International Journal of Computer Applications (0975 – 8887)
Volume 160 – No 6, February 2017
20
A Review on Optical Character Recognition Techniques
Hiral Modi
P. G. Scholar
CSE Department
Gujarat Technological University
Ahmedabad, India
M. C. Parikh, PhD
Associate Professor
CSE Department
Gujarat Technological University Ahmedabad,
India
ABSTRACT
At present scenario, there is growing demand for the software
system to recognize characters in a computer system when
information is scanned through paper documents. This paper
presents detailed review in the field of Optical Character
Recognition. Various techniques are determined that have
been proposed to realize the center of character recognition in
an optical character recognition system. OCR (Optical
Character Recognition) translates images of typewritten or
handwritten characters into the electronically editable format
and it preserves font properties. Different techniques for pre-
processing and segmentation have been surveyed and
discussed in this paper.
General Terms
Pattern Matching.
Keywords
Character Recognition System, Image Segmentation, OCR,
Preprocessing, Skew correction, Classifier.
1. INTRODUCTION
OCR (Optical Character Recognition) translates images of
typewritten or handwritten characters into machine editable
format. OCR reads damaged or low-quality codes and returns
the best guess at what the code is. It is widely used as a form
of information entry from printed paper data records, whether
passport documents, invoices, bank statements, computerized
receipts, business cards, mail, printouts of static data, or any
suitable documentation. OCR does not deal with quality and
sharpness of characters. To overcome the limitations of OCR
a new approach comes into picture which is OCV.
Projection Profile-based methods used makes segmentation
easy to separate the text in document image into lines, words,
and characters independent of the Language in the Text.
Different methods are used at each intermediate stage of
OCR. Text Segmentation is done using Projection Profile
method. They proposed an algorithm for correction of the
skew angle of the text document [1]. Blur is the important
factor that damages OCR accuracy. In this paper prediction
method based on a local blur estimation is proposed. The
relation between blur effect and character size is investigated
which is useful for the classifier. Classifier separates the given
document into three classes: readable, intermediate, non-
readable classes [2].
The grading system is used to evaluate the performance of
printed text using various quality measures. The recognition
results showed high recognition rate as the system was able to
perform a recognition rate of 98.69 % along with a precision
of 0.9857 and a sensitivity of 1 [3]. This paper presents
complete OCR (Optical Character Recognition) system for
camera captured image/graphics embedded textual documents
for handheld devices [4]. Paper [5] describes the skew
detection and correction of scanned document images written
in Assamese language using the horizontal and vertical
projection profile analysis OCR consists of many phases
such as Pre-processing, Segmentation, Feature Extraction,
Classifications and Recognition [6].
1.1 Digitization
Digitization is the process of converting a paper-based
handwritten document into electronic format. Here, each
document consists of only one character. The electronic
conversion is accomplished by using a method whereby a
document is scanned and an electronic representation of the
original document as an image file format is produced. The
author used various scanners for digitization, and the digital
image was going for next step that is a preprocessing phase.
1.2 Pre-processing
In The pre-processing phase, there is a series of operations
performed on the scanned input image. It enhances the image
rendering it suitable for segmentation the gray-level character
image is normalized into a window sized. After noise
reduction, a bitmap image is produced. Then, the bitmap
image was transformed into a thinned image.
1.3 Segmentation
The Segmentation phase is the most important process.
Segmentation is done by separation from the individual
characters of an image. Segmentation of handwritten
characters into different zones (upper, middle and lower zone)
and characters is more difficult than that of printed documents
that are in standard form. This is mainly because of variability
in a paragraph, words of line and characters of a word, skew,
slant, size and curved. Sometimes components of two adjacent
characters may be touched or overlapped and this situation
creates difficulties in the segmentation task. The touching or
overlapping problem occurs frequently because of modified
characters in upper-zone and lower-zone.
1.4 Feature Extraction
In this phase, features of individual character are extracted.
The performance of an each character recognition system that
depends on the features that are extracted. The extracted
features from input character should allow classification of a
character in a unique way. Different types of features are
available like diagonal features, intersection, open-ended
features, zoning features.
International Journal of Computer Applications (0975 – 8887)
Volume 160 – No 6, February 2017
21
Figure 1 Phases of General Character Recognition System
2. METHODOLOGY
One of the most important steps of offline character
recognition system is skew detection and correction which has
to be used in scanned documents as a pre-processing stage in
almost all document analysis and recognition systems. This
paper describes the skew detection and correction of scanned
document images written in Assamese language using the
horizontal and vertical projection profile analysis [5].
Documents with background images in OCR cause an error. A
non-linear transformation is used to enhance the contrast of
each channel image. The method was tested using Tesseract
(an open source OCR engine) and compared with two
commercial OCR software ABBYY Finereader and
HANWANG (OCR software for Chinese characters). The
experimental results show that the recognition accuracies are
improved significantly after removing background images [7].
For pre-processing Fourier Transform is used which
decomposes an image into sine and cosine components with
increasing frequencies. Fourier transform converts spatial
domain onto frequency domain which is easily used for
further processing [1]. Reading text from photographs is a
challenging problem. They applied recently developed
machine learning algorithms for learning the features
automatically from unlabeled data. They proposed text
detection and recognition system based on a scalable feature
learning algorithm and applied it to images of text in natural
scenes [8]. Since past few years, research has been performed
to develop machine printed Chinese/English characters. In this
paper, they described the search and fast match techniques.
High-performance Chinese/English OCR engine is used to
construct a large vocabulary. They have collected 1862 text
lines from varied sources such as newspapers, magazines,
journals, books, etc [9].
H. Wang and J. Kangas [10] proposed a method of
identifying character- like regions in order to extract and
recognize characters in natural color scene images
automatically. Connected component extraction is used to
check the block candidates. Priority adaptive segmentation
(PAS) is implemented to obtain accurate foreground pixels of
the character in each block. Paper [11] presented a system for
text extraction based on the open-source OCR algorithm. The
system is used for functional verification of TV sets. J. Diaz-
Escobar [12] proposed a new method for recognition of
content-less characters in degraded images using the phase
congruency and local energy model. The suggested phase
features are invariant to non-uniform illumination and slight
geometric distortions. Degraded images were compared with
that of the SIFT method in terms of recognition metrics.
Another approach in the paper [13] Hauling the scene text
from image and video is challenging due to the complex
background, changeable font size, dissimilar style, unknown
layout, poor resolution and blurring, position, viewing angle
and so on. For text extraction region and connected
component based methods are used. Artificial Neural network
(ANN) is used as the classifier to filter out the text and non-
text components.
There are natural variations in human writing so designing a
reliable OCR system is a challenging task. An algorithm
based on Kohonen Neural Network is presented in this paper.
Kohonen algorithm that is one of Artificial neural network
The experiments also demonstrated that system complexity
can be reduced significantly without degrading performance
by considering two-layered neural network rather than
multiple layered neural networks [14]. In this paper [15] a
complete OCR methodology for recognizing historical
documents, either printed or handwritten without any
knowledge of the font, is presented. The pre-processing and
segmentation approach is used in order to detect text lines,
words, and characters. Yaeger [16] has proposed a
handwritten character recognition system. The proposed
system works by using the neural network techniques. For the
recognition of characters, a multi-layer perceptron is used by
this system and it gives better results. J. Hu et. al [17]
proposed a system in which high-level features are combined
with low-level features on simple points and these are able to
cover a huge amount of input patterns. Also, these features
have invariance property which is used for normalizing the
curvature of features. Funanda [18] has proposed a system
which uses the HMM for the recognition of the online
handwritten recognition. The proposed system reduces the
usage of memory and also it improved the recognition rate of
online handwritten characters. In paper [1] Horizontal
Projection Profile and Vertical Projection Profile methods are
used for segmentation. Different methods are used at each
intermediate stage of OCR. Text Segmentation is done using
Projection Profile method. They proposed an algorithm for
correction of the skew angle of the text document.
J. r´ı Matas [19] presented an end-to-end real-time scene text
localization and recognition method. In the first stage of the
classification, the probability of each ER being a character is
estimated using novel features calculated with O(1)
complexity. In second stage only ERs with locally maximal
probability are selected.
Huei-Yung Lin and Chin-Yu Hsu [20] presented neural
network based approach which reduces the training time and
maintains the high recognition rate. Multi-stage approach and
Preprocessing
Feature to be extracted
Search
Image Library
A
Matched letter
Match
A
Image
see
International Journal of Computer Applications (0975 – 8887)
Volume 160 – No 6, February 2017
22
pre-processing are done for the experiment. Preprocessing is
performed to partition the training data prior to training stage.
In this paper [21], a computer vision and character recognition
algorithm for a license plate recognition (LPR) is presented to
be used as a core for intelligent infrastructure like electronic
payment systems (toll payment, parking fee payment),
freeway. Based on the connected component analysis and
novel adaptive image segmentation technique is presented
[21].
3. COMPARISON
Paper [5] presented that projection profile is used as a suitable
feature for skew detection. Vertical Projection Profile
Analysis allows small noise which produces error where
Horizontal Projection Profile Analysis reduces the effect of
noise. The time complexity of Vertical is high with compared
to horizontal projection profile. In paper [7] author proposed a
method which is used to remove the background image from
pilling up. In government agencies and independent
organizations, OCR simplifies data collection and analysis,
among other processes, document. The experiment is done
using three OCR software tool: HANWANG OCR, ABBYY,
and Tesseract. With compared to Tesseract OCR,
HANWANG OCR, and ABBYY OCR better because there
are built-in functions are available to preprocess image before
text extraction
In paper [8] they trained their character classifier with
features. They tested 5198 characters from 62 classes 26
upper- and 26 lower-case letters). Accuracy for the largest
system (1500 features) is the highest, at 81.7% for the 62-way
classification problem.
Segmentation is an important stage of OCR in image
processing. In this paper [22] they surveyed different
techniques which are available for segmentation. Most
methods are categorized into three groups: the analytical, the
empirical goodness and the empirical discrepancy groups.
Segmentation algorithms can be evaluated analytically or
empirically, so the evaluation methods can be divided into
two categories: the analytical methods and the empirical
methods. The analytical methods directly examine and assess
the segmentation algorithms themselves by analyzing their
principles and properties. The empirical methods indirectly
judge the segmentation algorithms by applying them to test
images and measuring the quality of segmentation results.
Empirical methods are classified into two types: empirical
goodness and empirical discrepancy method. In first method
properties of segmented images are measured
using”goodness” parameters. Where in the second type some
references that present the ideal or expected segmentation
results are first found.
Devices like Personal Data Assistants (PDA) which is pen
input devices require good online handwriting character
recognition algorithms. A. Funada et al. [18] proposed a new
algorithm to recognize on-line handwriting and it utilize
HMM(Hidden Markov Model). The memory reduction rate is
a function of the matrix size and the number of states. They
performed character segmentation, character classification
which is fairly standard multilayer perceptron trained with
error backpropagation provides the ANN character classify
4. APPLICATION
Optical Character Recognition is a vast field with a number of
varied application which is described below [23]. For OCR
enhanced image segmentation algorithm based on histogram
equalization using genetic algorithms are used.
4.1 Captcha
A CAPTCHA is a program that can generate and grade tests
that human can pass but current computers programmers’
cannot. In CAPTCHA, an image consisting of series of letters
of number is generated which is obscured by image distortion
techniques, size and font variation, distracting backgrounds,
random segments, highlights, and noise in the image. This
system can be used to remove this noise and segment the
image to make the image tractable for the OCR (Optical
Character Recognition) systems.
4.2 Institutional Repositories and Digital
Libraries
Institutional repositories are digital collections of the outputs
created within a university or research institution. It is an
online locale of intellectual data of an institution, especially a
research institution where it is collected, preserved and aired.
It helps to open up the outputs of an institution and give it
visibility and more impact on worldwide level
4.3 Invoice Imaging
Invoice imaging is widely used in many businesses
applications to keep track of financial records and prevent a
backlog of payments from pilling up. In government agencies
and independent organizations, OCR simplifies data
collection and analysis, among other processes.
4.4 Automatic Number Recognition
Automatic number plate recognition [6] is used as a mass
surveillance technique making use of optical character
recognition on images to identify vehicle registration plates.
ANPR has also been made to store the images captured by the
cameras including the numbers captured from the license
plate.
.
International Journal of Computer Applications (0975 – 8887)
Volume 160 – No 6, February 2017
23
Figure 2 Evaluation of Segmentation Algorithm
Table 1. Comparison of the different OCR Techniques
Author(s)
Data set
Method
Recognition rate
Yaeger et al. [16]
(1998)
(A-Z) characters, (0-9) digits, 23 symbols
with
writer independent system
Multi-Layer
Perceptron
21.3%
Hu et al.[17]
(2000)
(a) 500, 1000 and 2000 unipen database.
(b) 5000, 10000 and 20000 unipen
database.
Hidden Markov
Model
91.8%, 90.5% and 87.2%for
(a) dataset and 83.2%,
79.8%
and 76.3% for (b)
dataset.
Funanda et al.
[18]
(2004)
Kanji, Katakana, Hirangana, Western
alphabets and symbols with writer
Independent system.
Hidden Markov
Model
91.34%
A. F. Mollah et
al. [4]
(2011)
Set of 100 business cards images.
Segmentation using
Vertical Projection Profile
92.74%
M. Shen [7]
(2015)
1160 images with various resolutions, font
sizes and noise levels.
Image Enhancement using
non-linear Transformation
-
J. B. Pedersen et
al. [3]
(2016)
100 images with a total of 840 characters
Character based
segmentation and Nearest
Neighbour Classifier
98.69%
V. Kieu et al. [2]
(2016)
IPAD contains 297 document images and
PME contains 1998 document images
Fuzzy-C-Means clustering
method
90.57%
C. N. E.
Anagnostopoulos
et al. [21]
(2006)
1334 natural-scene gray-level vehicle images
probabilistic neural
network (PNN)
96.5% ( Segmentation)
89.1%(Entire Plate
Recognition)
A. Coates et al.
[8]
ICDAR data set 5198 test characters
Machine Learning
Algorithm
85.5%
4.5 Legal Industry
The legal industry is also one of the beneficiaries of the OCR
technology. OCR is used to digitize documents and directly
entered into a computer database.
4.6 Banking
Another important application of OCR is in banking, where it
is used to process cheques without human involvement
Cheque can be inserted into a machine where the system scans
the amount to be issued and the correct amount of accessed as
necessary
4.7 Healthcare
Healthcare has also seen an increase in the use of OCR
technology to process paperwork. Healthcare professionals
always have to deal with large volumes of forms for each
patient, including insurance forms as well as general health
forms. To keep up with all of this information, it is useful to
SEGMENTATION
Analytical
Method
Empirical
Method
Empirical Goodness
Method
Empirical Discrepancy
Method
International Journal of Computer Applications (0975 – 8887)
Volume 160 – No 6, February 2017
24
input relevant data into an electronic database that can be
accessed as necessary.
5. CONCLUSION
“This paper elaborated survey of disparate techniques for
OCR” has been studied. Handwritten character, natural scene
images, business cards and TV set images are selected for
experimentation. A systematic flow of OCR system is
discussed. In this paper projection profile based method for
segmentation, fourier transform technique is for pre-
processing, and nearest neighbour classifier for classification
are described. This paper can be helpful to the researcher for
selecting most appropriate techniques to achieve optimum
results for application according to a different parameter
described in the previous section.
6. REFERENCES
[1] A. S. Sawant, “Script Independent Text Pre-processing
and Segmentation for OCR,” Int. Conf. Electr. Electron.
Signals, Commun. Optim. - 2015, pp. 1–5, 2015.
[2] V. Kieu, F. Cloppet, and N. Vincent, “OCR Accuracy
Prediction Method Based on Blur Estimation,” 2016 12th
IAPR Work. Doc. Anal. Syst., pp. 317–322, 2016.
[3] J. B. Pedersen, K. Nasrollahi, and T. B. Moeslund,
“Quality Inspection of Printed Texts,” IWSSP 2016- 23rd
Int. Conf. Syst. Image Process. 23-25 May 2016,
Bratislava, Slovakia, pp. 6–9, 2016.
[4] A. F. Mollah, N. Majumder, S. Basu, and M. Nasipuri,
“Design of an Optical Character Recognition System for
Camera- based Handheld Devices,” IJCSI, vol. 8, no. 4,
pp. 283–289, 2011.
[5] B. Jain and M. Borah, “A Comparison Paper on Skew
Detection of Scanned Document Images Based on
Horizontal and Vertical,” IJSRP, vol. 4, no. 6, pp. 4–7,
2014.
[6] E. N. Bhatia, “Optical Character Recognition
Techniques : A Review,” IJARCSSE, vol. 4, no. 5, pp.
1219–1223, 2014.
[7] M. Shen, “Improving OCR Performance with
Background Image Elimination,” 2015 12th Int. Conf.
Fuzzy Syst. Knowl. Discov., pp. 1566–1570, 2015.
[8] A. Coates et al., “Text Detection and Character
Recognition in Scene Images with Unsupervised Feature
Learning.”
[9] P. Road, “Confidence Guided Progressive Search and
Fast Match Techniques for High Performance
ChineseEnglish OCR *,” IEEE, pp. 89–92, 2002.
[10] H. Wang and J. Kangas, “Character-Like Region
Verification for Extracting Text in Scene Images,” no.
11, 2001.
[11] I. Kastelan, S. Kukolj, V. Pekovic, V. Marinkovic, and Z.
Marceta, “Extraction of Text on TV Screen using Optical
Character Recognition,” IEEE, pp. 153–156, 2012.
[12] J. Diaz-escobar, “Optical Character Recognition based
on phase features,” IEEE, 2015.
[13] A. Thilagavathy, K. Aarthi, and A. Chilambuchelvan, “A
Hybrid Approach to Extract Scene Text from Videos,”
ICCEET, pp. 1017–1022, 2012.
[14] S. Goyal, “Optical Character Recognition,” IJARCSSE,
vol. 3, no. 11, pp. 982–985, 2013.
[15] G. Vamvakas, B. Gatos, N. Stamatopoulos, and S. J.
Perantonis, “A Complete Optical Character Recognition
Methodology for Historical Documents,” pp. 525–532,
2008.
[16] L. S. Yaeger, B. J. Webb, and R. F. Lyon, “Search for
Online , Printed Handwriting N EWTON,” Am. Assoc.
Artif. Intell., vol. 19, no. 1, pp. 73–90, 1998.
[17] J. Hu, S. G. Lim, and M. K. Brown, “Writer independent
on-line handwriting recognition using an HMM
approach,” J. PATTERN Recognit. Soc., vol. 33, pp. 133–
147, 2000.
[18] A. Funada, D. Muramatsu, and T. Matsumoto, “The
Reduction of Memory and the Improvement of
Recognition Rate for HMM On-line Handwriting
Recognition,” IEEE, pp. 0–5, 2004.
[19] J. r´ı Matas, “Real-Time Scene Text Localization and
Recognition,” IEEE, pp. 3538–3545, 2012.
[20] H. Lin and C. Hsu, “Optical Character Recognition with
Fast Training Neural Network,” IEEE, pp. 1458–1461,
2016.
[21] C. N. E. Anagnostopoulos, I. E. Anagnostopoulos, V.
Loumos, and E. Kayafas, “A License Plate-Recognition
Algorithm for Intelligent Transportation System
Applications,” IEEE, vol. 7, no. 3, pp. 377–392, 2006.
[22] Y. J. Zhang, “A survey on evaluation methods for image
segmentation,” pp. 1–13.
[23] A. Singh, K. Bacchuwar, and A. Bhasin, “A Survey of
OCR Applications,” Int. J. Mach. Learn. Comput., vol. 2,
no. 3, pp. 314–318, 2012
.
IJCA
TM
: www.ijcaonline.org