IRJET- Review on Optical Character Recognition

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 06 | June 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 3666
REVIEW ON OPTICAL CHARACTER RECOGNITION
Muna Ahmed Awel1, Ali Imam Abidi2
1Computer Science and Engineering, Sharda University, Greater Noida, India
2Assistant Professor, Computer Science and Engineering, Sharda University, Greater Noida, India
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract - Optical Character Recognition is the area of
Pattern Recognition that has a topic of studies over the past
some decades. Optical character recognition is technique of
automatically identifying of different character from a record
picture additionally provide full alphanumeric recognition of
printed or handwritten characters, textnumerical, letters, and
symbols in to a computer process able layout including ASCII,
Unicode and so forth. Optical character recognition is the
bottom for many distinct styles of programs in diverse fields, a
lot of which we use in our daily lives. Cost effective and less
time consuming, corporations, submit offices, banks, security
systems, and even the field of robotics hire this system as the
base in their Operations. These days, there are numerous
portions of research and making use of OCR technology. These
OCR technologies help to examine unique documents written
in English, Chinese, Hindu, Arabic, Russian, and others
languages. On This paper present review of some researches
has been made in English, Arabic and Devanagari characters.
And explained the methodology they use and challenge they
face during development of Optical character recognition.
Key Words: OCR, optical character recognition, character
recognition, handwriting character recognition.
1. INTRODUCTION
Character recognition, usually abbreviated to optical
character recognition or shortenedOCR,isthemechanical or
electronic translation of images of handwritten, typewritten
or printed text (usuallycaptured bya scanner)intomachine-
editable text [4]. It is a field of research in pattern
recognition, artificial intelligence and machine vision.
Though academic research in the field continues, the focus
on character recognition has shifted to implementation of
proven techniques. Optical characterrecognitiontechnology
was invented in the early 1800s, when it was patented as
reading aids for the blind. In 1870, C. R. Carey patented an
image transmission system using photocells, and in 1890
P.G. Nipkow inventedsequential scanningOCR.However, the
practical OCR technology used for reading characters was
introduced in the early 1950s as a replacement for
keypunching system [2]. A year later, D.H. Shephard
developed the first commercial OCR for typewritten data.
The 1980’s saw the emergence of OCR systems intended for
use with personal Computers. Nowadays, it is common to
find PC-based OCR systems that are commercially available.
However, most of these systems are developed to work with
Latin-based scripts. Optical character recognition systems
for Latin characters have been available for over a decade
and perform well on clear typed text. There are research has
also been directed at other non-Latin scripts such as Arabic,
Japanese, Chinese, Hindu, Tibetan. In order to develop an
OCR system it requires the development and integration of
many sub systems. The first step is preprocessing such as
skew detection and correction, noise detectionand removal,
binearization, thinning, and normalization. Then
segmentation of document images into line, word and
characters. This is followed by feature extraction for
representing character images and a classification module
that label characters to their proper class. Finally, post
processing i.e. applying
2. LITERATURE REVIEW
Character recognition technique has been completed
through studies on different charactersfor example,English,
Arabic, Chinese, Devanagari, Bangla, Farsi and Kannada and
so on. Totally, the complete method is carried out in three
phase Preprocessing, Feature extractionandrecognition[5].
In this paper only cover the study has been done on English,
Arabic and Devanagari scripture.
2.1 In English Scripter Character Recognition
In 2004 N. M. Noor, M. Razaz and P. Manley-CookeProposed
system using global geometrics feature extraction and
geometric density classifier for feature extraction then
neural fuzzy logic used for classification. Evaluation of the
system has achieve for Geometric Density 77.89% and
Geometric Feature 76.44% accuracy rate [6]. In 2010 Dewi
Nasien, Habibollah Haron and Siti Sophiayati Yuhaniz This
studies Take three datasets from NIST database considered
lowercase letters 189,411, uppercase letters 217,812 and
combination of uppercase and Lowercase letters 407,223
sample are used. Those Samples are divided into 80% for
training and 20% for testing. For feature extraction used
Freeman chain code (FCC). Support vector system (SVM) is
selected for recognition step. The method recognize for the
first dataset 86% accuracy, second dataset 88% ofaccuracy
and third dataset 73% accuracy achieved [7]. In 2011 Vijay
Patil and Sanjay Shimpi develop system that recognize
handwritten English character using neural network for
feature extraction system they used Character Matrix And
for recognition back propagation neural network used. The
result indicate that hand back propagation network provide
more than 70% of accuracy rate[3]. 2015 M. S. Sonawane
and Dr. C.A. Dhawale this study compare and evaluate two
classifier which is artificial neural network and nearest
neighbor. Used grid method to extract feature and the result

show nearest neighbor achieve 61.53% accuracy when
neural network gives 57.69%. Math lab tool was used for
features extracted and recognition. The evaluation outcome
suggests Nearest Neighbor is a better recognizer comparing
with artificial neural network when implemented to English
Characters[8].
2.2. Arabic Scripter Character Recognition
In 2002 Majid M. Altuwaijri and Magdy A. Bayoumi They
develop system to recognize Arabic text using neural
network used set of moment invariants descriptors (under
shift, scaling and rotation) and artificial neural network
(ANN) used for classification The study has shown 90% of a
high accuracy rate [9]. In 2015 Ashraf Abdel Raouf, Colin A.
Higgins, Tony Pridmoreand Mah-moudI.Khalil Haarstudied
approach for recognizing Arabic character using Haar
Cascade Classiﬁer (HCC) These classifiers were trained and
tested on some 2,000 images. To extract feature Haar-like
feature extraction used and boosting of a classifier cascade.
The system was tested with real text image and produces
87% accuracy rate for Arabic character recognition[10]. In
2017 N. Lamghari, · M. E. H. Charaf and · S. Raghay On this
research the data are divided into three parts. From 34,000
character 70% are used for training, 15% for testing phase
and 15% for validation. To extract feature hybrid feature
extraction used (pixel density, resize, freeman code,
structural features, invariant) for recognition used feed
forward-back propagation neural network. The system has
achieved 98.27% high recognition rate[11]. In 2018 Noor A.
Jebrila, Hussein R. Al-Zoubib and Qasem Abu Al-Haijac In
addition to the preprocessing step includes in particular
three levels. In the primary section, they employed word
segmentation to extract characters. In the second one
section, Histograms of Oriented Gradient (HOG) areusedfor
feature extraction. The very last phase employed Support
Vector Machine (SVM) for classifying characters. They have
carried out the proposed method for the recognition of
Jordanian metropolis, city, and village names as a case
examine, similarly to many other phrases that offers the
characters shapes that aren't included with Jordancites.The
set has cautiously been selected to include each Arabic
character in its all forms. To the conclusion, they have got
constructed their own dataset inclusive of greater than
43.000 handwritten Arabic phrases(30000usedfortraining
and 13000 used for testing stage). Recognition result show
99% rate of accuracy[12].
2.2 Devanagari Scripter Character Recognition
In 2011 Gyanendra K. Verma, Shitala Prasad, and Piyush
Kumar Curvelet present in approach for Hindi handwritten
character recognition using curvelet transformer. Thestudy
are used dataset that contain 200 images of character (each
image contains all Hindi characters). Feature extract using
curvelet transform and for recognition k-nearest neighbor
the experiment result show morethan90%accuracy [13].In
2013 Divakar Yadav, Sonia Sánchez-Cuadrado and Jorge
Morato develop optical character recognition system using
neural network for Hindi characters and trained with 1000
dataset. Feature extraction technique is histogram of
projection based on mean distance, on pixel values and
vertical zero crossing. Then classify using back-propagation
neural network with two hidden layers. Experimental result
show 98.5% correct recognition[14].In2015Akanksha Gaur
and Sunita Yadav this system extract feature using k-means
clustering and classified used support vector machine using
linear kernel and Euclidean distance. The evaluation show
that SVM has better results using linear kernel than
Euclidean distance. Maximum achieved using Euclidean
distance is 81.7% accuracy. Using linear kernel giving
95.86% result[5]. In 2018 Nikita Singh present system with
the title “An Efficient Approach for Handwritten Devanagari
Character Recognition Based on Artificial Neural Network”
for recognition hind character. For feature extraction they
used histogram oriented gradients (HOG) and recognition
used artificial neural network (ANN) classifier. The system
get 97.06% high accuracy [15].
3. MAJOR STEPS INVOLVE IN CHARACTER
RECOGNITION
Building an OCR engine is not an easy thing to do as the
main difficulty lies with – identifying each character and
word. For making an OCR engine from scratch below are the
steps which one can follow to make sure that the OCR meets
the desired expectation of character recognition and this is
the methodology and the steps most of researchers used.
3.1 Optical Scanning
To start with an OCR, image can be capture by digital
camera also but after seeing the challenge been faced in
privies work better to use scanner therefor consider first
need putting together a good optical scanner. With the help
of this scanner, an image of original file or document is
captured. It is commanding to select scanner with a good
sensing tool and transport mechanism.

3.2 Pre-processing:
Preprocessing is performing different operation on the
scanned or input image. It helps to remove noise fromimage
make character clear and It basically enhances the image
rendering suitable for segmentation. Preprocessing has
various task are such as converting gray scale, binarization,
thinning, skewing and normalization.
3.3 Segmentation:
Once the preprocessing produces noise free clean character
image, it’s then segmented into several subcomponents.
There are three stepsofsegmentationfirstlinesegmentation
divide the character in image horizontal second word
segmentation the divide words from line sentence last
character segmentation divide the characters from word.
Finally we get segmented charactersthosecharacterhelp for
feature extraction and recognition.
3.4 Feature extraction:
This is one of the riskiest components in an OCR
development. The main aim is to extract important patter
from characteristics. The selected features are expected to
contain pattern that differentiate one character from other
and relevant information from the input data, so that the
classification can be performed by using thosepatterextract
from segmented character this instead of the complete
original data.
3.5 Training and recognition:
Investigation of OCR’s pattern recognition can be done via
template matching, statistical technique, syntactic or
structural techniques, and artificial neural networks. The
system also have to be learn in such a way that the problem
associated to incomplete vocabulary is solved.
3.6 Post-processing:
In this final process, activities like grouping, error detection
and correction take place. During grouping, symbols in the
text are associated with strings. However, it’s impossible to
reach 100% accurate identification of characters, only some
of the errors can be detected and deleted as per the context.
4. CHALLENGES OF OPTICAL CHARACTER
RECOGNITION
For better and high character recognition accuracythere are
so many OCR techniques but still difficult to achieve 100%
correct recognition especially for character that has
similarity. The challenges I observe duringreviewismanyof
them related to the data collection and preprocessing if we
can identify and rid of those challenges we can get high
correct recognition. The following issues created due to
collecting input data using digitals camera. Instead of using
camera to capture characters or scripts prefer to scan the
document but let’s see what those challenges are.
4.1 Scene Complexity
Input data taken with camera may have other object is also
for example building, homes, panting and other objects to
separate those objects from text or character is very tough.
The data that content non textual contents make
preprocessing difficult there for affect the character
recognition process.
4.2 Conditions of Uneven Lighting
Many times image taken from road or outdoor affected by
light and shadows. This is another challenge for optical
character recognition. Itmakedifficulttodetectandsegment
characters. This kind of issues makes scanning document
more preferable than capturing it by camera. Camera light
flash also may help for additional lighting and create
shadows in images.
4.3 Skewness (Rotation)
Image taking using camera also disturb by this issue. The
angle of the image incorrect therefor when we fed this data
to optical character recognition system the outcome will be
incorrect. But there are techniques to solvethisproblemlike
Fourier transformer, projection profile, Hough transform
and so on.
4.4 Blurring and Degradation
This also caused by image taken with camera. This happen
when images are taken from distance, trying to capture on
movements and Lack of focusing. Image taking on this and
other circumstance face blurring and degradation. For
segmentation and accurate recognition sharpness of
characters is needed.
4.5 Fonts and style
Characters that are connected each other like Arabic, Hindi
and fonts style like Italic and other overlap each other this
make difficult for optical character recognition system
during segmentation process hard to detect and divide
words in to character.
4.6 Multilingual Environments
Characters that have multi environment such as, language
that has large number of character Ethiopian, Korean,
Chinese, Japanese and other. Characters that written
connectedly with each other Arabic language. Ethiopian
language Amharic alphabet similarity of characters it’s
difficult for computer to see the difference between most of
them. Therefor this kind of multi environmental characters
are challenges for OCR to divide and extract individual
characters and recognize correctly.

4.7 Damage documents
When the input document are very old and damage whether
we take it in camera or scanned will be very difficult to
observe the character, content many noise when we try to
remove those noise sometime the data or image lose it
necessary content or characters.
5. CONCLUSION
In the research works revised in this paper, character
recognition system use different approaches and many of
them get good accuracy. What we can understand from this
paper is feature extraction techniques should be choose
according to the character you working becauseeachscripts
or alphabets has its own nature therefor need to find
techniques which fit or suitable for characters. The better
able to extract features from character more we can detect
and recognize characters in highest accuracy.
ACKNOWLEDGMENT
First and foremost I thank my supervisor Dr. Ali Imam Abidi
for his guidance in my research work. He gave me the best
advice and helping me with provide me necessarydocument
for my researches” Review on Optical Character
Recognition” ,Who also helped me in the survey of the
related work of different authors .I came to know about so
many new things I am really thankful to them for his care
and support.
REFERENCES
[1] J. Cowell and F. Hussain, “Amharic character
recognition using a fast signature based algorithm,”
Proc. Int. Conf. Inf. Vis., vol. 2003–Janua,pp.384–389,
2003.
[2] I. Stoianov, “Optical Character Recognition of
Historical Documents,” Clover.Slavic.Pitt.Edu, 1995.
[3] V. Patil and S. Shimpi, “Handwritten English
character recognition using neural network,” Elixir
Comp. Sci. Engg, vol. 41, no. 3, pp. 5587–5591, 2011.
[4] K. A. Okrah, “Nyansapo (the wisdom knot): Toward
an African philosophy of education,” Nyansapo (The
Wisdom Knot) Towar. an African Philos. Educ., no.
224, pp. 1–121, 2003.
[5] A. Gaur and S. Yadav, “Handwritten Hindi character
recognition usingk-meansclusteringandSVM,”2015
4th Int. Symp. Emerg. Trends Technol. Libr. Inf. Serv.
ETTLIS 2015 - Proc., pp. 65–70, 2015.
[6] N. M. Noor, M. Razaz, and P. Manley-Cooke, “Global
geometry extraction for fuzzy logic based
handwritten character recognition,” Proc. - Int. Conf.
Pattern Recognit., vol. 2, pp. 513–516, 2004.
[7] D. Nasien, H. Haron, and S. S. Yuhaniz, “Support
Vector Machine (SVM) for english handwritten
character recognition,” 2010 2nd Int. Conf. Comput.
Eng. Appl. ICCEA 2010, vol. 1, pp. 249–252, 2010.
[8] M. S. Sonawane and C. A. Dhawale, “Evaluation of
Character Recognisers:Artificial Neural Network and
Nearest Neighbour Approach,” 2015 IEEE Int. Conf.
Comput. Intell. Commun. Technol.,pp.129–132,2015.
[9] M. M. Altuwaijri and M. A. Bayoumi, “Arabic text
recognition using neural networks,” pp. 415–418,
2002.
[10] A. AbdelRaouf, C. A. Higgins, T. Pridmore, and M. I.
Khalil, “Arabic character recognition using a Haar
cascade classifier approach (HCC),” Pattern Anal.
Appl., vol. 19, no. 2, pp. 411–426, 2016.
[11] N. Lamghari, M. E. H. Charaf, and S. Raghay, “Hybrid
Feature Vector for the Recognition of Arabic
Handwritten Characters UsingFeed-ForwardNeural
Network,” Arab. J. Sci. Eng., vol. 43, no. 12, pp. 7031–
7039, 2018.
[12] N. A. Jebril, H. R. Al-Zoubi, and Q. Abu Al-Haija,
“Recognition ofHandwritten ArabicCharactersusing
Histograms of Oriented Gradient (HOG),” Pattern
Recognit. Image Anal., vol. 28, no. 2, pp. 321–345,
2018.
[13] R. Rani, R. Dhir, and G. S. Lehal, “InformationSystems
for Indian Languages,” Commun. Comput. Inf. Sci.,vol.
139, no. January 2016, pp. 174–179, 2011.
[14] D. Yadav, S. Sánchez-Cuadrado, and J. Morato,
“Optical character recognition for Hindi language
using a Neural-network approach,” J. Inf. Process.
Syst., vol. 9, no. 1, pp. 117–140, 2013.
[15] N. Singh, “An Efficient Approach for Handwritten
Devanagari CharacterRecognitionbasedonArtificial
Neural Network,” 2018 5th Int. Conf. Signal Process.
Integr. Networks, SPIN 2018, pp. 894–897, 2018.

IRJET- Review on Optical Character Recognition

More Related Content

What's hot (18)

Similar to IRJET- Review on Optical Character Recognition (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Review on Optical Character Recognition