Handwritten Character Recognition: A Comprehensive Review on Geometrical Analysis

IOSR Journal of Computer Engineering (IOSR-JCE)
e-ISSN: 2278-0661,p-ISSN: 2278-8727, Volume 17, Issue 2, Ver. IV (Mar – Apr. 2015), PP 83-88
www.iosrjournals.org
DOI: 10.9790/0661-17248388 www.iosrjournals.org 83 | Page
Handwritten Character Recognition: A Comprehensive Review
on Geometrical Analysis
Meenu Mohan 1
, Jyothi R.L2
1
(Computer Science & Engineering, College of Engineering, Karunagappally/ Cusat, India)
2
(Computer Science & Engineering, College of Engineering, Karunagappally/ Cusat, India)
Abstract: This paper presents a detailed review of Offline Handwritten Character Recognition. HCR is an
optical character recognition, which convert the human readable character to machine readable format. In
HCR, to attain 99% accuracy is very difficult. Here a detailed study on Geometrical methods of feature
extraction in character recognition has been done by giving more emphasis to Zone based techniques and it has
been analyzed that the efficiency of HCR depends on the selection of appropriate feature extraction methods
and classifier. A comparative study in various steps in character recognition like Preprocessing, Segmentation,
Feature Extraction and Classification are carried out. Various application areas of HCR like Postal address
reading, mail sorting, office automation for text entry, person identification, signature verification, bank-check
processing etc. are also analyzed.
Keywords: OCR, Preprocessing, Segmentation, Feature Extraction, Classification.
I. Introduction
Character Recognition is an active research area in the field of image processing and pattern
recognition. It is the process of converting an image representation of document into digital format. Character
recognition is of 2 types: Magnetic character recognition and Optical character recognition. Optical character
recognition (OCR) is the translation of scanned images of handwritten, typewritten or printed document into
machine encoded form. The document image may be printed or handwritten. The printed document means that
the documents are written by electronic devices, which includes all the printed materials such as book,
newspaper, magazine etc. Handwritten documents are written by hand held equipments. The handwritten
recognition system can be classified into online and offline hand written recognition system as shown in Fig 1.
Fig 1: Classification of Character Recognition
In online handwritten character recognition (HCR), a special electronic pen samples the handwriting
input where the writing is done on electronic surface. Here recognition is done in real time. Here features that
are extracted depend on the dynamic information that has been used as input. In offline handwritten character
the information that serves as input does not exhibit any dynamic change but the most important challenge of
handwritten character recognition is the variability of writing style. Different person have their own
handwriting. So the handwritten text varies from person to person. This paper discusses various methodologies

Handwritten Character Recognition: A Comprehensive Review on Geometrical Analysis
that have been analyzed based on literature study of handwritten character recognition systems. The different
stages of Handwritten character recognition system are Pre-processing, Segmentation, Feature Extraction and
Classification.
II. Phases Of Character Recognition System
A. Image Acquisition:-
The offline recognition system acquires an optically scanned image as an input image. Digitization in
handwritten character recognition is the process of converting a handwritten document into a digital format. A
scanner or digital camera captures an image of text and converts it to an image files format such as a bitmap,
jpeg etc.
B. Preprocessing:-
Preprocessing is a series of operations that is performed on the scanned input image to improve the
quality of image for effective feature extraction. Major steps under pre-processing are:
1. Noise Removal
2. Binarization
3. Morphological Operations
4. Size Normalization
Noise is introduced in an image during image acquisition. It produces a random variation of image
intensity and sometimes will be visible as grains in the image. Noise removal is the process of removing or
reducing the noise from the image. There exist several algorithms and filters for noise reduction and removal.
The different types of noises that exist in document images are Salt and Pepper noise, Gaussian noise, Gamma
noise, Uniform noise etc. Various type of filtering methods like Gaussian filtering method, Min-max filtering
method etc. are applied for noise removal. Median filter is used to remove salt and pepper noise. Binarization is
the process of converting colour or gray-scale image into binary image with the help of thresholding. The
different methods of binarization are Global thresholding, Local thresholding, Adaptive thresholding, Otsu’s
method etc. Morphological operations are also used in preprocessing. Dilation and Erosion are commonly used
morphological operation that increase or decrease character size of an image. Dilation is the process of adding
pixels to the character boundary. In erosion, the pixels are removed from the boundary of character.
Skeletonization is the process of reducing the character image to single pixel wide representation.
Fig 2: Offline Handwritten Character Recognition Architecture
Normalization is the process that reduces the range of pixels intensity values present in an image. Size
normalization is the preprocessing step that resizes the character image into a standard size. Skew detection and
correction is also a part of character image preprocessing. During document scanning, skew is introduced in the
image. Skew angle is an angle that the text lines of the image make with horizontal direction. The aim of skew
detection is to align an image text before processing. Commonly used skew elimination techniques are
projection profile method and Hough transform method.

C. Segmentation:-
Segmentation is the process that isolates individual character from handwritten character image.
Segmentation is classified into Implicit and Explicit segmentation. In implicit segmentation, the words are
predicted directly without segmenting the word as individual letters but the explicit segmentation, the word is
segmented into individual character. Segmentation is carried out using threshold based, edge based, region
based, clustering techniques etc. The different steps in segmentation are line, word and character segmentation.
In line segmentation, horizontal projection profile method is used. It separates the boundaries between lines.
Word segmentation is done by applying vertical projection profile method on the separated lines. Finally, the
characters are isolated from these words using connected component labeling.
D. Feature Extraction:-
The feature extraction method is the most vital and conclusive one and therefore the features should be
extracted correctly, that decides the effectiveness of the classification. Feature extraction methods are classified
into three major groups:
1. Statistical features.
2. Global transformation and Series expansion
3. Structural features.
Statistical features represent the character image as statistical distribution of points. Zoning, Crossing
and Distances, Projections etc. are the various methods used for statistical feature extraction. Global
transformation and series expansion includes various techniques like Fourier transform, Gabor transforms,
Wavelets, Moments and Karhunen-Loeve Expansion etc. Structural features are based on geometrical and
topological properties of the character. Loops, curves, lines, T-point, cross, aspect ratio, strokes and their
directions, inflection between two points etc. are used as structural features.
E. Classification:-
Classification is the decision making part of the any recognition system. Various approaches for
classification in character recognition systems are analyzed. Most commonly seen classifiers are Artificial
Neural Network, SVM, and Nearest Neighbor classifier. The classifiers compare the given vector with the
stored pattern and give the best match as an output. The various pattern classification methods can be
successfully applied to character recognition. The classification methods that are used in handwritten character
recognition systems are categorized into statistical methods, ANN, SVM, structural methods and multiple
classifier methods. In case of Statistical methods, ANN and SVM the input feature vectors should be of same
dimensionality for a single recognition system. In multiple classifier methods, the classification results of
multiple classifiers are combined to reorder the classes.
III. Literature Review
There are many researches that have been done in the field of image processing and pattern recognition
which is related to handwritten character recognition. This section describes an extensive review for handwritten
character recognition:
J.Pradeep et.al. [1] focus on recognition of English offline handwritten character using Neural
Network. Noise Removal is done using median filter, Binarization is using Otsu’s global technique, Detection of
edges are done using Sobel filter, dilation and filling are also carried out as the part of preprocessing. Diagonal
feature extraction method is used for feature extraction stage. Divide the enhanced image into 54 equal zones.
So 54 features are obtained from each character. In classification, feed forward back propagation neural network
is used. The diagonal features provide good recognition accuracy compared to the conventional horizontal and
vertical methods of feature extraction. Using this 54 feature based system it yield a recognition efficiency of
98%.
S.V. Rajashekararadhya et. al. [2] proposed the Image centroid and Zone centroid based distance
metric feature extraction system handwritten numeral recognition for four popular South Indian Scripts. The
four languages are Malayalam, Tamil, Kannada and Telugu. Preprocessing stage concentrated on Noise
reduction, Slant correction, Normalization and Thinning. In feature extraction stage, image centroid and zone
centroid and hybrid methods are used. Feed forward back propagation neural network and nearest neighbor
classifiers are used for subsequent classification and recognition purpose. The recognition accuracy obtained for
Kannada and Telugu numerals is 99%. For Tamil and Malayalam numerals have obtained 96% and 95%
respectively.
S.L.Mhetre et.al. [3] have proposed two different approaches for recognition of Devanagari
handwritten numerals. In the first method, Grid features are used. In the second method, ICZ (Image Centroid
Zone) & ZCZ (Zone Centroid Zone) features based on distance information are extracted. Here ANN and

matching score are used for classification and the accuracies obtained using two approaches are evaluated. In
classification ANN provide better accuracy compared to matching score.
S. V. Rajashekararadhya et. al. [4] has described Zone based feature extraction system (ICZVDD-
ICZHRD method for handwritten numeral/mixed numerals recognition of South-Indian scripts. The nearest
neighbor, feed forward back propagation neural network and support vector machine classifiers are used for
classification. The recognition rate obtained of 98.65 % for Kannada numerals, 96.1 % for Tamil numerals, 98.6
% for Telugu numerals and 96.5% for Malayalam numerals using the Support vector machine.
S. V. Rajashekararadhya et. al. [5] have described Image Centroid Zone (ICZ) based Angle feature
extraction for handwritten numeral recognition of Kannada script. The numerals image centroid is computed and
the image is further divided into n equal zones. Average angle from the character centroid to the pixels present
in the zone is computed. This procedure is repeated sequentially for all zones present in the numeral image.
Finally n such features are extracted. For classification purpose, nearest neighbor classifier and support vector
machines are used. The recognition accuracy achieved 96.05% for Kannada numerals using Support vector
machines.
Gita Sinha et. al. [6] had taken care of Arabic numeral recognition. In preprocessing stage, binarization,
dilation, erosion, noise removal and normalization are used. Three features extraction techniques that has been
used are Image Centroid Zone (ICZ), Zone Centroid Zone (ZCZ) and Hybrid feature extraction techniques.
Hybrid feature extraction techniques are combination of ICZ+ZCZ. SVM classifier is used for classification.
The recognition rate is 97.21% on handwritten Arabic numeral.
Seema A. Dongare et.al. [7] have proposed Devanagari character recognition works in stages as
document preprocessing, segmentation using line segmentation, word and character segmentation, feature
extraction using zone based approach followed by recognition using feed forward neural network. Recognition
of handwritten Devanagari character is quite difficult due to presence of shirorekha, conjunct characters and
similarity in shapes for multiple characters. Here an attempt is carried out to increase the accuracy and
performance.
Gita Sinha et.al. [8] presented Gurumukhi handwritten character recognition. Preprocessing stage
includes steps like Gray scale conversion, Binarization using Otsu’s method, filtering and morphological
operation, Noise removal, Skeletonization, Skew detection. Zone-based feature extraction technique is used for
extracting the feature and SVM classifier is used for Gurumukhi handwritten character recognition. The
recognition accuracy obtained is 95.11%.
Sandeep Saha et.al.[9] proposed 40-point feature extraction for English handwritten character
recognition using multilayer feed forward neural network. The whole image is divided into 16 zones and then
computed the average intensities of each zones. Then the entire image is divided diagonally from left top to
bottom, right top to bottom, left bottom to top and right bottom to top and innermost cell features are extracted.
Finally features vectors consisting of 40 features are tested using the artificial neural network and has a better
recognition efficiency is reported.
Sangeetha Sasidharan et. al. [10] describes that segmentation of offline Malayalam handwritten
character recognition. Preprocessing stage includes noise removal and binarization. In segmentation stage, line
segmentation using horizontal projection profile method is used. Character segmentation focus on the
segmentation of untouched characters, segmentation of consonants touching to Valli (special Malayalam
character) and segmentation of consonants touching to Chandrakala(special Malayalam character). The
efficiency obtained in this work is 94.08%.
Anita Pal et. al. 11] have proposed boundary tracing along with Fourier Descriptor for handwritten
English character recognition. In preprocessing stage skeletonization and normalization is performed. In feature
extraction stage, boundary detection is done using 8-neighbor adjacent method. Neural Network is used for
classification.
Reetika Verma et. al. [12] describes the surf feature extraction and neural network .This paper
demonstrated capability for solving complex problems of character recognition. Preprocessing stage includes
noise removal and image enhancement. Surf feature technique and neural network is used for feature extraction
and classification. This technique is fast, low cost and more accurate result can be obtained.
In Abdul Rahiman M et. al.[13] proposed a handwritten character recognition system based on vertical
and horizontal line positional analyzer algorithm. In preprocessing, median filter is used for noise removal. In
segmentation stage, line and character separation are used, which gives isolated character. The features are
extracted based on horizontal and vertical line count and position. Decision tree classifier is used for
classification. Recognition accuracy obtained 91%.
In Pranchi Mukherji et. al. [14] proposed a Shape feature extraction techniques for handwritten
character recognition. Preprocessing includes noise removal using Gaussian filter, binarization using Ostu’s
method, skeletoniztion. Average Compressed Direction Coding Algorithm for stroke is used for feature
extraction method. In classification, Decision Tree classifier is used. 86.4% is the overall recognition accuracy.

In Parikh Nirav Tushar et. al.[15] describes that a Chain Code based handwritten character recognition.
In preprocessing stage, binarization and slant correction are used. In feature extraction stage Chain Codes are
constructed which form the features of character. ANN is used for classification. Recognition accuracy is 80% is
reported. Amritha Sampath et.al [16] proposed the same method of feature extraction.
Comparison between the various literatures that is mentioned in the section is summarized in the following
table1.
Table 1: Comparison of Various Geometrical Techniques in HCR
Author Preprocessing Segmentation Feature Extraction Classification Recognition
Accuracy
J.Pradeep[1]
Noise Removal,
Binarization,
Edge detection,
Dilation and filling. -
Diagonal feature
Feed Forward Back
propagation Neural
Network.
98%
S.V.Rajashekararadhya
[2]
Noise Removal,
Slant correction,
Normalization,
Thinning.
-
Image centroid zone
Zone centroid zone
Hybrid centroid zone
Feed Forward Back
propagation Neural
Network and
Nearest Neighbor
99% for
Kannada
96% for Tamil
95% for
Malayalam
S.L.Mhetre [3]
Colour to gray
conversion, Noise
Removal,
Thresholding,
Thinning,
Size
Normalization.
-
Grid based method;
ICZ &ZCZ method
ANN & Matching
Score
-
[4]
Noise Removal,
Slant correction,
Normalization,
Thinning
-
Image centroid zone
Zone centroid zone
Hybrid centroid zone SVM
98.65% for
Kannada
98.6% for
Telugu
96.1% for
Tamil
96.5% for
Malayalam.
[5]
Noise Removal,
Slant correction,
Normalization,
Thinning.
- Zone based angle SVM 96.05%
Gita Sinha[6]
Binarization,
Dilation,
Erosion,
Noise removal,
Normalization.
-
Image centroid zone
Zone centroid zone
Hybrid centroid zone.
SVM 97.21%
Seema A Dongare[7]
Colour to gray
conversion,
Noise removal,
Binarization.
Line
Word
Character
Image centroid zone
Zone centroid zone
ANN
-
Gita Sinha [8]
Colour to gray
conversion,
Noise Removal,
Binarization, Filtering
operation,
Contour smoothing,
Skew detection,
Skeletonization.
Line
Word
Character
Image centroid zone
Zone centroid zone
SVM 95.11%
Sandeep Saha[9]
Colour to gray
conversion,
Binarization.
Image cropping 40-point feature ANN -
Sangeetha
Sasidharan[10]
Noise Removal,
Binarization
Line segmentation
using projection
profile; Character
segmentation of
untouched
characters,
consonants
touching to Valli
and Chandrakala.
- - 94.08%
Anita Pal[11]
Skeletonization,
Normalization. -
Fourier descriptors ( 8-
Neighbour Adjacent
Method)
Multilayer Perceptron
Network 94%
Reetika Verma [12] Noise Removal - Surf feature extraction Back Propagation
Neural Network
-

Abdul Rahiman M[13] Noise Removal Line & Character
separation
Horizontal and Vertical
Line count and position
Decision Tree 91%
Pranchi Mukherji [14]
Noise Removal,
Binarization,
Skeletonization.
-
Average Compressed
Direction Coding
Algorithm
Decision Tree 86.4%
Parikh Nirav Tushar
[15]
Binarization
Slant correction
- Chain code ANN 80%
IV. Conclusion
This paper presented a detailed study of offline handwritten character recognition systems developed in
different languages. From the literature review it has been analyzed that the recognition accuracy mainly
depends on proper selection of feature extraction methods. This work mainly concentrated on geometrical based
character analysis methods. The recognition efficiency of Feature vectors are can be improved by selection of
appropriate preprocessing methods. It has been also analyzed that in case of character recognition neural
network and SVM provide better classification efficiency compared to other classification methods.
References
[1]. J.Pradeep, E.Srinivasan and S.Himavathi, Diagonal Feature Extraction Based Handwritten Character System Using Neural
Network, International Journal of Computer Applications (0975-8887), vol.8, no.9, pp.17-22, October 2010.
[2]. S.V.Rajashekararadhya and P.Vanaja Ranjan, Handwritten Numeral/Mixed Numerals Recognition of South-Indian Scripts: The
Zone Based Feature Extraction Method, Journal of Theoretical and Applied Information Technology, vol.7, no.1, pp. 063-079, 2005
- 2009.
[3]. S.L.Mhetre and M.M.Patil, A Comparative Study of Two Methods for Handwitten Devanagari Numeral Recognition, IOSR
Journal of Computer Engineering, vol.15, pp. 49-53, Nov-Dec.2013.
[4]. S.V. Rajashekararadhya and P. Vanaja Ranjan, Efficient Zone Based Feature Extraction Algorithm for Handwritten Numeral
Recognition of Four Popular South Indian Scripts, Journal of Theoretical and Applied Information Technology, pp. 1171-1181,
2005 - 2008.
[5]. S.V. Rajashekararadhya and P. Vanaja Ranjan, Handwritten numeral recognition of Kannada script, Proceedings of the
International Workshop on Machine Intelligence Research, pp. 80-86, 2009.
[6]. Gita Sinha and Jitendra kumar, Arabic Numeral Recognition Using SVM Classifier, International Journal of Emerging Research in
Management &Technology, vol.2, pp.62-67, May 2013.
[7]. Seema A.Dongare, Dhananjay B.Kshirsagar and Snehal V. Waghchaure, Handwritten Devanagari Character Recognition using
Neural Network, IOSR Journal of Computer Engineering (IOSR-JCE), vol.16, pp. 74-79, Mar-Apr. 2014.
[8]. Gita Sinha, Anita Rani, Renu Dhir and Rajneesh Rani, Zone-Based Feature Extraction Techniques and SVM for Handwritten
Gurmukhi Character Recognition, International Journal of Advanced Research in Computer Science and Software Engineering,
vol. 6, pp. 106-111, June 2012.
[9]. Sandeep Saha, Nabarag Paul, Sayam Kumar Das and Sandip Kundu, Optical Character Recognition using 40-point Feature
Extraction and Artificial Neural Network, International Journal of Advanced Research in Computer Science and Software
Engineering, vol.3, pp. 495-502, Apr. 2013.
[10]. Sangeetha Sasidharan, Anjitha Mary Paul, Segmentation of Offline Malayalam Handwritten Character Recognition, International
Jouurnal of Advanced Research in Computer Science and Software Engineering, vol.3, pp. 761-766, Nov. 2013.
[11]. Anita Pal & Dayashankar Singh, Handwritten English Character Recognition Using Neural Network, International Journal of
Computer Science & Communication, vol.1, pp. 141-144, Jul.-Dec. 2010.
[12]. Reetika Verma, Rupinder Kaur, An Efficient Technique for Character Recognition using Neural Network & Surf Feature
Extraction, International Journal of Computer Science & Information Technologies, vol.5, pp. 1995-1997, 2014.
[13]. Abdul Rahiman M and M.S. Rajasree, Recognition of Handwritten Malayalam Character using Vertical & Horizontal Line
Positional Analyzer Algorithm, in Proc. ICMLC, 2013.
[14]. Prachi Mukherji and Priti.P.Rege, Shape Feature and Fuzzy Logic Based Offline Devanagari Handwritten Optical Character
Recognition, Proc. Journal of Pattern Recogntiion Research 4, Jun.2009.
[15]. Parikh Nirav Tushar and Saurabh Upadhyay, Chain Code Based Handwritten Cursive Character Recogntiion System with Better
Segmentation Using Neural Network , in Proc. International Journal of Computational Engineering Research, vol.3, May.2013.
[16]. Amritha Sampath, Tripti.C and Govindaru.V, Freeman Code Based Online Handwritten Techiniques for Handwritten Character
Recognition for Malayalam using Backpropagation Neural Networks, ACIJ, vol.3, no.4, Jul.2012.

Handwritten Character Recognition: A Comprehensive Review on Geometrical Analysis

More Related Content

What's hot (20)

Similar to Handwritten Character Recognition: A Comprehensive Review on Geometrical Analysis (20)

More from iosrjce (20)

Recently uploaded (20)

Handwritten Character Recognition: A Comprehensive Review on Geometrical Analysis