OBTAINED FROM IMPLEMENTATION OF THE EMPLOYED METHOD
Letter
name
Contour
image
Image Normalized
chain code
AA
1005443622
BA
2213400766
JA
2134460078
MA
0013276654
NA
0000327644
OO
1007653222
TA
0404672426
7. CONCLUSION
A template matching chain code-based approach for identification of
Kannada characters was introduced in this paper. After obtaining the contour
of the characters, features are extracted using Freeman chain code.
Obtained chain code is compared with the database to identify the character.
The proposed method is tested on all simple characters of Kannada text.
Experimental results using standard fonts show that the accuracy of the
proposed method is 98%. In the future work, we will try to complete our
work for all compound characters like
ottaksharas(mixed character) and
handwritten Kannada character .
8.References:
1.Ashwin TV, Sastry P S 2002Afont and size-independent OCR system for
printed Kannada documents using support vector machines. Sadhana
27:
35–58
2.Chong Chee-Way, Raveendran P, Mukundan R 2003 A comparative
analysis of algorithms for fast computation of Zernike moments. Pattern Rec.
36: 731–742
3.Multi-Font Farsi/Arabic Isolated Character Recognition Using Chain Codes
H. Izakian, S. A. Monadjemi, B. Tork Ladani, and K. Zamanifar
4.Girosi F, Poggio T 1990 Networks and the best approximation property.
Bio. Cybernetics. 63: 169–176
5.Gonzalez R C, Woods R E 1993 Digital image processing (Boston, MA,
USA: Addison Wesley
Longman Publishing Co. Inc.)
6.B M Sagar , Dr Shobha , Dr Ramakanth Kumar P, converting printed
Kannada text image file to machine editable format using database
approach,International Journal Of Computers, Issue 2 , Volume 2,2008
7.HuM-K 1962 Visual pattern recognition by moment invariants. IRE Trans.
Inf. Theory. IT-8: 179–187
8.Jawahar C V, Pavan Kumar, Ravi Kiran S S 2003 A Bilingual OCR for HindiTelugu documents and its applications. Proc. Seventh Int. Confer . on
Document Anal. and Rec. 408–412
9.Khotanzad A 1998 Rotation invariant pattern recognition using Zernike
moments. Proc. Int. Conf. on Pattern Rec. 326–328
10.Kunte Sanjeev R, Sudhaker Samuel R D 2006 A two-stage character
segmentation scheme for Printed Kannada text. J. Graphics, Vision and
Image Processing 6: 1–8
11.Moody J, Darken C J 1989 Fast learning in network of locally-tuned
processing units. J. Neural
Comput. 1: 281–294
12.Mukundan R, Ong S H, Lee P A 2001 Image analysis by Tchebichef
moments. IEEE Trans. Image Processing 10: 1357–1364
13.Mohammed Al-Rawi, Yang Jie 2002 Practical fast computation of Zernike
moments. J. Comput. Sci. and Technol. 17: 181–188
14.Nagabhushan P, Pai RadhikaM1999Modi?ed region decomposition method
and optimal depth decision tree in the recognition of non-uniform sized
characters—An experimentation with Kannada characters. Pattern Rec. Lett.
20: 1467–1475
15.Negi Atul, Chakravarthy Bhagavathi, Krishna B 2001 An OCR system for
Telugu. Proc. Sixth Inter .Confer . on Document Anal. and Rec. 1110–1114
16.VijayaKumar B, Ramakrishnan A G 2004 Radial basis function and subspace approach for printed Kannada text recognition. Proc. IEEE ICASSP
2004 5: 321–324
17.S.Basavaraj Patil &N V Subbareddy, Neural Network based system for
script identification in Indian documents,Sadhana vol.27,part 1,February
2002,PP83-97
18.
A.Montazer,N.Jafari
and
H.Ebrahimnezhad
Recognition
of
Persian
Numeral Fonts by combining the Entropy Minimised Fuzzifer and Fuzzy
Grammar.6th
WSEAS
Int.Conf.
on
Artificial
Intelligence,
Knowledge
Engineering and Data Bases, Corfu Island, Greece
19.M S Khorsheed Recognising Cursive Arabic Text using a Speech
Recogniton System,7th WSEAS International Conference on Automation
&Information,Cavtat,Croatia