It gives a detailed description of OCR and the recent advancements in that fields and the applications of OCR

1. A Detailed study and recent research on OCR
Proddutur Shruthi Dr. Devaraj Verma C
(M.Tech) in Data Science ORCID-ID: 0000-0002-1504-4263
Department of specialization Associate Professor in Computer science
Jain (Deemed to be University), Jain (Deemed to be University),
Bangalore, Karnataka, India Bangalore, Karnataka, India Email:
shruthiamar.2012@gmail.com Email: devarajverma04@gmail.com
Abstract:
This paper provides a total overview
of OCR. Optical character recognition is
nothing but the ability of the computer to
collect and decipher the handwritten inputs
from documents, photos or any other
devices. Over these many years, many
researchers have been researching and
paying attention on this topic and proposed
many methods which can be solved. This
research provides a historical view and the
summarization of the research which done
on this field.
Keywords: Historical view, handwritten
recognition, OCR
1. Introduction:
The handwritten recognition
problems are addressed by many researchers
from time to 𝑡𝑖𝑚𝑒1,2,3,4
. OCR has been
expanded to many languages other than
English like Chinese, Japanese, Devangiri,
Arabic, Hangul etc. In Japanese it can also
be applied on Katakana syllabic characters,
Kanji characters5
. Although so many
decades passed, the most difficult in OCR is
the Arabic handwritten texts than the other
language handwritten 𝑡𝑒𝑥𝑡𝑠6
. Handwritten
recognition is described as changing over
the manually written content into a
notational portrayal. Character recognition
can be isolated into two methods: 1. Online
2. Offline. It can be differentiated with type
the type of the data what we have. Online
character recognition is defined as
identification of the characters which are
obtained from the equipments like smart
pen, pressure tablets etc and then we will be
the characteristics like velocity and pressure.
While we come to Offline character
recognition, it is nothing but recognizing the
characters or numerical which are written on
paper and given as input in the form of
digital image which are captured by the
camera.
Offline character recognition has
many applications like postal sorting, bank
cheque, postal address recognition, tax
forms etc, so they carry a lot of research,
and also presented a historical view on this.
There were also a survey on the techniques
some of the 𝑝𝑎𝑝𝑒𝑟𝑠7,8,9,10
. There is also a
survey on online and offline handwritten
𝑟𝑒𝑐𝑜𝑔𝑛𝑖𝑡𝑖𝑜𝑛11
. The most difficult part in
recognition is the cursive handwritings,
these are the most difficult 𝑡𝑎𝑠𝑘12
. In OCR,
basically we need to find the text in the
image and then we should determine what
texts are present in the image. So it is a two
step process. So, there are basically there

2. strategies: 1. Computer vision technique. 2.
Deep learning Approach. 3. Advance deep
learning.
OCR manages the issue of
perceiving optically prepared characters.
Optical acknowledgment is performed
disconnected after the composition or
printing has been finished, instead of on-line
acknowledgment where the PC perceives the
characters as they are drawn. Both hand
printed and printed characters might be
perceived, yet the presentation is
straightforwardly reliant upon the nature of
the information reports.
Figure 1. Various regions of CR
This paper consists of 5 sections.
Section 2 describes about the historic
development, section 3 discusses about the
achievements in OCR, section 4 discusses
about the phases in OCR, section 5 discusses
about the applications, and last section is
nothing but a conclusion.
2. History of OCR:
There are a lot of dialects on this
planet. So many, truth be told, that some of
them have blurred from memory and require
a considerable amount of criminologist work
to interpret. Old Egyptian hieroglyphics
were a serious riddle until the revelation of
the Rosetta stone. At the point when we can
see a case of something close to a case of
something different, it is conceivable to
think about those things. At the point when
we see, for instance, Greek (a language we
know) close to Ancient Egyptian (a
language we lost), we can methodically
make inferences. Shockingly, that is the
means by which OCR does something
amazing.
In earlier days lots of people
dreamed about OCR device which can read
and write a human written characters and
numbers, and then in late 1920s an
Australian engineer Gustav 𝑇𝑎𝑢𝑠ℎ𝑒𝑐𝑘13
developed a reading machine in Germany
which helped the blinds to read and the next
machine was invented by Paul 𝐻𝑎𝑛𝑑𝑒𝑙14
in
USA in 1933 and named that machine as
Statistical machine. These were the first
OCR concepts at that time. Around then
certain individuals longed for a device
which could understand letters and numbers.
This stayed a fantasy until the period of PCs
showed up, in the 1950's. Notwithstanding,
we think their essential thought merits
referencing, on the grounds that it is as yet
alive. Back in time many papers were
published on the history of 𝑂𝐶𝑅21,22,23
.
Despite the fact that the inception of OCR
can be found as ahead of schedule as 1870,it
originally showed up as a guide to the
outwardly disabled, and the earliest fruitful
endeavor which was invented by a Russian
researcher Tyurin in 190018
.
Character
Recognition
Offline
Recognition
Single
Characters
Imprint
Manually
written
Handwritten
scripts
Recognition Validation
Online
Recognition

3. In 1940’s, a much more advanced
version of OCR came into existence as there
was a development in the computers. Thus it
was acknowledged as an information
handling approach with significance to the
career society. The chief inspiration for the
advancement of OCR frameworks is the
need to adapt to the tremendous surge of
paper, for example, cheque, business
structures, administration records, credit
engravings and mail arranging produced by
the extending mechanical community. OCR
devices have been industrially accessible
from the center of the 1950s. From that
point forward broad examination has been
done and countless specialized articles and
studies have been distributed by different
analysts in the region of character
acknowledgment. A few publications have
been distributed on optical character
𝑎𝑐𝑘𝑛𝑜𝑤𝑙𝑒𝑑𝑔𝑚𝑒𝑛𝑡19,20,21
. During the most
recent years a huge amount of investigation
has been put on in the branch of Optical
Character Recognition (OCR).
In 1917, an 18-year-old Mary
Jameson exhibits the Optophone. Light from
a printed page thinks about a selenium cell,
and the machine lets out a melodic harmony.
Ms. Jameson, who is visually impaired, can
peruse the print at a record-breaking rate of
single word every moment. It's a decent
beginning, yet the future has better toys
coming up. In 1951, David Shepard builds
up a machine that could perceive every one
of the 26 letters of the Latin letters in order,
as delivered by a standard sort author. He
calls it "Gismo", which later advances into
the Farrington Machine. By the 1960s, OCR
innovation is being utilized all at once in
mail-arranging by the U.S. postal help. In
1974, Kurzweil Technologies delivers the
CCD flatbed scanner, the first Omni-text
style optical character acknowledgment
framework. Prior to 1974, OCR could just
peruse textual styles exceptionally intended
for machine 𝑐𝑜ℎ𝑒𝑟𝑒𝑛𝑐𝑒22
. During the 70s,
the extent of OCR is broadening, and the
mechanical universe is starting to envision
its future effect. During the period of 1931-
1954 the Intelligent Machines Research
Corporation was the first company to create
a machine which was able to interpret the
Morse code and read the text aloud and
named GISMO, they were also able to sell
these machines. These were the first OCR
tools which were invented in the
𝑖𝑛𝑑𝑢𝑠𝑡𝑟𝑦23
.
Various business items have been
delivered that transform digitized records
into word documents, typically in ASCII
design. In spite of the fact that particular
items cycle machine printed records
effectively, with regards to manually written
reports the outcomes are not agreeable
enough. Additionally, such items can't deal
with chronicled archives because of their
inferior quality, absence of standard letter
sets and nearness of obscure fonts. To this
end, acknowledgment of recorded reports is
one of the most testing undertakings in
𝑂𝐶𝑅24
.
A. The Generations of OCR:
A few business OCR's showed up in the
start of the 1960's; these were yields of the
time of slash and attempt. The original can
be portrayed by the obliged characters
outline which the OCR's perused. These
straightforward techniques were viable. The

4. most normal in particular was the
NCR 42033
, which in particular has a
unique textual style for numerals and five
images, called NOF or bicodes, had been
planned. The later normal one is the
Farrington 301034
, of Farrington
Electronics Inc. The storyline was
equivalent to the NCR's Optical character
recognition; i.e., they additionally utilized an
extraordinary text style, called Selfchek 12F,
7B. IBM was additionally extremely
dynamic in creating OCR, perceiving that
OCR was a significant information gadget
for the PC. The principal popularized OCR
was the International Business
Machines 141835
, which was intended to
peruse an uncommon International Business
Machines text style, 407. An info character
was taken care of to a 17 segment by 10
column move record that made the devices
safe to upright posture variety. The character
shape focused more common than the duplet
word method referenced previously. The
acknowledgment strategy was legitimate
format coordinating; be that as it may,
positional relationship was completely
utilized. The rationale was very intricate so
as to adapt to report varieties.
The first generation of OCR is the
business OCR frameworks showing up in
the epoch of 1960 - 1965 might be referred
as the original of Optical character
recognition. This age of Optical character
recognition devices were predominantly
described by the obliged character shapes
read. The images were extraordinarily
intended for machine perusing, and the
initial ones didn't look normal. With time
multifont machines begun to show up, this
could peruse up to 10 distinct textual styles.
The quantity of textual styles were restricted
by the example acknowledgment strategy
applied, format coordinating, which thinks
about the letter images with a study of
model snaps for every one character of each
textual mode.
The second generation of OCR is the
perusing machines of the subsequent age
showed up in the center of the 1960's and
mid 1970's. These frameworks had the
option to perceive ordinary machine carved
characters and furthermore had manually-
carved character acknowledgment abilities.
At the point when hand-printed characters
were thought of, the letters set was
compelled to numerical and a couple of
words and images. The foremost and well
known arrangement of the particular sort
was the International Business Machines
Corporation 1287, which was shown at the
World Fair in New York in 1965.
Additionally, in this era Toshiba built up a
main programmed character arranging
device for ZIP code numerical and Hitachi
made the primary Optical character
recognition device for elite as well as
minimal effort. In this era huge tasks was
completed in the territory of normalization.
During 1966, an exhaustive investigation of
Optical character recognition necessities was
finished and an U.S level Optical character
recognition letter set was characterized;
Optical character recognition-A. In this text
style was profoundly adapted and intended
to encourage imaged acknowledgment,
albeit even now comprehensible to people.
An EU text style was additionally
structured, Optical character recognition-B,
that had many common textual styles than
the U.S norm. A few endeavors were made

5. to combine the two text styles into one
norm, however rather machines having the
option to peruse the two guidelines showed
up.
Figure 2. OCR-A
Figure 3. OCR-B
The third wave of OCR frameworks,
showing up in the center of the 1970's, the
test was archives of low standard and huge
printed and manually written letter sets.
Minimal effort and superior were
additionally significant targets, which were
helped by the emotional development in
equipment innovation. Albeit more modern
OCR-machines began to show up at the
market straightforward OCR gadgets were
still extremely valuable. The era prior to the
PCs and laser printers began to overwhelm
the territory of text creation, composing was
an extraordinary specialty for OCR. The
constant print separating and modest count
of textual styles made basically planned
OCR gadgets exceptionally valuable.
Unfinished copies could be made on normal
typewriters and took care of into the PC
through an OCR gadget for definite altering.
Thusly word processors, which were a
costly asset right now, could uphold a few
people and the expenses for hardware could
be cut.
TimeLine Details
1870 The absolute first
endeavors to OCR
1940 The advanced variant
of optical character
recognition
1950 The initial OCR
devices show up
1960-1965 First wave of optical
character recognition
1965-1975 Second wave of
optical character
recognition
1975-1985 Third wave of optical
character recognition
1986 - Now optical character
recognition to
individuals
Table 1: A brief optical character
recognition sequence
3. Recent Achievements:
Specialists everywhere on
over the world have accomplished
victories in 𝑂𝐶𝑅3,25,26 ,27
. We can
refer to table 2 to see some of the
results. As we can see the table 2 is
divided into 3 main divisions:
Authors, Algorithms used and
recognition rate.

6. Table 2: Summary of recognition rates
Authors Algorithms
Used
Recognitio
n rate
Yang, J., Ren, P.,
& Kong, 𝑋28
Faster R-
CNN
97%
Ahmed Talat
Sahlol; Mohamed
Abd
Elaziz; Mohamme
d A. A. Al-
Qaness; Sunghwa
n 𝐾𝑖𝑚29
Hybrid
whale
optimization
algorithm
using
neighboring
rough set
96%
T. T. Zin, S. Z.
Maw and P.
𝑇𝑖𝑛30
Mobile Tutor 90%
Z. Huang and Q.
𝑍ℎ𝑎𝑛𝑔31
.
Skew
correction
algorithm
based on
RESNET
95%
B. Dessai and A.
𝑃𝑎𝑡𝑖𝑙32
Convolution
al neural
network
89%
The 𝑝𝑎𝑝𝑒𝑟28
written by Yang, J.,
Ren, P., & Kong, X really gives us a new
method with a new dimension which is
transforming the many problems by their
new design. And this model showed a better
result than the traditional OCR. The dataset
used in this paper are 3 different datasets
which are Letter, word, EMNIST datasets.
The 𝑝𝑎𝑝𝑒𝑟29
written by Ahmed
Talat Sahlol; Mohamed Abd
Elaziz; Mohammed A. A. Al-
Qaness; Sunghwan Kim was written in
MATLAB with the stages preprocessing,
feature extraction and selection by the
algorithm Hybrid whale optimization
algorithm using neighboring rough set,
while the last stage classification was done
in python. The dataset used in this was
CENPARAMI which was published in
Canada. The dataset contains Arabic
characters. The number of features in the
dataset is 261 which give an accuracy of
96% with Precision, Recall, and time as 97,
86 and 1.91.
The 𝑝𝑎𝑝𝑒𝑟30
written by T. T. Zin, S.
Z. Maw and P. Tin uses EMNIST (Extended
MNIST) handwritten datasets. They used
deep convolutional neural networks for
character classification and segmentation.
This method is most effective for school
children and this approach makes the
children to self learn and self correcting the
numerical and character so that they can
improve the hand writing.
Another approach which was
proposed by Z. Huang and Q. Zhang is
Skew correction algorithm based on residual
neural network. This was for Chinese
handwritten characters and two models were
proposed by the one was 4-D classification
and other was 181-angle classification and
these two frameworks were formed on
RESNET. While precisely anticipating the
heading and the edge of the character
pictures, it likewise demonstrates that
rearranging the picture by anticipating the
heading of the character picture will bring
about higher acknowledgment 𝑝𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛31
.
The method proposed by B. Dessai
and A. Patil was nothing but implementing
CNN for OCR which are very efficient in
image classification than SVM’s. CNN are
prepared with trained dataset. They gathered
information tests of 15 characters from
various people. 1500 examples of each
character were utilized for preparing and

7. 250 examples of each character were
utilized for testing. After the system was
prepared obscure characters from word were
given as contribution to the
acknowledgment framework. Working with
Devanagari characters were more
unpredictable on the grounds that of their
compound structures when contrasted with
the roman characters. Likewise, the nearness
of a header line for each word furthermore,
uniqueness recorded as hard copy styles of
each individual increment the intricacy. The
accuracy from this method was 89.34%
including d [ x c j l n ; j e r t y u श ण and
excluding श x ण the accuracy was 91.11%
.To make the OCR framework more
proficient compound manually written
characters which are framed by mixes of
two unique characters can be incorporated
for 𝑎𝑐𝑘𝑛𝑜𝑤𝑙𝑒𝑑𝑔𝑚𝑒𝑛𝑡32
.
4. Phases of recognition
The phases of recognizing a
character or a numerical is basically divided
into nine phases: Acquisition of Image,
Binarization, Slant Correction, Smoothing,
Size Normalization, Preprocess the image,
Feature Extraction, Feature Selection,
Classification and recognition.
A. Acquisition of Image:
Acquisition of image is the first step.
Here, we should get an image or
picture from the camera or other
sources. The image which we get
will be in the form of JPEG or PNG
format. The input may be in colored,
gray or black and white.
B. Binarization:
Binarization of image means we
convert the image into 0 or 1 form
which can alphabet or numerical.
The process becomes much easier
when we do Binarization on an
image which we get as an input and
it also provides us a good result if we
do this process.
Figure 3. Phases of recognition
C. Slant Correction:
Slant is characterized as incline of
the overall composing pattern as for
the vertical line. The picture
framework is isolated into upper and
lower parts. The focuses of gravity
of the lower furthermore, upper parts
are figured and associated. The slant
Image
Binarization
Slant
Correction
Smoothing
Size
Normalization
Preprocessing Feature
Extraction
Feature
Selection
Classification
and Recognition
Output

8. of the interfacing line characterizes
the incline of the 𝑤𝑖𝑛𝑑𝑜𝑤36,37
.
D. Smoothing and noise removal:
Smoothing is a procedure is
proposed to allocate non-zero
likelihood to alter tasks not present
in the preparation corpus. It also
suggests both filling and thinning.
Filling takes out little breaks, holes
and openings in digitized characters
while diminishing decreases width of
𝑙𝑖𝑛𝑒38
.
We should also remove the noise
present in the image it is one of the
methods in preprocessing. If the
noise is present in the image it will
affect the accuracy and by removing,
it enhances the image quality. On the
off chance that commotion signals
between segments of the lines are
definitely not taken out, incredible
holes will emerge; these clamors
ought to be eliminated to acquire all
the critical information. In a lot of
pictures, one of them may have
numerous 𝑛𝑜𝑖𝑠𝑒𝑠39
.
E. Size normalization:
It is required as the size of the
character differs starting with one
individual then onto the next and
furthermore now and again in any
event, when the individual is same.
Standardization helps in likening the
size of the character picture (double
network) so highlights can be
separated on a similar 𝑏𝑎𝑙𝑎𝑛𝑐𝑒12
.
F. Feature Extraction and Feature
Selection:
Feature Extraction is the strategy of
gathering huge materialistic data
from the substance of crude data.
Significant materialistic data is the
exact and viable delineation of
characters. The arrangement of
qualities got from crude data is
alluded to as highlight extraction to
augment the character
acknowledgment rate including
minimal amount of parts. There are
several techniques used to do this
part as it is most crucial part, and its
selection of the right amount of set
of features is the important step in
the classification 𝑝𝑟𝑜𝑐𝑒𝑠𝑠40
. The
algorithms like SIFT, PCA, Genetic
Algorithms, LDA, Histograms etc.
Feature Selection is a procedure to
choose the highlights that is pertinent
for arrangement stage. The objective
of highlight choice (FS) is that of
diminishing the quantity of
highlights to be considered in the
order stage. This assignment is
performed by eliminating unessential
or uproarious highlights from the
entire arrangement of the accessible
ones. Highlight choice is achieved by
decreasing however much as could
reasonably be expected the data
misfortune because of the list of
capabilities decrease: in this manner,
at list on a fundamental level, the
determination cycle ought not to
lessen grouping 𝑒𝑥𝑒𝑐𝑢𝑡𝑖𝑜𝑛41
.

9. G. Classification and recognition:
Classification is characterized as the
way toward grouping a character into
its fitting classification. The basic
way to deal with grouping depends
on connections present in picture
segments. The factual methodologies
depend on utilization of a segregate
capacity to characterize the picture.
Some of the characterization
approaches are Bayesian classifier,
Decision tree, neural system
classifier, SVM’s and so forth. At
long last, there are classifiers
dependent on syntactic methodology
that expects a syntactic way to deal
with make a picture from its sub-
constituents.
There are basically two types of
methods 1. Decision-theoretic
methods and 2. Structural Methods.
Decision-theoretic methods:
Matching: It covers the
gatherings of methods dependent on
likeness estimates where the
separation linking a element vector,
depicting the extricated letter and the
portrayal of every class is
determined. Various steps might be
utilized, yet the regular is the
Euclidean separation. The indicated
base separation distinguisher
functions admirably when the
categories are all around isolated,
that is the point at which the
separation between the methods is
enormous contrasted with the spread
of each class.
At the point when the whole
character is utilized as contribution
to the characterization, and no
highlights are removed (format
coordinating), a relationship
approach is utilized. Here the
separation between the character
picture and model pictures speaking
to every letter category is registered.
Optimum statistical
classifiers: In Statistical order
arrangement a probabilistic way to
deal with acknowledgment is
utilized. By and large, its utilization
gives the most reduced likelihood of
making characterization mistakes.
A classifier that limits the
complete normal misfortune is
known as the Bayes' classifier.
Assume an obscure image portrayed
by its component vector, the
likelihood, this image has a place
with class c is processed for all
categories c=1, 2, 3,…, N. An image
is then relegated the class which
gives the most extreme likelihood.
For this plan to be ideal, the
likelihood thickness elements of the
images of every category must be
understood, alongside the likelihood
of event of each class. The last is
generally fathomed by expecting that
all classes are similarly plausible.
The thickness work is generally
thought to be regularly dispersed,
and the nearer this supposition that is
to the real world, the closer the
Bayes' classifier comes to ideal
conduct.
The base separation classifier
depicted above is determined totally
by the mean vector of every

10. category, and the Bayes classifier for
Gaussian categories is indicated
totally by the mean vector and
covariance network of every
category. These boundaries
determining the classifiers are gotten
through a preparation cycle. In the
course of this cycle, preparing
examples of every category is
utilized to figure these boundaries
and portrayals of every category are
acquired
Neural-Networks: As of
now, the utilization of neural systems
to perceive letters has reemerged. In
view of a back-spread system, this
system is made out of a few layers of
interconnected components. A
component vector enters the system
at the info sheets. Every component
of the layer processes a weighted
whole of its info and changes it into
a yield by a nonlinear capacity. In
preparing the loads at every
association are balanced till an ideal
yield is acquired. An issue of neural
systems in optical character
recognition might be belonging to
restricted consistency and consensus,
phase a bit of leeway is their
versatile nature.
Structural Methods: Syntactic
methods are more common prevalent
approaches.
Syntactic Methods:
Proportions of likeness dependent on
connections between basic segments
might be figured by utilizing
linguistic ideas. The thought is that
every category will have their own
punctuation characterizing the
synthesis about the letter. The syntax
might be spoken to as strings or
trees, and the auxiliary parts
separated originating at an obscure
letter are coordinated contrary to the
sentence structures of each category.
Assume in particular, we have dual
diverse character categories which
could be produced over the two
punctuation G1 as well as G2,
individually. Consider an obscure
letter; we state that it is similar to the
top of the line in the event that it
might be created by the language G1,
yet not by G2.
5. Applications of OCR:
1. Handwritten recognition: It is
the capacity of a PC to get and decipher
coherent manually written contribution from
sources, for instance, articles archives,
images, contact monitor and various
appliances. The snaps made substance may
be distinguished "disconnected" from a
touch of note-paper by optical examining or
smart word affirmation. On the other hand,
the advancements of the writing instrument
tip might be recognized "on line", for
instance by a writing instrument-based
computer display 𝑠𝑢𝑟𝑓𝑎𝑐𝑒11
.
2. Automatic number-plate
recognition: It is a technology where it uses
a technique called OCR which reads the
numbers on the number plate through
images which can be used for vehicle
location. ANPR has likewise been made to
store the photos caught by the cameras

11. containing the numerical caught from
𝑝𝑒𝑟𝑚𝑖𝑡42 plate. Automatic number-plate
recognition advancement own to plate
assortment all around as it is a district
explicit innovation. These are used by
different law enforcement officer powers
and as a strategy for voltaic cost assortment
on pay-as-you-use streets and recording the
developments of congestion or people.
3. Data Entry: This zone covers
innovations for entering a lot of limited
information. At first such record perusing
devices were utilized for banking use. The
frameworks are described by perusing just a
very restricted arrangement of printed
characters, typically numerals and a couple
of exceptional images. They are intended to
peruse information like record numbers,
clients distinguishing proof, article numbers,
and measures of cash and so on. The paper
designs are obliged with a set number of
fixed lines to peruse per record.
4. Traffic-sign recognition: It is a
mechanism where the vehicle automatically
detects the traffic signs which can be
classified by the shape, color etc. The signs
can be schools, accident prone zone, speed
limit etc. In this, the camera which is present
in the car scans the signs in the road and it is
showed in the LCD screen warning the
driver about the sign. Here there are two
steps: 1. Localization and 2. Recognition.
This has been a compulsory attachment to
cars in Europe from May 2022.
5. Banking: The employments of
optical character recognition fluctuate over
various areas. In banking one realizes its
application, where optical character
recognition uses automation technologies
and for this the best example is ATM. And
another example in banking which we can
think is handle cheque where a handwritten
cheque is scanned and verified. A register
can be embedded with a machine, the
composition on it is filtered immediately,
and the right measure of cash is moved. This
𝑖𝑛𝑛𝑜𝑣𝑎𝑡𝑖𝑜𝑛44,45
has almost been idealized
for printed cheque, and is originally exact
for transcribed cheque too; however it
periodically requires manual affirmation.
Generally speaking, this decreases hold up
times in numerous banks.
6. HealthCare: It Healthcare has
likewise observed a growth in the
deployment of Optical character recognition
alteration to handle administrative work. A
medical care expert consistently requires
taking care of huge bulk of structures for
every victim, involving defense frames just
as broad wellbeing structures. To stay aware
of the entirety of this information, it is
helpful to enter pertinent data into a gadget
information base that can be gotten to as
vital. Structure 𝑝𝑟𝑒𝑝𝑎𝑟𝑖𝑛𝑔44,45
devices,
controlled by OCR, can isolate information
from structures and place it into information
bases, thus every sufferer's data is
immediately registered. Therefore, medical
care suppliers can zero in on conveying the
most ideal support of each patient.
7. Assistance for visually impaired:
Back in days, previous to the advanced PCs
and the requirement towards contribution, a
lot of information rose, this was the
envisioned zone of utilization for
understanding machines. Joined with a

12. discourse union framework such a reader
would empower the oblivious to
comprehend printed archives. With now
technology, it can scan, read and recognize
the text. We can also convert the scanned
text into voiced speech which is an
advantage for the blind people.
6. Conclusion
In this paper, we have discussed the
history of OCR and the various phases like
image acquisition, Binarization, slant
correction, noise removal, smoothing,
normalization, preprocessing, feature
extraction, selection. Each phase is
discussed in detail. Recent achievements are
also discussed here. Disregarding the
immense proportion of assessment in
Optical character recognition, affirmation of
characters for language, for instance, Arabic,
Sindhi Urdu despite everything stays an
open test. A review of Optical character
recognition methods for these tongues has
been organized as an upcoming work.
Another significant zone of research is
multi-lingual character acknowledgment
framework. At last, the work of OCR
frameworks in viable applications stays a
functioning are of exploration.
7. Acknowledgement
This work is supported by the
department of specialization, Jain
University, Bangalore. The authors would
like to thank all the staffs for their support in
completion of this work.
8. References:
[1]. J-C. Simon, 1992, Off-Line Cursive
Word Recognition, Proceedings of the
IEEE, 80, 1150-1161
[2]. A. W. Senior, 1994, Off-Line Cursive
Handwriting Recognition Using NNs, PhD
Dissertation, University of Cambridge,
England.
[3]. C.Y. Suen et al., 1993, Building a New
Generation of Handwriting Recognition
Systems, Pattern Recognition Letters, 14,
303-315.
[4]. S.N. Srihari, 1992, High-Performance
Reading Machines, Proceedings of the
IEEE, 80, 1120-1132.
[5]. Talaat, Ahmed & Suen, Cheng. (2014).
A Novel Method for the Recognition of
Isolated Handwritten Arabic Characters.
[6]. M. Cheriet, Visual recognition of Arabic
handwriting: challenges and new directions.
In Arabic and Chinese Handwriting
Recognition, Lecture Notes in Computer
Science, vol. 4768, Springer, 2008, pp 1-21.
[7]. Mori S, Suen CY, Yamamoto K.
Historical review of OCR research and
development. Proc IEEE. 1992;80:1029–
1057. Doi: 10.1109/5.156468.
[8].Govindan K, Shivaprasad AP. Character
recognition—a review. Pattern
Recogn. 1990;23(7):671–683. Doi:
10.1016/0031-3203(90)90091-X.
[9]. Mehfuz S, Katiyar G. Intelligent system
for off-line handwritten character
recognition: a review. Int J Emerg Technol
Adv Eng. 2012;2:538–543.

13. [10]. Katiyar G, Mehfuz S. Evolutionary
computing techniques in off-line
handwritten character recognition: a
review. UACEE Int J Computer Science
Appl. 2012;1:133–137.
[11]. Plamondon R, Srihari SN. On-line and
off-line handwriting recognition: a
comprehensive survey. IEEE Trans Pattern
Anal Mach Intell. 2000;22(1):63–84. Doi:
10.1109/34.824821.
[12]. Gauri Katiyar and Shabana Mehfuz A
hybrid recognition system for off-line
handwritten characters.
[13]. G. Tauschek, “Reading machine,” U.S.
Patent 2 026 329, Dec. 1935.
[14]. P. W. Handel, “Statistical machine,”
US. Patent 1915 993, June 1933
[15]. S. N. Srihari and S. W. Lam, 1995,
Character Recognition, Technical Report,
CEDAR-TR-95-1.
[16]. V.K. Govindan, 1990, Character
Recognition - A Review, Pattern
Recognition, 23, 671-683
[17]. S. Impevedo et al., 1991, Optical
Recognition - A Survey, International
Journal of PR & AI, 5, 1-24.
[18]. J.Mantas, An overview of character
recognition methodologies, pattern
recognition 19, 425-430 (1986).
[19]. Character Recognition. British
Computer Society, London, England (1971).
[20]. K.S.Fu, Syntactic Pattern Recognition
and Applications. Prentice Hall, Engiewood
Cliffs, New Jersey (1982).
[21]. G.Naggy, Optical character
recognition: theory and practice, Handbooks
of statistics, P.R.Kilshnaiah and L.N.Kanal,
Eds, Vol. 2, pp. 621-649 (1982).
[22]. "Optical character recognition -
History". ABBYY Technology. Retrieved 18
September 2016.
[23]. “The First OCR System: ‘GISMO’
(1951): HistoryofInformation.com”.
www.historyofinformation.com. Retrieved
2016-09-17.
[24]. Vamvakas, G. & Gatos, B. &
Stamatopoulos, Nikolaos & Perantonis,
Stavros. (2008). A Complete Optical
Character Recognition Methodology for
Historical Documents. Document Analysis
Systems, IAPR International Workshop on.
525-532. 10.1109/DAS.2008.73.
[25]. S-B. Cho, 1997, Neural-Network
Classifiers for Recognizing Totally
Unconstrained Handwritten Numerals, IEEE
Trans. on Neural Networks, 8, 43-53.
[26]. S-W. Lee, 1996, Off-Line Recognition
of Totally Unconstrained Handwritten
Numerals Using MCNN, PAMI, 18, 648-
652.
[27]. S-W. Lee, 1995, Multilayer Cluster
Neural Network for Totally Unconstrained
Handwritten Numeral Recognition, NNs, 8,
783-792.
[28]. J. Yang, P. Ren and X. Kong,
"Handwriting Text Recognition Based on
Faster R-CNN," 2019 Chinese Automation
Congress (CAC), Hangzhou, China, 2019,
pp. 2450-2454, doi:
10.1109/CAC48633.2019.8997382.

14. [29]. A. T. Sahlol, M. Abd Elaziz, M. A. A.
Al-Qaness and S. Kim, "Handwritten Arabic
Optical Character Recognition Approach
Based on Hybrid Whale Optimization
Algorithm With Neighborhood Rough Set,"
in IEEE Access, vol. 8, pp. 23011-23021,
2020, doi: 10.1109/ACCESS.2020.2970438.
[30]. T. T. Zin, S. Z. Maw and P. Tin, "OCR
Perspectives in Mobile Teaching and
Learning for Early School Years in Basic
Education," 2019 IEEE 1st Global
Conference on Life Sciences and
Technologies (LifeTech), Osaka, Japan,
2019, pp. 173-174, doi:
10.1109/LifeTech.2019.8883978.
[31]. Z. Huang and Q. Zhang, "Skew
Correction of Handwritten Chinese
Character Based on ResNet," 2019
International Conference on High
Performance Big Data and Intelligent
Systems (HPBD&IS), Shenzhen, China,
2019, pp. 223-227, doi:
10.1109/HPBDIS.2019.8735469.
[32]. B. Dessai and A. Patil, "A Deep
Learning Approach for Optical Character
Recognition of Handwritten Devangiri
Script," 2019 2nd International Conference
on Intelligent Computing, Instrumentation
and Control Technologies (ICICICT),
Kannur,Kerala, India, 2019, pp. 1160-1165,
doi: 10.1109/ICICICT46008.2019.8993342.
[33]. R. K. Gerlach, “Wide-tolerance optical
character recognition for existing printing
mechanisms,” in Optical Character
Recognition, G. L. Fischer et al., Eds.
McGregor & Wemer, 1962, pp. 93-114.
[34]. C. C. Heasly, Jr, and G. L. Fischer, Jr,
“Some elements of optical scanning,” in
Optical Character Recognition, G. L Fischer
et al., Eds, McGregor & Wemer, 1962, pp.
15-26.
[35]. E. C. Greanias, “Some important
factors in the practical utilization of optical
character readers,” in Optical Character
Recognition, G. L. Fischer et al., Eds.
McGregor & Wemer, 1962 pp, 129-146.
[36]. Hanmandlu M, Mohan KRM,
Chakraborty S, Goyal S, Choudhury DR
(2003) Unconstrained handwritten character
recognition based on fuzzy logic. Pattern
Recogn 36:603–623
[37]. Hanmandlu M, Murthy OVR (2007)
Fuzzy model based recognition of
handwritten numerals. Pattern Recogn
40:1840–1854
[38]. Hládek, D., Staš, J., Ondáš, S. et
al. Learning string distance with smoothing
for OCR spelling correction. Multimed
Tools Appl 76, 24549–24567 (2017).
https://doi.org/10.1007/s11042-016-4185-5
[39]. S. Lamba, S. Gupta and N. Soni,
"Handwriting Recognition System- A
Review," 2019 International Conference on
Computing, Communication, and Intelligent
Systems (ICCCIS), Greater Noida, India,
2019, pp. 46-50, doi:
10.1109/ICCCIS48478.2019.8974547.
[40]. Trier OD, Jain AK, Taxt J (1996)
feature extraction methods for character
recognition: a survey. Pattern Recogn
29(4):641–662

15. [41]. Arif, Muhammad & Hassan, Haswadi
& Nasien, Dewi & Haron, Habibollah.
(2015). A Review on Feature Extraction and
Feature Selection for Handwritten Character
Recognition. International Journal of
Advanced Computer Science and
Applications. 6.
10.14569/IJACSA.2015.060230.
[42]. S.L. Chang, T. Taiwan , L.S. Chen,
Y.C. Chung, S.W. Chen, ―Automatic
license plate recognition‖, IEEE
Transactions on Intelligent Transportation
Systems, Vol: 5 , Issue: 1, p.p. 42 – 53,
2004.
[44].
http://www.cvisiontech.com/reference/gener
al-information/ocr-applications.html
[45]. M.D. Ganis, C.L. Wilson, J.L. Blue,
―Neural network-based systems for
handprint OCR applications‖, IEEE
Transactions on Image Processing, Vol: 7,
Issue: 8, p.p. 1097 – 1112, 1998