You are on page 1of 4

Application of Artificial Neural Network Model for Optical, Character

Recognition

Nallasamy Mani’ and Bala Srinivasan’


‘Department of Electrical and Computer Systems Engineering
2
Department of Computer Technology
Monash University, Caulfield Campus
Victoria 3145, Australia

ABSTRACT system. Several algorithms for character


recognition have been developed based on
Many artificial neural network models feature selection [2,3]. Some of them have
(ANN’S)have been proposed to mimic the been found commercialliy viable and have
human brain in solving problems involving gone into production like OmniPage,
human-like intelligence. An application of Wordscan, TypeReader etc [l]. The
artificial neural network approach for optical performance of the systems have been
character recognition (OCR) is discussed in constrained by the dependence on font, size
this paper. We examine a simple pattern- and orientation. The reco’gnition rate in these
recognition system using artificial neural algorithms depends on the choice of features.
network to simulate character recognition. A Most of the existing algorithms involve
simple feed-forward neural network model extensive processing on the image before the
has been trained with different set of noisy features are extracted that results in increased
data. The back-propagation method is used computational time.
for learning in neural network.
In this paper, we discuss a neural network
1. INTRODUCTION based method for character recognition that
would effectively reduces the image
Optical character recognition (OCR) is a processing time while miaintaining efficiency
process of converting a printed document or and versatility. We also discuss an
scanned page into ASCII characters that a enhancement to the previous approach for
computer can recognise. Computer systems character recognition. The parallel
equipped with such an OCR system improve computational capabilities of neural networks
the speed of input operation, decrease some ensures a high speed of recognition which is
possible human errors and enable compact critical to a commercial environment. Neural
storage, fast retrieval and other file network approach have been used for
manipulations. The range of applications character recognition [ 1,473, but a complete
include postal code recognition, automatic system which encompasses all the features of
data entry into large administrative systems, a practical OCR system is yet to be realised.
banking, automatic cartography and reading The key factors involved in the
devices for blind. implementation are: an optimal selection of
features which categorically defines the
Accuracy, flexibility and speed are the main details of the characters, the number of
features that characterise a good OCR features and a low image processing time.

0-7803-4053-1/97/$10.00 @ 1997 IEEE 251 7


line segments and smeared images. A pre-
processor is used to smooth the digitised
2. NEURAL NETWORK characters. Moreover, the system must be
ARCHITECTUFU3 able to handle touching characters,
proportional spacing, variable line spacing
The architecture of a neural network and change of font style in the scanned text,
determines how a neural network transforms in addition to the problems of multi-fonts.
its input into an output. This transformation
can be viewed as a computation. We have image acquisition J
implemented a multi-layer feed forward
neural network with one hidden layer as
shown in figure 1. mage pre-processing
I
hidden
layer
[ feature extraction ]
n
neural networks
I I I

input features
BP classification

.
BP traning I[ results 1
Figure 2 System Block Diagram

weight 3.1. Feature Extraction


- connection
Feature extraction is the process of getting
Figure 1 The Network Model information about an object or a group of
object in order to facilitate classification. This
is an important part in our system. The
3. OCR SYSTEM DESIGN character from the scanned image is
normalised from 60 X 60 pixel into 32 X 32
The main functional modules in our OCR pixel as in figure 3.
systems are: image acquisition module, image 60
Pre-processing module, feature extraction

a
module and neural network module. The r--7
block diagram -in Figure 2 shows these
modules. 6o

The main task of image acquisition module is U


to obtain text image from a scanner. It is 32
called ‘image’ because scanner inherently n
scans pixel of the text and not characters. The
input file format is PCX. When patterns are
scannedand digitised, the data may carry
I
U
a132

some unwanted noise. For example, a scanner


with low resolution may produce touching Figure 3

2518
The horizontal and vertical vectors (Vh and
Vv respectively) are added together to form
the input vector of the neural network.
Finaily, an input vector that contains 64
(horizontal + vertical) unique features of the
character is evaluated. A histogram is the
distribution of the pixel intensity values of an
image or portion of an image. It indicates the
overall brightness and contrast of an image.
Histogram techniques are used for automatic
processing of lines, words and characters
extraction in the sequence.

The erosion and dilation operations make the


object smaller and larger respectively. The advantage is to train the network with
Erosion makes an object smaller by removing user defined character sets, numerals and
or eroding away the pixel on its edges. even with other languages. Once the network
Dilation makes an object larger by adding is trained it would create: an associated weight
pixel around its edges. Dilation technique is of the particular training file. The systems has
used for extracting a word from the original been implemented usiing C++ and Turbo
image (gray scale). Image dilation is applied vision. The performance of the system is
to make the characters in a word thicker until reported.
they join together. The image erosion Learning rate = 0.45
techniques are used for extracting each
character from a word. -7
Error tolerance

0.07500
Iteration

4. ENHANCED MODEL 0.05000


0.02500
0.01000
An enhancement to our OCR system is an in- 0.00750
built training pattern editor and a better 0.00500 72
graphics user interface. The pattern editor is
very useful in creating the training data files. i"."""'"rI
0.00050

- Screen of Parrern Manoger


I 1
Input Pattern :Pattern 3 of 5
Learning rate = 0.90
"-1 . - I

0.02500
A m w Keys
Entn :
0.01 000
n: Crcatc noisy panem. 0.00750
q. Qm editing &is plrem.
0.00500

399
I530

0.0001

251 9
5. REFERENCES

[l]. A. Rajavelu, M.T. Musavi, and


Shirvaikar, M. V, “A Neural Network
Approach to Character Recognition”, Neural
Networks, V01.2, 1989, pp.387-393.
[2]. S. T. Kahan, T. Pavlidis, and W. Baird,
“On recognition of printed characters of any
font and size”, IEEE Transactions of Pattern
Recognition and Machine Intelligence,
PAMI-9, 1987, pp.274-285.
[3]. B. Hussain, and M. R. Kabuka, “A novel
feature recognition neural network and its
application to character recognition”, IEEE
Transactions of Pattern Recognition and
Machine Intelligence, Vol. 16, No. 1, 1994,
pp.98 - 106.
[4]. H.I. Avi-Itzhak, T.A. Diep, and H.
Garland, “High accuracy optical character
recognition using neural networks with
centroid dithering”, IEEE Transactions of
Pattem Recognition and Machine
Intelligence, Vol. 17, No.2, 1995, pp.218-
224.
[SI. D.E. Rumelhart, G.E. Hinton, R.J.
4. CONCLUSION Williams, ‘‘Learning Representation by Error
Backpropagation”, In Parallel Distributed
We have shown that artificial neural network Processing, Vol. 1, MIT Press, Cambridge,
can be implemented successfully in optical Chapter.8, 1986, pp.318-362.
character recognition. The system has image [6]. W.P. Jones, and J. Hoskins, “Back
pre- and post processing modules for text Propagation: A generalised learning rule”,
image. The experiment result shows Byte, 12, 1987, pp.155-158.
recognition rate is 70% for noisy data to upto [7]. A.K. Jain, J. Mao, and K.M. Mohiuddin,
99% . Further work is initiated for multiple “ Artificial Neural Networks: A Tutorial”,
font and size characters and hand written IEEE Computer, Vo1.29, No.3, 1996, pp.31-
character recognition. 44.

2520

You might also like