Using optimized features for modified optical backpropagation

Journal of Information Engineering and Applications www.iiste.org
ISSN 2224-5782 (print) ISSN 2225-0506 (online)
Vol.4, No.8, 2014
Using Optimized Features for Modified Optical Backpropagation
Neural Network Model in Online Handwritten Character
Recognition System
Fenwa O.D*, Adetunji A.B* and Ajala F.A*
Department of Computer Science and Engineering, Ladoke Akintola University of Technology, Ogbomoso,
Nigeria
(*Email of Corresponding authors - fenwadeborah@yahoo.com, abadetunji@lautech.edu.ng,
faajala@lautech.edu.ng)
Abstract
One major problem encountered by researchers in developing character recognition system is selection of
efficient features (optimal features). In this paper, Particle Swarm Optimization (PSO) is proposed for feature
selection. However, backpropagation algorithm has been reported to be an effective and most widely used
supervised training algorithm for multi-layered feedforward neural networks but has the shortcomings of longer
training time and entrapment into a local minimal. Several research works have been proposed to improve this
algorithm but some of these research works were based on ‘learning parameter’ which in some cases slowed
down the training process. Hence, this paper has focused on alleviating the problem of standard backpropagation
algorithm based on ‘error adjustment’. To this effect, PSO is integrated with a ‘Modified Optical
Backpropagation (MOBP)’ neural network to enhancement the performance of the classifier in terms of
recognition accuracy and recognition time. Experiments were conducted on MOBP neural network and PSO-based
MOBP classifiers using 6,200 handwritten character samples (uppercase (A-Z), lowercase (a-z) English
alphabet and 10 digits (0-9)) collected from 100 subjects using G-Pen 450 digitizer and the system was tested
with 100 character samples written by people who did not participate in the initial data acquisition process.
Experimental results show promising results for the PSO-based MOBP classifier in terms of the performance
measures.
Keywords: Artificial Neural Network, Feature Extraction, Feature Selection, Particle Swarm Optimization,
Modified Optical Backpropagation.
1
1. Introduction
Advancement in computing technology has greatly influenced the lives of human beings and the usage of
computer is increasing at a tremendous rate. As computer systems become increasingly integrated into our
everyday life, it is therefore necessary to make them more easily accessible and user friendly. The ease with
which we can exchange information between user and computer is of immense importance today because input
devices such as keyboard and mouse have limitations. Owing to these limitations, researchers for over decades
have been attracted to device a quick and natural way of communication between computer systems and human
beings (Anita and Dayashankar, 2010; Fenwa, Omidiora and Fakolujo, 2012a; Fenwa et.al., 2012b).

Vol.4, No.8, 2014
The classes of recognition systems that are usually distinguished are online systems for which handwritten data
are captured during the writing process (which makes available the information on the ordering of the strokes)
and offline systems for which recognition takes place on a static image captured once the writing process is over
(Anoop and Anil, 2004; Liu, Stefan and Masaki, 2004; Mohamad and Zafar, 2004; Naser, Adnan, Arefin, Golam
and Naushad, 2009; Pradeep, Srinivasan and Himavathi, 2011). The online methods have been shown to be
superior to their offline counterpart in recognizing handwritten characters due the temporal information available
with the former (Pradeep, Srinivasan and Himavathi, 2011). Handwriting recognition system can further be
broken down into two categories: writer-independent recognition system which recognizes wide range of
possible writing styles and a writer-dependent recognition system which recognizes writing styles only from
specific users (Santosh and Nattee, 2009).
Statistical classifiers, Probabilistic classifiers, Artificial Neural Networks (ANN) are some of the widely used
image classifiers. The major drawback of the statistical classifiers is its inability to classify accurately. On the
other hand, probabilistic classifiers suffer from the setback of difficulty in estimating the conditional
probabilities. Artificial intelligence has been a major contribution to the advancement of computer science in that
it focuses on the creation of machines/systems that can mimic human thoughts, understand speech and countless
other feat that were assumed never to be possible. However, ANNs outperform the other classifiers because of its
flexibility, scalability, tolerance to faults, accuracy and learning. Generally speaking, the practical use of neural
networks has been recognized mainly because of such distinguished features as: general nonlinear mapping
between a subset of the past time series values and the future time series values. The capability of capturing
essential functional relationships among the data, which is valuable when such relationships are not a priori
known or are very difficult to describe mathematically and/or when the collected observation data are corrupted
by noise universal function approximation capability that enables modeling of arbitrary nonlinear continuous
functions to any degree of accuracy.
However, backpropagation algorithm has been reported to be an effective and most widely used supervised
training algorithm for multi-layered feedforward neural networks but has the shortcomings of longer training
time and entrapment into a local minimal (Freeman and Skapura, 1992). Several research proposals have been
made to improve this algorithm. Some of these research works were based on ‘learning parameter’ which in
some cases slowed down the training process (Minai, 1990; Riedmiller and Braun, 1993; Otair and Salameh,
2005). Thus, this paper will employ a modified optical backpropagation proposed by Fenwa et al. 2012 which
focused on alleviating the problem of standard backpropagation algorithm based on ‘error adjustment’.
2. Related Work
Otair and Salameh (2005) proposed an online handwritten character recognition system using an Optical
Backpropagation network. Two neural networks were developed and trained to recognize handwritten
characters. Also two algorithms were use, namely classical Backpropagation (BP) and Optical Backpropagation
(OBP), which applies a non-linear function on the error from each output unit before applying the
Backpropagation phase. The OBP aimed at speeding up the training process and escape from local minima and
was successful at that.
2

Vol.4, No.8, 2014
In 2012, Fenwa et.al proposed a modified Optical Backpropagation neural network model in online handwritten
character recognition system using hybrid of geometrical and statistical features. The proposed system showed
better performance when compared with the existing Optical backpropagation neural network in literature.
In this paper, Particle Swarm Optimization (PSO) algorithm is integrated with a ‘Modified Optical
Backpropagation (MOBP)’ neural network to enhancement the performance of the classifier in terms of
recognition accuracy and recognition time.
3. Research Methodology
In this paper, four stages of development of the proposed character recognition system which include; data
acquisition, pre-processing, character processing which consists of feature extraction and feature selection and
classification using MOBP, and PSO-based MOBP classifiers as shown in Figure 3.1. Experiments were
performed with 6,200 handwriting character samples (uppercase (A-Z) lowercase (a-z) English alphabet) and
digits (0-9) collected from 100 subjects using G-Pen 450 digitizer and the system was tested with 100 character
samples written by people who did not participate in the initial data acquisition process. The performance of the
system was evaluated based on convergence time and recognition accuracy.
3.1 Character Acquisition
The data used in this work were collected using Digitizer tablet (G-Pen 450). The G-Pen has an electric pen with
sensing writing board. An interface was developed using C# to acquire geometrical features such as stroke
number, pressure used in writing the strokes of the characters, number of junctions and the location in the written
characters and the horizontal projection count of the character from different subjects using the Digitizer tablet.
Characters considered were 26 upper case (A-Z), 26 lower case (a-z) English alphabets and 10 digits (0-9)
making a total number of 62 characters. 6,200 characters were collected from 100 and this serves as the training
data set which was the input data that was fed into the neural network. Sample data used were as shown in figure
3.2
3.2 Feature Extraction
This research focuses on a feature extraction technique that combined three characteristics of the handwritten
character to create a global feature vector. A hybrid feature extraction algorithm was developed using
Geometrical and Statistical features. Integration of Geometrical and Statistical features was used to highlight
different character properties, since these types of features are considered to be complementary. Eleven features
from two categories (Geometrical features and Statistical features) were used in this work.
The Hybrid (Geom-Statistical) Feature Extraction Algorithm proposed by Fenwa, Omidiora and Fakolujo, 2012a
was used:
Step 1: Get the stroke information of the input characters from the digitizer (G-pen 450)
These include:
(i) Pressure used in writing the strokes of the characters
3

Vol.4, No.8, 2014
(ii) Number (s) of strokes used in writing the characters
(iii) Number of junctions and the location in the written characters
(iv) The horizontal projection count of the character
Step 2: Apply Contour tracing algorithm to trace out the contour of the characters
Step 3: Run Hybrid Zoning algorithm on the contours of the characters
Step 4: Feed the outputs of the extracted features of the characters into the digitization stage in order to convert
4
all the extracted features into digital forms
The Hybrid Zoning Algorithm: Hybrid of Modified ICZ and Modified ZCZ
Input: Pre-processed character image
Output: Features for Classification and Recognition
Method Begins
Step 1: Divide the input image into 25 equal zones.
Step 2: Compute the input image centroid
Step 3: Compute the distance between the image centroid to each pixel present in the zone.
Step 4: Repeat step 3 for the entire pixel present in the zone.
Step 5: Compute average distance between these points.
Step 6: Compute the zone centroid
Step 7: Compute the distance between the zone centroid to each pixel present in the zone.
Step 8: Repeat step 7 for the entire pixel present in the zone
Step 9: Compute average distance between these points.
Step 10: Repeat the steps 3-9 sequentially for the entire zones.
Step 11: Finally, 2xn (50) such features were obtained for classification and recognition.
Method Ends
3.3 Feature Selection
Feature selection refers to the problem of dimensionality reduction of data, which initially consists of large
number of features. The objective is to choose optimal subsets of the original features which still contain the
information essential for the classification task while reducing the computational burden imposed by using many
features. In this work, Particle Swarm Optimization is proposed for feature selection.

Vol.4, No.8, 2014
Particle Swarm Optimization (PSO)
The PSO method is a member of wide category of Swarm Intelligence methods for solving the optimization
problems. It is a population based search algorithm where each individual is referred to as particle and represents
a candidate solution. In this paper the PSO algorithm proposed by Anita and Jude 2010 was employed. Each
single candidate solution is “an individual bird of the flock”, that is, a particle in the search space. Each particle
makes use of its individual memory and knowledge to find the best solution. All the particles have fitness values,
which are evaluated by fitness function to be optimized and have velocities which direct the movement of the
particles. The particles move through the problem space by following a current of optimum particles. The initial
swarm is generally created in such a way that the population of the particles is distributed randomly over the
search space. At every iteration, each particle is updated by following two “best” values, called pbest and gbest.
Each particle keeps track of its coordinates in the problem space, which are associated with the best solution
(fitness value). This fitness value is called pbest. When a particle takes the whole population as its topological
neighbor, the best value is a global best value and is called gbest. The detailed algorithm is given as follows:
Step 1: Set the constants kmax, c1, c2, r1, r2, w.
Randomly initialize particle positions x0(i) for i = 1, 2………p.
Randomly initialize particle velocities v0(i) for i = 1, 2……..p.
Step 2: Set k =1.
Step 3: Evaluate function value fk using design space coordinates xk(i)
If fk ≥ fpbest, then pbest(i) = xk(i)
If fk ≥ fgbest, then gbest= xk(i)
Step 4: Update particle velocity using the following equation
vk+1(i)= w*(vk(i)) + c1r1*(pbestk(i) - xk(i)) + c2r2(gbestk – xk(i)) (1)
Update particle position vector using the following equation
xk + 1(i) = xk(i) + vk + 1(i) (2)
Step 5: Increment i. If i > p, then increment k and set i = 1.
Step 6: Repeat steps 3 to 5 until kmax is reached.
The notations used in this algorithm are:
kmax = maximum iteration number
w = inertia weight factor
c1, c2 = cognitive and social acceleration factors
r1, r2 = random numbers in the range (0, 1).
In this paper, each of the eleven features are represented by a chromosome (string of bits) with 11 genes (bits)
corresponding to the number of features. An initial random population of 20 chromosomes is formed to initiate
the genetic optimization. The initial coding for each particle is randomly generated. The order of position of the
features in each string is pressure of the stroke, stroke number, horizontal projection count, contour pixel, image
centroid, zone centroid, distance between zone centroid, distance between image centroid, horizontal centre of
gravity and vertical centre of gravity respectively. A suitable fitness function is estimated for each individual.
5

Vol.4, No.8, 2014
The fittest individuals are selected and the crossover and the mutation operations are performed to generate the
new population. This process continues for a particular number of generations and finally the fittest chromosome
is calculated based on the fitness function. The features with a bit value “1” are accepted and the features with
the bit value of “0” are rejected. The fitness function used in this work is given by
Fitness (α * γ ) + β * (3)
where γ = classification accuracy
6
c = total number of features
r = length of the chromosome (number of ‘1’s)
α א[0, 1] and β = 1- α
This formula shows that the classification accuracy and the feature subset length have different significance for
feature selection. A high value of α assures that the best position is at least a rough set reduction. The goodness
of each position is evaluated by this fitness function. The criteria are to maximize the fitness values. An optimal
solution is obtained at the end of the maximum iteration. This value is binary coded with eleven bits. The bit
value of “1” represents a selected feature whereas the bit value of “0” represents a rejected feature. Thus an
optimal set of features are selected from the PSO technique. Out of the eleven features extracted, seven optimal
set of features are selected from the PSO algorithm.
3.4 Classification Method
In this work, two types of neural network classifiers are used and these are: the Modified Optical
Backpropagation neural network and the PSO-Based Modified Optical Backpropagation neural network.
3.4.1 The Optical Backpropagation Neural Network:
The difficulty encountered in the standard Backpropagation algorithm is when the actual value fo
k(neto
pk)
approaches either extreme value, the factor fo
k ((neto pk) . (1 - fo
k(neto
pk))) makes the error signal very small. The
Optical Backpropagation algorithm (an enhanced Backpropagation) focused on this delay of the convergence
that is caused by the derivative of the activation function. However, a slight modification of the error signal
function of the standard Backpropagation algorithm has resolved this shortcoming and indeed greatly accelerates
the convergence to a solution. The adjustment of the new algorithm (OBP) is described to improve the
performance of the BP algorithm. The convergence speed of the training process was improved significantly by
OBP through maximizing the error signal, which was transmitted backward from the output layer to each unit in
the intermediate layer.
In BP, the error at a single output unit is defined according to equation (4) as:
δo
pk = (Ypk - Opk ) . fo'
k (neto
pk) (4)
where the subscript “p” refers to the pth training vector, and “k” refers to the kth output unit. In this case,
Ypk is the desired output value, and Opk is the actual output from kth unit, then δo
pk will propagate backward to
update the output-layer weights and the hidden-layer weights while the error at a single output unit in adjusted
OBP is given as (Otair, & Salameh, 2005):

Vol.4, No.8, 2014
7
New δo
pk = (1+e (Ypk - Opk )2 . fo'
k (neto
pk)), if (Y – O) >= zero (5a)
pk = - (1+e (Ypk - Opk )2 . fo'
New δo
k (neto
pk)), if (Y – O) < zero (5b)
An OBP uses two forms of Newδo
pk, because the exponential function always return zero or positive values,
while adapts operation for many output units need to decrease the actual outputs rather than increasing it. The
Newδo
pk will propagate backward to update the output-layer weights and the hidden-layer weights. This Newδo
pk
minimized the errors of each output unit more quickly than the old δo
pk, and the weights on certain units change
very large from their starting values.
The steps of an OBP (Otair and Salameh, 2005)
1. Apply the input example to the input units.
2. Calculate the net-input values to the hidden layer units.
3. Calculate the outputs from the hidden layer.
4. Calculate the net-input values to the output layer units
5. Calculate the outputs from the output units
6. Calculate the error term for the output units, using the Newδo
pk (using equations 5a and 5b)
instead of δo
pk in equation (4)
7. Calculate the error term for the hidden units, through applying New δo
pk, also
Newδh
pj = fh'
j(neth
pj) . ( δo
pk . Wo
kj) (6)
8. Update weights on the output layer.
Wo kj(t+1) = Wo kj(t) + (η .Newδo
pk . ipj) (7)
9. Update weights on the hidden layer.
Wh
ji(t+1) = Wh
ji(t) + (η . Newδo
pj . Xi) (8)
10. Repeat steps from step 1 to step 9 until the error (Ypk – Opk) is acceptably small for each
training vector pairs.
The proposed algorithm as classical BP is stopped when the squares of the differences between the actual and
target values summed over units and all patterns are acceptably small.
3.4.2 Modified Optical Backpropagation Neural Network
The error function defined in Optical Backpropagation earlier is proportional to the square of the Euclidean
distance between the desired output and the actual output of the network for a particular input pattern. As an
alternative, other error functions whose derivatives exist and can be calculated at the output layer can replace the

Vol.4, No.8, 2014
traditional square error criterion (Haykin, 2003). In this research work, error of the third order (Cubic error) had
been adopted to replace the traditional square error criterion used in Optical Backpropagation. The equation of
the cubic error is given as:
j(neth
pj) .( δo
8
δo
pk = -3(Ypk – Opk)2 . fo'
k(neto
pk) (9)
The cubic error in equation (6) was manipulated mathematically and this further maximized the error signal of
each output unit which was transmitted backward from the output layer to each unit in the intermediate layers
(Fenwa, Omidiora, Fakolujo and Ganiyu, 2012).
The derived equations were as shown in equations (10a) and (10b) below:
Modified δo
pk = 3((1+ et)2 . fo'
k(neto
pk)) If (Ypk – Opk)2 >= 0 (10a)
Modified δo
pk = -3((1+ et)2 . fo'
k(neto
pk)) If (Ypk – Opk)2 <=0 (10b)
where Ypk = Target or Desired output
Opk = Network output
t = (Ypk – Opk)2
However, one of the ways to reduce the training time is through the use of momentum, as it enhances the
stability of the training process. The momentum was used to keep the training process going in the same general
direction (Haykin, 2003). In the modified Optical Backpropagation network, momentum was introduced. Hence,
equation (6) becomes
Wo
kj(t+1) = Wo
kj(t) + μWo
kj(t) + (η . Modified δo
pk. ipj) (11)
where μ is the momentum coefficient typically about 0.9 and η is the learning rate.
The Modified Optical Backpropagation Algorithm:
Modifications of the algorithm are in terms of Error Signal Function
With the introduction of Cubic error function and Momentum, the modified Optical Backpropagation is given as:
1. Apply the input example to the input units.
2. Calculate the net-input values to the hidden layer units.
3. Calculate the outputs from the hidden layer.
4. Calculate the net-input values to the output layer units
5. Calculate the outputs from the output units
6. Calculate the error term for the output units, using equation (10a) and (10b) instead of equations (5a)
and (5b)
7. Calculate the error term for the hidden units, through applying improve δo
pk also
Modified δh
pj = fh'
pk . Wo
kj) (12)

Vol.4, No.8, 2014
9
8. Update weights on the output layer.
Wo
kj(t+1) = Wo
kj(t) + μWo
kj(t) + (η . Modified δo
pk . ipj) (13)
9. Update weights on the hidden layer.
Wh
ji(t+1) = Wh
ji(t) + (η . Modified δh
pj . Xi) (14)
Repeat steps from step 1 to step 9 until the error (Ypk – Opk) is acceptably small for each of the training vector
pair. The proposed algorithm as OBP is stopped when the cubes of the differences between the actual and target
values summed over units and all patterns were acceptably small.
3.4.3 The PSO-Based Modified Optical Backpropagation
The second classifier used in this work is PSO-Based MOBP classifier. The objective for using the optimization
algorithm is two folds: (i) dimensionality reduction which improves the convergence rate and (ii) elimination of
insignificant features which improves the classification accuracy. In this work Particle Swarm Optimization
algorithm is used for optimal feature selection. The extracted features are subjected to this optimization
technique which finally yields the optimal feature set. The number of neurons used in the input layer for this
PSO-Based MOBP classifier is reduced since the number of optimal features is lesser than the complete feature
set. Also, the mathematical calculations are minimized because of the reduced size of the weight matrix. Hence,
a significant reduction in the time period is achieved for the weight adjustment of the hidden layer neurons.
Thus, the PSO-Based MOBP neural network is better than MOBP neural network.
4. Experiment
Experiments were carried out on a Hewlett Packard system with the configuration: 64 bits operating system,
4.00G RAM and Intel(R) CORE(TM) i5-3210M CPU @ 2.50GHz processor. The system was implemented
using C# programming language.
5. Results and Discussions
Experiments were performed with 6,200 handwriting character samples (uppercase (A-Z) lowercase (a-z)
English alphabet) and digits (0-9) collected from 100 subjects using G-Pen 450 digitizer and the system was
tested with 100 character samples written by people who did not participate in the initial data acquisition
process. The performance of the system was evaluated based on convergence time and recognition accuracy.
It was shown in Table 1 that the more the dimensional input vector (character matrix size), the more the number
of epochs. Usually, the complex and large sized input sets require a large topology network with more number of
iterations (Epochs). The epochs is directly proportional to the training time, this implies that the larger the image
size, the more the training time. Three different image sizes (5 by 7, 10 by 14 and 20 by 28) were considered in
this paper and it was shown from the results that the higher the image size, the higher the number of epochs
required to train the network due to increment in vector space to be processed by the network. From the above
table, it is evident that the PSO-Based MOBP is superior to MOBP classifier in terms of convergence rate.

Vol.4, No.8, 2014
From table 2, the training time of PSO-Based MCPN is smaller when compared with the MOBP classifier due to
its ability to achieve dimensionality reduction and removal of irrelevant features of character images.
Classification accuracy is the ratio of the number of correctly classified images to the total number of images.
Convergence time period is the time taken for training process and testing process. From table 3, it is clearly
understood that the architecture of the PSO-Based MOBP is highly simplified and is less prone to error in
classification than the MOBP classifier. It also reveals the less number of mathematical computational operations
involved in PSO-Based MOBP. The PSO-Based MOBP classifier has better classification accuracy than MOBP
classifier.
6. Conclusion and Future Work
This paper explores the need for optimization algorithms to enhance the performance of the classifiers. In this
work, PSO is used as the optimization algorithm and it is used along with the modified Optical Backpropagation
classifier. Experimental results suggest better improvement in the classification accuracy for the PSO-Based
MOBP than the other classifier (MOBP). However, an increase in the convergence rate is also achieved by the
PSO-based MOBP classifier which is highly essential for real-time applications. Therefore an optimization
technique is highly essential irrespective of the classifiers under consideration.
Finally, the application of PSO optimization algorithm for performance improvement of the neural classifier has
been explored in the context of online character image classification. Future can be tailored towards
hybridization of other classifiers to further enhance the performance of the system. The work can also be
extended by using different optimization algorithms to estimate the performance of the classifiers. However,
different set of features can be used to improve the classification accuracy and experiments can be carried out on
a different set of database in order to generalize the technique. Irrespective of the modifications and the systems
used, this paper has been able to present the significance of optimization algorithm for accurate and quick image
classification systems.
10
7. References
Anita, P. and Dayashankar S. (2010): ‘’Handwritten English Character Recognition using Neural Network’’,
International Journal of Computer Science and Communication, 1(2): 141-144.
Anoop, M. N. and Anil K.J. (2004): ‘’Online Handwritten Script Recognition’’, IEEE Trans. PAMI, 26(1): 124-
130.
Fenwa O.D., Omidiora E.O., Fakolujo O.A. (2012a): ‘’Development of a Feature Extraction Technique for
Online Character Recognition System’’, Journal of Innovation System Design and Engineering, 3(3):10-23.
Fenwa, O.D., Omidiorah, E.O., Fakolujo, O.A. and Ganiyu, R.A. (2012b): ‘’Development of a Writer-
Independent Online Handwritten Character Recognition System using modified Hybrid Neural Network
Model’’, Accepted Manuscript for Publication in Journal of Computer Engineering and Intelligent System,
International Institute of Science, Technology and Education, New York, USA, Paper ID: 2011471827.
Freeman, J.A. and Skapura, D.M. (1992): Backpropagation Neural Networks Algorithm Applications and
Programming Techniques, Addison-Wesley Publishing Company: 89-125.
Haykin, S. (2003): Neural Networks: A Comprehensive Foundation, PHI, New –Delhi, India
Hecht-Nielsen, R. (1990): Neurocomputing: Addison-Wesley Publishing Company.
Jude, D. H, Vijila, C. K. and Anitha, J (2010): ‘’Performance Improved PSO Based Modified
Counterpropagation Neural Network for Abnormal Brain Image Classification’’, International Journal Advanced
Soft Computing Application, 2(1): 65-84.

Vol.4, No.8, 2014
Liu, C.L., Nakashima, K., Sako, H. and Fujisawa, H. (2004): ‘’Handwritten Digit Recognition: Investigation of
Normalization and Feature Extraction Techniques’’, Pattern Recognition, 37(2): 265-279.
Minai, A.A. and Williams, R.D. (1990): ‘’Acceleration of Backpropagation through learning Rate Momentum
Adaptation’’, Proceeding of the International Joint Conference on Neural Networks: 1676-1679.
Mohamad, D. and Zafar, M.F. (2004): ‘’Comparative Study of Two Novel Feature Vectors for Complex Image
Matching Using Counterpropagation Neural Network’’, Journal of Information Technology, FSKSM, UTM,
16(1): 2073-2081.
Morita M., Sabourin R., Bortolozzi F., Suen C. Y. (2003): ‘’ A Recognition and Verification Strategy for
Handwritten Word Recognition’’, ICDAR'03), Edinburgh-Scotland: 482-486.
Naser, M.A., Adnan, M., Arefin, T.M., Golam, S.M. and Naushad, A. (2009): Comparative Analysis of Radon
and Fan-beam based Feature Extraction Techniques for Bangla Character Recognition, IJCSNS International
Journal of Computer Science and Network Security, 9(9): 120-135.
Otair, M.A. and Salameh, W.A. (2005): ‘’ Speeding up Backpropagation Neural Network’’ Proceedings 0f the
2005 Informing Science and IT Education Joint Conference 2: 167-173.
Pradeep, J., Srinivasan, E. and Himavathi, S. (2011): ‘’Diagonal Based Feature Extraction for Handwritten
Alphabets Recognition using Neural Network’’, International Journal of Computer Science and Information
Technology (IJCS11), 3(1): 27-37.
Riedmiller, M. and Braun, H. (1993): ‘’ A Direct Adaptive Method for Faster Backpropagation learning the
PROP Algorithm’’, Proceedings of the IEEE International Conference on Neural Networks (ICNN), 1: 586-591,
Francisco.
Santosh, K.C. and Nattee, C. (2009): ‘’A Comprehensive Survey on Online Handwriting Recognition
Technology and Its Real Application to the Nepalese Natural Handwriting’’, Kathmandu University Journal of
Science, Engineering Technology, 5(1): 31-55.
11

Vol.4, No.8, 2014
CHARACTER IMAGE
DATABASE
Collection of Online character
Images using G-pen 450 Digitizer
PRE- PROCESSING
- Extreme coordinate measurement
- Grid Resizing
CHARACTER PROCESSING
Feature Extraction
Using Geometric and Statistical Features
Feature Selection
Using PSO (Particle Swam Optimization)
TRANING/CLASSIFICATION
MOBP Classifier PSO-Based MOBP Classifier
Testing
Performance
Evaluation
FIG. 3.1: THE BLOCK DIAGRAM OF THE PROPOSED PSO-BASED MOBP CHARACTER RECOGNITION SYSTEM
12

Vol.4, No.8, 2014
Figure 3.2: Sample Data collected using G-pen 450 Digitizer
Table 1: Epochs Values with different Image Sizes for the two Classifiers
Image Sizes
MOBP
PSO-Based MOBP
5 by 7 1120 920
10 by 14 1600 1200
20 by 28 1864 1417
Table 2: The Training Time (in minutes) of the two classifiers under different datasets
13
Character
Samples
MOBP
Training Time
PSO-Based
MOBP Training
Time
1,200 0.84 2.28
2,480 7.01 4.01
3,720 9.87 6.36
4,960 18.69 13.98
6,200 52.68 36.79

Vol.4, No.8, 2014
Table 3: Classification Accuracies of the three Classifiers
14
Modified Optical
Backpropagation
(MOBP)
PSO-Based Modified
Optical
Backpropagation
(PSO-Based MOBP)
Character
Samples
CR
(%)
FR
(%)
RF
(%)
CR
(%)
FR
(%)
RF
(%)
1,240 75 6 2 79 3 1
2,480 80 3 1 85 2 1
3720 86 1 1 89 2 0
4,960 92 1 0 95 1 0
6,200 95 1 0 98 1 0
CR: Correct Recognition, FR: False Recognition, RF: Recognition Failure

The IISTE is a pioneer in the Open-Access hosting service and academic event
management. The aim of the firm is Accelerating Global Knowledge Sharing.
More information about the firm can be found on the homepage:
http://www.iiste.org
CALL FOR JOURNAL PAPERS
There are more than 30 peer-reviewed academic journals hosted under the hosting
platform.
Prospective authors of journals can find the submission instruction on the
following page: http://www.iiste.org/journals/ All the journals articles are available
online to the readers all over the world without financial, legal, or technical barriers
other than those inseparable from gaining access to the internet itself. Paper version
of the journals is also available upon request of readers and authors.
MORE RESOURCES
Book publication information: http://www.iiste.org/book/
IISTE Knowledge Sharing Partners
EBSCO, Index Copernicus, Ulrich's Periodicals Directory, JournalTOCS, PKP Open
Archives Harvester, Bielefeld Academic Search Engine, Elektronische
Zeitschriftenbibliothek EZB, Open J-Gate, OCLC WorldCat, Universe Digtial
Library , NewJour, Google Scholar

Using optimized features for modified optical backpropagation

More Related Content

What's hot (20)

Viewers also liked (8)

Similar to Using optimized features for modified optical backpropagation (20)

More from Alexander Decker (20)

Using optimized features for modified optical backpropagation