Image Processing Methods OCR, ICR AND OMR

Arya MirΛογισμικό & κατασκευή λογ/κού

6 Ιουλ 2011 (πριν από 6 χρόνια και 1 μήνα)

3.463 εμφανίσεις

The image processing is a branch of computer science in which the digital signal, representing a digital image taken by a digital camera, or generated by a digital scanner, is processed.






Arya Mirdjalali
Department of Electronics and Computer Science
University of Southampton


The image processing is a branch of computer
science in which the digital signal, representing a
digital image taken by a digital camera, or
generated by a digital scanner, is processed.

In most of image processing techniques the source
image is considered as a two-dimensional signal.
Therefore the processing is a step toward applying
signal processing methods which involve
processing, interpretation and amplification of the

There are three common processing methods to
extract information from images for further machine
processing. ICR stands for Intelligent Character
Recognition, OCR stands for Optical Character
Recognition and OMR stands for Optical Mark
Recognition. Despite the underlying similarities of
these technologies there exist various differences
between them which make them suitable for
different tasks and different applications.

Digital Image Processing- Signal Processing - OMR
1. I

Image processing has been one of the fast growing
fields in Computer Science in resent years. The
increased interest in Computer vision, robotics,
neural networks, Image Morphology, Image
compression, Knowledge-based Image Analysis
systems, which all involve image processing
knowledge and techniques proof the importance
and growth of this interesting field.

In general, all applications involving image
processing can be sub categorized under two
categories based on their main goals. Those
concerns with development and extraction of
information for human interpretation and
understating, and those concerns with generating
data for further machine interpretation.

Any image processing, usually involves three steps.
The first step is to import the image using an optical
scanner or directly through a digital camera. The
second step concerns with image manipulation
including enhancement, noise reduction and
compression of data. The final step is to generate
an appropriate form of output, which could be
simply a modified version of the source image or a
sequence of character representing some data
extracted from the processed image. [1]

In this paper, three methods which are widely used
in processing images in order to extract data and
information for the purpose of further machine
interpretation will be studied. These three
techniques are:
• OMR: Optical Mark Recognition
• OCR: Optical Character Recognition
• ICR: Intelligent Character Recognition

In Section 2 of this paper some general aspects of
digital images and how they are represented by
computers will be studied. The section will continue
by discussing the basic steps of image processing
and finish by highlighting some key components of
every image processing system.
Permission to make digital or hard copies of all or part of this
work for personal or classroom use is granted without fee
provided that copies are not made or distributed for profit or
commercial advantage and that copies bear this notice and the
full citation on the first page. To copy otherwise, to republish, to
post on servers or to redistribute to lists, requires prior specific
Annual Multimedia Systems, Electronics and Computer
Science, University of Southampton
© 2007 Electronics and Computer Science, University of

Section 3, 4 and 5 will respectively focus on OCR,
ICR and OMR and the underlying principal of each
and technical details involved in each of these

The paper will continue in Section 6 by discussing
the differences between the three techniques and
comparing them against each other. In section 7,
usage of image processing in future industry is
discussed by a real world example.

2. B


The history of image processing goes back to early
1920s when the need for improving the quality of
the newspaper images transferred through a sub
sea cable between New York and London became
a major problem not only in terms of actual quality
of the images but also in terms of the transmission
time. At that time increasing the quality by a certain
degree and decreasing the transmission time
across Atlantic Ocean from more than a week to
only 3 hours was a great achievement. [2]

Now after over 80 years by enhancement in
technology and available resources such as huge
memories or super computers which are capable of
fast processing of huge amount of information, the
expectation and the goals of image processing has
moved to a higher level.

2.1 Digital Image
A digital image is a finite set of digital values
representing a two dimensional image. These digital
values or pixels, are stored in computer memory as
a two dimensional array of integers. Each pixel
represents a specific coordinate in a two
dimensional region on the image and contains some
information about that position. Digital images can
be classified according to the type of this
information. For example a colour image is an
image in which each pixel holds the information
about the colour of the point it represents. [3]

2.2 Display of Digital Image
As it been mentioned at the beginning of this
section an image represents a two-dimension
illumination magnitude function f (x, y), where x and
y are location Coordination and in any point (x, y)
the value of function f is proportional to the
illumination or the Grey level of that point. [4]

Figure 1 shows use of conventional axis in
displaying digital images.
Origin Y

Figure 1: Use of Conventional axis in displaying
digital images

2.3 Image Processing Fundamentals
The first step toward processing an image to extract
the information is called Image Acquisition which
means acquiring the digital image either directly
from a digital camera or by using an optical scanner
or any other photography device capable of
digitalization of the output signal.

Pre-processing is the second step. The main task in
pre-processing is using different techniques to
improve the quality of the image which increase the
chance of success in later processing activities.
Pre-processing normally involves techniques to
adjust the contrast level, eliminate the existing
noises and adjust the minor rotations. Rotations are
usually fixed by taking two points which are
supposed to be on a horizontal line and find the
slope of the line between them. And then this slop
will be considered in all arithmetic and calculations.

The next step is called Segmentation. This is the
process of dividing the input image into its
comprising segments or components. Segmentation
is regarded as one of the most difficult tasks of
image processing. The use of powerful and
appropriate segmentation algorithms in this phase
considerably increase the chance of overall
success. And weak or erroneous algorithms almost
always lead to dramatic failure.

The output of segmentation phase is normally rough
pixel data which an area border or whole points
inside it are recognized. This data should be
formatted in an appropriate and proper way for
computer processing. Hence one of the first

decisions to make is whether the data should be
displayed as a border or a whole area.

The border display is usually used where an
external specification of the figure such as curves
and corners are required. Whereas area display is
preferred when some internal features and
specification of the image segments such as image
texture are required. [6]

There are four well known forms of segmentation:

• Pixel Based Segmentation
• Model Based Segmentation
• Multi scale Segmentation
• Semi automatic Segmentation

All are widely used in different applications. [7] But
in general there are two basic classes of models
which are used to determine if an image satisfy
some assumptions which is required in order for a
particular technique to be applicable to it. Statistical
models, that describe the pixel population in an
image or region and Spatial models that describe
the decomposition of an image into regions. [8]

Description which is also called feature selection is
concerns with extraction of features which gives
some of the required quantitative information, or is
fundamental to recognize something from other.

The final step consists of recognition and
interpretation. Recognition is a process which,
based on the obtained information from descriptors,
assigns a label to an object. Interpretation consists
of definition assignment to a set of recognized

Knowledge and information about problem is kept in
a database inside the image processing system.
This knowledge may be simply some information
about the location of the areas containing details of
interest, or might contains a more complex set of
information. [9]

Figure 2, shows the interaction between these
different steps. The double arrows in Figure 2
represent the constant interaction between
knowledge database and different steps of the

Figure 2: Image processing life cycle

2.4 Components of the System
In general the basic components of every image
processing system are:

• Image Acquisition device:
Digital camera
Optical Scanner

• Storage unite:
Optical Disc
Video Tape
Magnetic Disc
Magnetic Tape

• Processing device:
PC computer

• Communication and Display:
Television Display





Display &

Photography film
Considering the fact that, an 8-bit image of 1024 x
1024 pixels requires one million bytes of memory, it
is very important to provide sufficient memory and
resources for any system dealing and processing
images. [4]

There are usually three groups of memories used
during an image processing procedure:

• Short term memory which is used during
• Working memory for fast rereading of data.
• Archive memory which is frequently low
access memory.

Some image processing systems use the computer
memory for short term storage, whereas some other
systems use a specially designed board which is
called Frame Buffer. Frame buffers are capable of
storing multiple images and can be accessed in a
very high rate.

3. O
Optical Character recognition (OCR) is the
technology which makes it possible for scanned
pictures or pictures created with typing to be
converted to machine characters and stored in a
computer. This method is based on pixel processing
and scanning image pixels.

3.1 OCR Processing Steps
The first step in OCR processing is Information
Scanning. The speed of this process depends on
various elements such as the quality of the scanner,
the number of output colour, paper quality,
cleanliness, weights and character recognition
system appearance.

The next step is Image Recognition in which images
are translated. Accuracy of OCR method is
determined by threshold value and amount of

The final step is called Review. In this phase the
value of translated image is compared with the
value of real image.

3.2 OCR Processing Methods
Basically there are two methods which are widely
used in OCR Systems.

• Matrix matching
• Feature Extraction

Matrix matching is the simplest and more common
method used in OCR systems. It compares the
character output of OCR with a library of character
and templates. When a matching candidate with a
predefined level of similarity is found the character
image will be replaced by the ASCII code of the
matching entry.

Feature Extraction is a more flexible method which
is used in Intelligent Character Recognition. This
method will be explained in the next Section of this

3.3 Benefits of OCR
There are different benefits in using OCR which
makes it more suitable than other methods of
entering data.
Reducing the amount of data entry errors,
strengthening data entry, handling peak loads,
human readability, compatibility with most of printing
techniques and scanning correction, are among the
most important benefits of OCR.

4. I
This technology is involved in providing machine or
software with the ability to read images of printed
characters or non cursive handwritings and store
them in the computer by converting them into ASCII

The image of handwriting characters is the result of
a pre-processed image which is processed pixel by
pixel and point to point, such as reading received
frames through fax. Today this method has a great
application in market.

4.1 ICR Self Learning Ability
ICR is in fact a more advanced version of OCR with
the extended ability to learn different font and styles
of handwriting during processing to improve the
performance and accuracy of recognition.

ICR systems usually have an automatic self
learning procedure, which updates the recognition
knowledge base database for new handwriting
patterns and styles.

4.2 ICR Processing Method
In ICR systems, Feature Extraction method is used
instead of Matrix Matching. This method has not got
the restriction of matching with some predefined

There are different variations of this method
according to the amount of computer intelligence
used. In this method computer or processing unite
looks for some features such as open areas, closed
shapes, diagonal lines and intersection of lines.

4.3 OCR / ICR Usage
Now a day, archiving is one of the increasing topics
in academic, industry and governmental

Providing online access to publications and books
through online libraries or demand for fast access to
the information and data through keeping huge
databases, are some of the fast growing issues
which lead to an increase need for more
sophisticated and advanced automatic data entry

“…Google is pumping $200m (£110m)
into creating a digital archive of millions of
books from four top US libraries - the
libraries of Stanford, Michigan and
Harvard universities, and of the New York
Public Library - by 2015. It is also
digitizing out-of-copyright books from the
UK's Oxford University… “

[10] [BBC NEW –published on

5. O
Optical Mark Recognition is a method of capturing
information by comparing reflectivity at some
predefined coordination on an image.

This technology is used for finding and recognising
a mark on an image. Based on predefined forms, an
OMR system is capable of storing and retrieving
information from the form and analysing them
This is a method which is used widely in industry
and universities. Usually this method is combined
with OCR, ICR and other processing method to
provide a powerful tool for processing forms.

OMR is the fastest and most reliable method of
correcting multiple choice questions. It is also used
for large scale statistical analysis. It is a powerful
automatic method for retrieving and storing of

Using an OMR based system usually starts with
designing a form and then providing a piece of
software to read and recognise the form. it is
followed by scanning the forms and storing data in a
field. And finish with analysing the retrieved data

Some of the OMR based forms which can be used
in OMR systems are:

• Multiple choice questions (MCQ)
• Multiple response questions (MRQ)
• True/False
• Matching
• Rank order method (ROM)
• Numerical
• Visual identification

5.1 Components of OMR System
The basic components of an OMR system are:

• Scanner capable of fast scanning of forms.
• Standard answer sheets.
• Software capable of converting and
processing image to data.
• Printer.

5.2 Benefits of OMR
Some of the key benefits of OMR system which
make it suitable for a wide range of activities are:

• Capability of controlling and modifying
• Cost reduction.
• Easy to use in office/home.

• Fast rough data processing.
• Obtaining results and analyzing data are
fast and easy.
• Easy back up and support through the
• Fast and easy installation.
• No need for skill to use.
• Reduction of human resources.
• Portability of software in comparison with
large punched card reader machines.

6. C
Although the fundamental principal and underlying
technology of OMR, OCR and ICR are very identical
but in fact there are some differences between each
of these methods which make each suitable for use
in certain situation and application.

OCR and ICR are regarded as recognition engines
which can work on images. But, OMR is regarded
as a technology for recognition of marks without the
need of any recognition engine.

Therefore Optical Mark Recognition can not be
used for recognising printed or hand written
characters, whereas OCR is capable of reading
printed and handwritten characters.

In OMR there is a need for a special form and
special method for mark reading. This special form
has a number of boxes with the same size on sides
and top. The rulers are used to guide the system to
where it can find the information of interest.

In OMR there is no need for scanning or sorting of
information images.

Usually accuracy and performance of OMR systems
are more than OCR based systems as long as a
well designed form and system be used.

In OCR/ICR, very flexible forms can be used and
there is no need for accurate recognition of boxes.
And basically the image could be floating.

Colour is a very important factor in OMR systems,
since using an omitted colour causes the output of
the scanner to reduce and the accuracy to increase.

Usually OCR/ICR systems are used where there is
a need for storing information. As in OMR machines
images are scanned, sorted, and written as optical

Hand writing Recognition

Machine print Recognition

Mark Recognition

Requiring special mark

Requiring registered marks

Electronic store and retrieve

Table 1: Comparison between OMR and
OCR/ICR methods

7. U

One of the current undergoing image processing
projects is dealing with conversion of paper
cheques to digital copies which will revolutionise the
future of cheques processing industry and banking.

This process will reduce the cost by early
conversion of paper cheques to electronic copies
which saves the cost of transportation and the need
of installation of reader, sorter and encoding

There are two types of electronic cheques:

• Cheque Truncation: in which a digital image
from front and back of the cheque is
combined with an electronic message
containing the information on the cheque.
• Cheque Conversion: which replace the
cheque with an Accounts Receivable (ARC)
or Point of Purchase (POP) transaction. In
ARC a digital image from front of the
cheque is retained for two years. Whereas
in POP the original cheque is given back to
the consumer and there is no need for an
image to be kept.

The digital images need to be processed with very
high accuracy and archived for later referencing.
There is also an obvious need for a very reliable

mechanism for finding and sorting identical images.
And a way to digitally sign the legally created
images to eliminate the chance of fraud.

8. C

In this paper the fundamental steps of every image
processing has been discussed in detail and some
solutions for common problems involved with each
of these steps has been described. And it became
clear that there is still room for improvement and
development of more advanced technologies for
pre-processing and improving the quality of images
and reconstruction of digital images.

The underlying principal behind three widely used
image processing algorithms and technology, OMR,
OCR and ICR has been covered. We came to the
conclusion that OCR and ICR are regarded as
recognition engines and ICR has the self learning
ability which makes it a powerful tool for recognising
new patterns and hand writings.

Although a large amount of work and effort have
been done toward converting the Latin handwriting
into machine characters. But there is still a demand
for engines and machines capable of processing
and recognising other scripting such as Chinese or

9. R

[1] J. Mantas, C. Daskalakis, A.G. Heaton, Feature
Generation of handwritten Characters by a
shape analysis method. The University of
Manchester Institute of Science and
Technology. International Conference on
Electronic Image Processing 25-28 July 1982.

[2] Gonzalez R. C. and Woods R. E Digital Image
Processing, 7th Edition. Prentice Hall (2004).

[3] Image Processing Fundamentals,
Last accessed November

[4] G. Auber and P Kornprobst, mathematical
problems in image processing. Springer (2002).

[5] Tech Vision, Intelligence in document Imaging.
Last accessed November 2006

[6] Robert Kermens, Image Processing System
Applications, Rochester Institute for
Technology, published May 2001

[7] Fisher J.L, Hinds S.C, Dapos Amato, D.P. Mitre
Corp, McLean, VA. A rule-based System for
document image segmentation. Published Jun

[8] Azriel Rosenfeld, Larry S. Davis, Image
Segmentation and Image Models. IEEE, V.67,
Number 5, May 1979

[9] Ian T. Young, Jan J. Gerbrands, Lucas J. Van
Vliet, Fundamentals of Image Processing,

[10] BBC News article, Google restarts online books
plan, published on 2005/11/01,
Last accessed October 2006