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 Many artificial neural network models (ANN’S)have been proposed to mimic the human brain in solving problems involving human-like intelligence. An application of artificial neural network approach for optical character recognition (OCR) is discussed in this paper. We examine a simple patternrecognition system using artificial neural network to simulate character recognition. A simple feed-forward neural network model has been trained with different set of noisy data. The back-propagation method is used for learning in neural network.
1. INTRODUCTION Optical character recognition (OCR) is a process of converting a printed document or scanned page into ASCII characters that a computer can recognise. Computer systems equipped with such an OCR system improve the speed of input operation, decrease some possible human errors and enable compact storage, fast retrieval and other file manipulations. The range of applications include postal code recognition, automatic data entry into large administrative systems, banking, automatic cartography and reading devices for blind. Accuracy, flexibility and speed are the main features that characterise a good OCR
0-7803-4053-1/97/$10.00
@
1997 IEEE
system. Several algorithms for character recognition have been developed based on feature selection [2,3]. Some of them have been found commercialliy viable and have gone into production like OmniPage, The Wordscan, TypeReader etc [l]. performance of the systems have been constrained by the dependence on font, size and orientation. The reco’gnition rate in these algorithms depends on the choice of features. Most of the existing algorithms involve extensive processing on the image before the features are extracted that results in increased computational time.
In this paper, we discuss a neural network based method for character recognition that would effectively reduces the image processing time while miaintaining efficiency and versatility. We also discuss an enhancement to the previous approach for character recognition. The parallel computational capabilities of neural networks ensures a high speed of recognition which is critical to a commercial environment. Neural network approach have been used for character recognition [ 1,473, but a complete system which encompasses all the features of a practical OCR system is yet to be realised. The key factors involved in the implementation are: an optimal selection of features which categorically defines the details of the characters, the number of features and a low image processing time.
251 7
2. NEURAL NETWORK ARCHITECTUFU3 The architecture of a neural network determines how a neural network transforms its input into an output. This transformation can be viewed as a computation. We have implemented a multi-layer feed forward neural network with one hidden layer as shown in figure 1.
n
line segments and smeared images. A preprocessor is used to smooth the digitised characters. Moreover, the system must be able to handle touching characters, proportional spacing, variable line spacing and change of font style in the scanned text, in addition to the problems of multi-fonts. image acquisition J mage pre-processing
I
[ feature extraction ]
hidden layer
neural networks I
I
I
input features
BP classification
I[
BP traning
results
.
1
Figure 2 System Block Diagram
-
weight connection
Figure 1 The Network Model
3. OCR SYSTEM DESIGN The main functional modules in our OCR systems are: image acquisition module, image Pre-processing module, feature extraction module and neural network module. The block diagram -in Figure 2 shows these modules. The main task of image acquisition module is to obtain text image from a scanner. It is called ‘image’ because scanner inherently scans pixel of the text and not characters. The input file format is PCX. When patterns are scannedand digitised, the data may carry some unwanted noise. For example, a scanner with low resolution may produce touching
3.1. Feature Extraction Feature extraction is the process of getting information about an object or a group of object in order to facilitate classification. This is an important part in our system. The character from the scanned image is normalised from 60 X 60 pixel into 32 X 32 pixel as in figure 3.
60
a
r--7
6o
U 32
n
I
a 1 3 2
U
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. Erosion makes an object smaller by removing or eroding away the pixel on its edges. Dilation makes an object larger by adding pixel around its edges. Dilation technique is used for extracting a word from the original image (gray scale). Image dilation is applied to make the characters in a word thicker until they join together. The image erosion techniques are used for extracting each character from a word.
The advantage is to train the network with user defined character sets, numerals and even with other languages. Once the network is trained it would create: an associated weight of the particular training file. The systems has been implemented usiing C++ and Turbo vision. The performance of the system is reported. Learning rate = 0.45
7 Error tolerance
Iteration
0.07500
4. ENHANCED MODEL
0.05000 0.02500 0.01000 0.00750
An enhancement to our OCR system is an inbuilt training pattern editor and a better graphics user interface. The pattern editor is very useful in creating the training data files.
0.00500
72
i"."""'"rI 0.00050
- Screen of Parrern Manoger
1
I Input Pattern :Pattern 3 of 5 "-1
Learning rate = 0.90
. - I
0.02500 0.01 000 0.00750 0.00500
A m w Keys Entn :
n: q.
Crcatc noisy panem. Qm editing &is plrem.
399
I530 0.0001
251 9
5. REFERENCES
4. CONCLUSION We have shown that artificial neural network can be implemented successfully in optical character recognition. The system has image pre- and post processing modules for text image. The experiment result shows recognition rate is 70% for noisy data to upto 99% . Further work is initiated for multiple font and size characters and hand written character recognition.
[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.218224. [SI. D.E. Rumelhart, G.E. Hinton, R.J. Williams, ‘‘Learning Representation by Error Backpropagation”, In Parallel Distributed Processing, Vol. 1, MIT Press, Cambridge, Chapter.8, 1986, pp.318-362. [6]. W.P. Jones, and J. Hoskins, “Back Propagation: A generalised learning rule”, Byte, 12, 1987, pp.155-158. [7]. A.K. Jain, J. Mao, and K.M. Mohiuddin, “ Artificial Neural Networks: A Tutorial”, IEEE Computer, Vo1.29, No.3, 1996, pp.3144.
2520