Pattern Recognition 58 (2016) 172–189

Contents lists available at ScienceDirect

Pattern Recognition
journal homepage: www.elsevier.com/locate/pr

A multi-objective approach towards cost effective isolated handwritten

Bangla character and digit recognition
Ritesh Sarkhel, Nibaran Das n, Amit K. Saha, Mita Nasipuri
Computer Science and Engineering Department, Jadavpur University, Kolkata 700032, India

art ic l e i nf o a b s t r a c t

Article history: Identifying the most informative local regions of a handwritten character image is necessary for a robust
Received 22 September 2015 handwritten character recognition system. But identifying them from a character image is a difficult task.
Received in revised form If this task were to be performed incurring minimum possible cost, it becomes more challenging due to
22 March 2016
having two independent, apparently contradicting objectives which need to be optimized simulta-
Accepted 13 April 2016
Available online 22 April 2016
neously, i.e. maximizing the recognition accuracy and minimizing the associated recognition cost. To
address the problem a multi-objective approach is required. In the present task, two popular multi-
Keywords: objective optimization Algorithm (1) a Non-Dominated Sorting Harmony-Search Algorithm (NSHA) and
Feature set (2) a Non-Dominated Sorting Genetic Algorithm-II (NSGA-II, Deb et al., 2002 [18]) are employed for
Region sampling
region sampling separately. The method objectively selects the most informative set of local regions
Handwritten character recognition
using the framework of Axiomatic Fuzzy Set (AFS) theory, from the sets of pareto-optimal solutions
Multi-objective evolutionary algorithm
Harmony search provided by the multi-objective region sampling algorithms. The system has been evaluated on two
NSGA-II isolated handwritten Bangla datasets, (1) a dataset of randomly mixed handwritten Bangla Basic and
AFS theory Compound characters and (2) a dataset of handwritten Bangla numerals separately, with SVM based
classifier, using a feature set containing convex-hull based features and CG based quad-tree partitioned
longest-run based local features extracted from the selected local regions. The results have shown a
significant increase in recognition accuracy and decrease in recognition cost for all the datasets. Thus the
present system introduces a cost effective approach towards isolated handwritten character recognition
systems.
& 2016 Elsevier Ltd. All rights reserved.

1. Introduction
Optical Character Recognition (OCR) is an active area of research.
While there are many systems commercially available for recog-
nizing printed text [1–4], their success is yet to be extended to
handwritten characters. Several reasons can be cited to explain this
apparent anomaly. Shape and size of handwritten characters vary
from one individual to another. It may even vary for a single indi-
vidual from time to time, depending on various factors. These
challenges make the task of recognizing handwritten characters
very difficult. Researchers all around the world have proposed
several methods [5] for handwritten character recognition, but
most of them are focused on Roman scripts [6], concentrating on
English and other European languages. Among Asian languages,
Chinese [7], Japanese, Korean languages are dominant in the lit-
erature. Indian scripts like Malayalam, Tamil, Telugu, and Hindi have
started to get attention of the researchers during past decade [8,9],
n
Corresponding author. Tel./fax: þ 91 3324146766.
E-mail address: nibaran@gmail.com (N. Das).
but development of OCR for complete Bangla script [10] has not
received much attention from researchers until recently. Bangla is
the second most popular script in India and the fifth most popular
script in the world [11]. Bangla alphabet contains some of the most
intricate and complex characters, which differ from one another
only by a single period, a modifier ref or an upper horizontal line or
Matra, as shown in an example in Fig. 1. Bangla alphabet contains
about 50 Basic characters (11 vowels and 39 consonants) and more
than 334 Compound characters [12]. Samples of a few of Bangla
Basic and Compound characters is shown in Fig. 2.
One of the most common approaches taken up by OCR
researchers is zoning, i.e. dividing the character image into several
zones or local regions [13] and generating the invariant local feature
set by extracting features from every local region. There are several
different zoning methods [13] mentioned in the literature, but most
of them can be classified into two major categories: static [4,10]
and dynamic zoning methods [13]. Static zoning methods divide a
handwritten character image into a fixed set of overlapping or
non-overlapping windows, where the number of windows is fixed.
Basu et al. used static zoning method in [14] and sub-divided the
handwritten numerals’ image into 9 fixed-sized, overlapping local

http://dx.doi.org/10.1016/j.patcog.2016.04.010
0031-3203/& 2016 Elsevier Ltd. All rights reserved.
 R. Sarkhel et al. / Pattern Recognition 58 (2016) 172–189
Fig. 1. Similarity in shape and size between different Bangla characters. (a). Bangla Basic character ‘ ’ (b) Bangla Basic character ‘ ’.
Fig. 2. Samples of handwritten Bangla characters.
regions and extracted longest-run based features form each sub-
region. On other hand, dynamic zoning methods sub-divide a
handwritten character image into local regions by dynamically
creating windows based on some statistical or topological feature
of that specific character. Cao et al. [8] proposed a similar techni-
que to generate a hierarchical feature-space based on a quin-tree
partition of the character image, where zones were dynamically
created based on the centroid of the contour segment of the
character residing in the parent zone. Das et al. [10,15] have used a
GA based selection mechanism to find out the most optimal set of
local regions for recognition of handwritten Bangla numerals.
In those papers, the researchers have emphasized on achieving
better recognition accuracies, but associated recognition costs
incurred in the process were not taken into consideration. For
example, Das et al. presented a two pass approach towards
handwritten character recognition in [10], which produced a sig-
nificant increase in the recognition accuracy but at the cost of a
recognition cost which is almost 8.5 times than the average per
character recognition cost incurred by traditional single pass
approach towards handwritten character recognition. This may
prove to be undesirable to users who want to use such a system for
real-life applications. An extensive study of recognition accuracy
versus associated recognition cost is undertaken in our experi-
mental setup to investigate the scope of a practical optical char-
acter recognition system, in terms of both recognition accuracy
and associated recognition cost.
In the present work, a multi-objective approach towards optical
character recognition (OCR) is proposed, which attempts to find a
trade-off between the recognition accuracy achieved by the sys-
tem and its associated recognition costs. In real life applications of
an OCR system, insignificant increase in recognition accuracy at
the expense of high recognition cost may not be acceptable to the
users of the system. In such cases, a multi-objective approach can
provide the user with a set of good solutions. In the present work,
framework of a novel, multi-objective isolated handwritten char-
acter recognition system is proposed. There are several variants of
multi-objective Evolutionary Algorithms [16] present in the lit-
erature. A Non-dominated Sorting Harmony-search Algorithm (NSHA
[17]) based region sampling method and a Non-dominated Sorting
Genetic Algorithm – II (NSGA-II [18]) based region sampling method
is introduced in our present work. These two multi-objective
173
region sampling algorithms mark one of the contributions of the
present work. Both of the region sampling algorithms are
employed over the decision space separately. These algorithms
have two objective functions: – (1) maximizing handwritten
character recognition accuracy and (2) minimizing associated
recognition costs. In our experimental setup, recognition accuracy
is measured using an SVM based classifier and recognition costs
are measured by: (i) average time taken by the recognition system
to recognize each handwritten character in the test-set and (ii) the
number of local regions used to represent each handwritten
character in the test-set. Two sets of pareto-optimal solutions
provided by these two algorithms are then combined using Axio-
matic Fuzzy Set (AFS) Theory [19]. The multi-objective region
sampling algorithms and the AFS theory based approach to
objectively combine the pareto-optimal solutions provided by the
multi-objective algorithms mark one of the contributions of the
present work.
The proposed method tries to find an objective solution over
the decision space, while providing an optimal trade-off between
recognition accuracy and corresponding recognition costs, making
it suitable to use in practical applications. The present work has
been evaluated on datasets of isolated handwritten Bangla char-
acters and handwritten Bangla numerals separately. Results from
these experiments have been compared with some of the other
popular handwritten character recognition methods present in the
literature, to prove its superiority.
The rest of the paper is organized as follows: in Section 2, a
brief overview on multi-objective evolutionary algorithms based
region sampling techniques is presented, basics of Axiomatic
Fuzzy Set (AFS) theory is introduced in Section 3; Section 4
describes the featureset and our present work is discussed in
details in Section 5, experimental results are presented in Section
6. Finally, a brief conclusion is drawn based on the results gathered
from the experiments.
2. Motivation behind using multi-objective evolutionary
algorithms for region sampling
Region sampling based OCR systems try to identify the most
discriminative set of local regions from handwritten character
174 R. Sarkhel et al. / Pattern Recognition 58 (2016) 172–189
images. The easiest approach is to exhaustively enumerate every
possible combination of local regions until the best combination is
found. This approach however, would take time of exponential
order. Therefore, Evolutionary Algorithm based meta-heuristic
approaches are generally employed, so that a good enough solu-
tion is found within a reasonable amount of time.
Most of the proposed methods present in handwritten char-
acter recognition literature [11] however, focus only on increasing
the accuracy of the recognition system. Sometimes such incre-
ments in recognition accuracy may come at an expense of
increased recognition cost. In such a scenario, where moderate
increments in recognition accuracy is achieved by incurring high
recognition costs, the solutions provided by a recognition system
may not be acceptable to the end-user. A handwritten character
recognition system, to be used in real-life applications, should not
just increase the average recognition accuracy of its test dataset
but keep a check on the associated recognition cost as well. Our
present work uses Multi-objective Evolutionary Algorithms to find
such a set of solutions. The multi-objective Evolutionary Algorithm
based region sampling technique, used in our experimental setup,
tries to identify the most informative set of local regions by
heuristically searching the solution space, while incurring mini-
mum possible recognition cost.
To formalize, in our experimental setup, the multi-objective
region sampling techniques try to optimize two separate objec-
tives simultaneously, which are: (a) maximizing the recognition
accuracy and (b) minimizing the cost associated with the recog-
nition process, as described before. The recognition cost is repre-
sented in terms of two major factors: (i) number of local regions
needed to represent the handwritten characters and (ii) average
time taken by the proposed system to recognize each handwritten
character. This set of solutions, which is also called the pareto-
optimal solutions [20] presents the users of the system with a set
of choices. Users may choose a single solution from the pareto-
optimal set, based on their real-time requirement and domain
expertize.
There are several variants of multi-objective Evolutionary
Algorithms [21] present in the literature. In our present work, we
have used a widely popular multi-objective Evolutionary Algo-
rithm, Non-dominated Sorting Genetic Algorithm (NSGA-II), pro-
posed by Deb et al. [18]. Since its introduction, NSGA-II has been
applied in various applications such as water network design [22],
construction management [23], economics [24], population plan-
ning [25] etc. To the best of our knowledge, the performance of
NSGA-II in a region sampling based handwritten character recog-
nition system is yet to be explored. In the present work, an NSGA-
II based region sampling algorithm is proposed. We have included
an extensive study of another multi-objective Evolutionary Algo-
rithm in our experimental setup. Harmony Search Algorithm is a
relatively new, music inspired meta-heuristics algorithm, pro-
posed by Geem et al. [26,44]. It has been proved [27] to provide
better performance than Genetic Algorithms for some applica-
tions. Several variants of multi-objective Harmony Search Algo-
rithms [28] are presented in the literature. They have been used to
address problems such as economic/environmental dispatch [29],
optimal power flow problem [17] etc. but to the best of our
knowledge, performance of a multi-objective Harmony Search
Algorithm in a region sampling based handwritten character
recognition system is yet to be explored. Here, we have introduced
a Non-dominated Sorting Harmony Search Algorithm (NSHA) based
region sampling technique for this purpose.
Both NSGA-II and NSHA based region sampling technique,
proposed in our present work, have been tested extensively on
publicly available datasets of isolated handwritten Bangla char-
acters and isolated handwritten Bangla numerals. Details about
the datasets used in the present work is provided in Section 4.
Using multiple region sampling algorithms and comparing their
results hereafter, help us to reach a conclusive decision about the
performance of the proposed system, as the results become
independent of the specificities of any particular algorithm in
consideration.
In our present work, an Axiomatic Fuzzy Set (AFS) Theory
based fuzzy logic is proposed to combine and finally return a
single solution from the candidate set of pareto-optimal solutions
extracted from the multi-objective region sampling algorithms
used in our experimental setup. AFS Theory has been employed in
our experimental setup, because its framework provides much
greater flexibility [19] in computing the interaction between a set
of local regions. It is also easier to compute the combined class-
separating power of a set of local regions, compared to other fuzzy
logic based systems. Important aspects of a fuzzy logic based
system like definition of membership function, logical operation
between a set of elements in the fuzzy featureset is already
defined [19] the framework of AFS theory.
3. A Brief overview on axiomatic set theory based fuzzy logic
A priori identification of the set of local regions containing the
most informative set of features is a difficult task. Local region or
feature selection methods, described in the literature, can be
broadly classified into two categories: wrappers and filters [30].
Wrappers identify the most informative set of features from the
discourse with the help of a training dataset and an efficient
learning algorithm; whereas filters use some kind of heuristics to
identify the feature set that has the most promise. For applications
like handwritten character recognition, merit of a local region is
interpreted by its contribution to the recognition system’s class-
separability power. Although there has been some literature that
uses entropy based fuzzy interpretation [31] of local features in
feature-ranking techniques for recognition of handwritten Hindi
numerals, fuzzification of features to determine the feature set
with most class-separability power for handwritten character
recognition has not been addressed that much. Collection of
individually good features does not ensure a good feature set [32].
Finding out all the possible combinations of local features is also a
difficult and time consuming task [33]. Our present work uses the
framework of AFS theory [19] based algebra to define fuzzy
semantics, membership functions and logical operations on a set
of local regions to identify the subset of local regions which has
the most class-separating power.
3.1. AFS algebra based fuzzy logic
AFS algebra mainly focuses on two things: how fuzzy sets are
created i.e. the definition of membership functions and how
logical operations on fuzzy sets are defined. The concepts extrac-
ted from a given dataset depend strongly on the observed data
[19]. For example, the concept of “heavy” (in terms of a person’s
weight) is not interpreted the same way in a dojo of Japanese
Sumo wrestlers, as it would be interpreted in a Ballet studio of
Broadway. Strong additional background knowledge on the data-
set is thus necessary to extract a concept.
A fuzzy concept is defined based on one or more features of the
dataset. Let M be the set of fuzzy concepts, M ¼ m1 ; m2 …::mn ,
where a concept mi is defined on a feature of X i A F. M can be
viewed as a building block, containing the elementary concepts
associated to each feature. Using M, every possible concept on X
can be easily formulated and represented, which essentially means
that every possible concept Y on X can be easily formulated using
 R. Sarkhel et al. / Pattern Recognition 58 (2016) 172–189 175

M, where C т ðx0 Þ is the set of all elements whose degree of belonging to

X   ∏m A C m is less than or equal to that of x0 and Mγ ð:Þ is a function
Y¼ j AJ
∏imi A C j mi where mi D C j : ð1Þ whose value always lies between 0 and 1.

3.1.1. Definitions 4. Design of the featureset

Let M be a non-empty set. Then the set EM* can be defined as
nX   4.1. Dataset of the experiment
EM  ¼ jAJ
∏imi A C j mi j C j A 2m ;
o The proposed method has been evaluated on four publicly
jA J; J maybe any non  empty index set : ð2Þ
available benchmark [12,34] datasets. A brief overview of the
A binary relation R between two fuzzy concepts is defined as datasets used in the present work is described in Table 1. More
follows: details about these databases such as sample collection, prepara-
P   P   tion techniques etc. can be found in the ‘Reference’ column pro-
For j A J ∏imi A C j mi ; i A I ∏kmk A Di mk A EM ;
X   X   vided in Table 1.
i k
j AJ
∏ m i A C j
m i R i AI
∏m k A Di
m k
( 4.2. Design of the feature set
8 C j ðj A J Þ ( Di ði A I Þsuch that C j + Di
 ð3Þ
8 Di ði A I Þ ( C j ðj A J Þ such that Di + C j
The extracted features, used in the present work can be clas-
It is clear from (3) that R is an equivalence relation and the sified into two categories: global and local. While the number of
P  i P  k  global features extracted from a handwritten character or digit
semantics j A J ∏mi A C j mi and i A I ∏mk A Di mk is equivalent
image is fixed, in case of local features it may vary based on the
under R.
number of local regions considered. Global features are extracted
The quotient EM  =R is called EM and each from the whole image, whereas local features are extracted from a
X  
∏kmk A Di mk ϵ EM is called a f uzzy concept: ð4Þ sub-region or local region [13] of the image being considered. Both
i AI
global and local features are used in the feature set so that the
inherent distinguishing pattern of a character or digit can be suf-
3.2. AFS structure ficiently quantified.

AFS structure formalizes a mathematical description of the data
structure used by the AFS algebra, a completely distributive lattice
[19] generated by the fuzzy datasets and the concepts behind
them. An AFS structure can be defined as a triplet ðM; т; X 0 Þ:
3.2.1. Definition
Let X 0 ; M are two finite sets. Let т be a relation defined as:
т: X 0  X 0 -2m . т is called an AFS structure if it follows two axioms:
ðaÞ ðx0 ; y0 Þ A X 0  X 0 ; тðx0 ; y0 Þ D тðx0 ; x0 Þ
ð5Þ
ðbÞ 8 ðx0 ; y0 Þ; ðy0 ; z0 Þ A X 0  X 0 ; тðx0 ; y0 Þ \ Tðy0 ; z0 Þ D тðx0 ; z0 Þ
0
In this definition, X is the universe of discourse, M is the concept
set and т is the axiomatic structure. In real world applications, т
can be defined as follows:
тðx0 ; y0 Þ ¼ fmj m A M; x0 Rm y0 g A 2m ;
where Rm represents binary relation of simple conceptm A M, and
x0 Rm y0 means the degree of x0 belonging to attribute m is larger
than or equal to that ofy0 .
P
The membership function of a fuzzy concept γ ¼ γ ¼ j A J
 
∏imi A C j mi is defined as follows:
  
 
θγ ¼ supj ∏γ A C j ℳγ C тj ðx0 Þ ; where C T ðx0 Þ ¼ z0 A X 0 j Tðx0 ; z0 Þ + C
4.2.1. Global features
The number of global features used in our experimental setup
is 175, out of which 155 are convex-hull based features and the
rest are quad-tree based longest run based features. Convex hull
based features, proposed by Das et al. [40] are used in the present
work. The feature set includes maximum dcp, total number of
rows having dcp 4 0, average dcp, number of visible bays, etc. CG
based quad-tree partitioned longest-run features were suggested
by Basu et al. in the year 2009. Longest run features extracted from
the root node (1*4 ¼ 4) and the first-level child-nodes (4*4 ¼ 16) of
the resultant quad-tree also contribute to the global feature set
(4 þ16 ¼20). Details about the feature extraction techniques can
be found in reference [41].
4.2.2. Local features
ð6Þ
The features extracted from the second level of child nodes of
the CG based quad-tree [42] contribute to the local feature set. The
number of nodes in the second level of the quad-tree is 16
(42 ¼16). Longest run based features are extracted from each local
region represented by a node in the quad-tree, along four axes,
horizontal, vertical and two diagonals. Each local region is repre-
sented by 4 local features. Hence, total number of local features in
the feature set of our present work is 64 (16*4¼ 64) and the
aggregate number of features in the integrated feature set of the
ð7Þ
experimental setup is 239 (175þ 64 ¼239).

Table 1
Datasets used in the present work.

Index Name of the dataset Dataset type Number of Number of training Number of test Reference
classes samples samples

DB1 CMATERdb 3.1.1 Isolated, handwritten Bangla numerals 10 4000 2000 [35,36]
DB2 ISI handwritten Bangla numeral Isolated, handwritten Bangla numerals 10 19392 4000 [37,38]
database
DB3 CMATERdb 3.2.1 Isolated, handwritten Bangla Basic characters 50 12000 3000 [35,36]
DB4 CMATERdb 3.2.1 þ CMATERdb Randomly mixed, isolated, handwritten Bangla 384 46919 11661 [36,39]
3.1.3.3 Basic and Compound characters
176 R. Sarkhel et al. / Pattern Recognition 58 (2016) 172–189
A brief overview of the feature extraction techniques employed
in our present work is presented in Fig. 3.
5. Present work
As discussed before, objectives of our present work is threefold.
They are: (a) proposing a region sampling methodology that
returns a subset of local regions containing the most dis-
criminating features, while incurring minimum possible recogni-
tion cost (b) evaluating the performance of the proposed method
on two separate, publicly available datasets of Bangla handwritten
characters and (c) comparing the performance of the proposed
method against some of the popular contemporaries present in
the literature. Fig. 4 presents a block diagram of the proposed
system.
5.1. Definitions and notations
Let I HW is a 2D array that denotes a digital image M of
 the first phase of the proposed system, a non-dominated sorting
dimensions H  W, such that I HW ¼ f ði; jÞ0 r i rH  1and0 r j
rW  1g, where f(i,j) denotes the intensity of the pixel at position
ði; jÞ. In our present work, only binary images of Bangla hand-
written characters and numerals are considered. For a binary
image the value of f ði; jÞ A ½0; 1. As discussed before, a 2D binary
image is assumed to be a combination of a number of overlapping/
non-overlapping regions Rk , k ¼{1, 2... n}, i.e. M ¼⋃k Rk . Here the
regions Rk , k ¼{1, 2... n} are rectangular in shape with edges par-
allel to corresponding edges of M. Thus, a region Rk is defined by
the pixel-pairs
n at o the bottom-left and top-right corners
i.e. Rk  iTL TL BR BR
k ; jk ; ik ; jk , where the pixel at (iTL TL
k ; jk ) denotes the
pixel at the top-left corner of the region and the pixel at (iBR BR
k ; jk )
denotes the pixel at the bottom-right corner of the region. Goal of
the present work is to find a subset M i D M, such that the recog-
nition cost of the character image (when described by the set of
regions M i ) is minimal and the recognition accuracy achieved by
the proposed system is maximum.
5.2. Region sampling methodology
As discussed before, the local feature set used in the present
work is comprised of longest run based features [15] computed
along 4 directions, extracted from the second level of the CG based
Fig. 3. Feature extraction techniques implemented under present work.
quad-tree partition of a sample handwritten character or digit
image. Let i denotes the partition level of a CG based quad-tree,
where i A f0; 1; 2g. Si denotes
n the
 set
n of local regions
oo at i-th level of

the quad-tree; then Si ¼ Rij  j A 0; 1…4i  1 . Clearly, a sample
image M can be defined as M ¼ ⋃j Rij . Let the set of local features
extracted
n from the oregion Rij of M is denoted by V ij , where
V ij ¼ F H V D1 D2
ij ; F ij ; F ij ; F ij . FH V D1 D2
ij ,F ij ,F ij ,F ij denote the values of the
longest run based feature extracted from the local region Rij of the
handwritten character or numeral image, along its four major axes
i.e. horizontal axis, vertical axis, principal diagonal axis-1 (south-
west diagonal) and principal diagonal axis-2 (north-east diagonal)
respectively, as shown in Fig. 8. Global feature set of M is defined
as, G ¼ ð⋃ij A f0;1g V ij Þ [ CF, where CF denotes the convex-hull based
features extracted from the entire image. Local feature set of M is
defined as L ¼ ⋃ij ¼ 2 V ij . Fitness value of a set of regions Lm  L is
denoted by f ðLm [ GÞ, where f ð:Þ denotes the recognition accuracy
by a SVM is based classifier of the sample handwritten character or
digit image, described by only the set of local regions Lm .
The present work performs region sampling in two phases. In
harmony-search algorithm (NSHA) based region sampling method
and a non-dominated sorting genetic algorithm (NSGA-II by Deb
et al. [18]) based region sampling method are employed on the
dataset separately and the non-dominated points are extracted
from both of the pareto-optimal fronts. In the final phase, a quality
consensus is performed among the extracted points. Local regions
with majority of votes (more than 50% vote share) are preserved
separately and the rest of the candidate regions are reinterpreted
as fuzzy features with the help of AFS theory [43]. The subset of
regions with the most class-separating power is selected. Finally,
the union of the set of fuzzily selected subset of local regions and
the set of previously separated regions are returned. Fig. 4 shows a
schematic diagram of the proposed system.
5.2.1. The Non-dominated Sorting Genetic Algorithm (NSGA-II) based
region sampling methodology
Since its introduction, NSGA-II [18] has been widely used in
various applications such as water network design [22], con-
struction management [23], economics [24], population planning
[25] etc. In our present work, we have investigated the efficacy of
NSGA-II for the selection of the most discriminating set of local
regions from a handwritten character or digit image, while
incurring minimum possible recognition cost. In the following
 R. Sarkhel et al. / Pattern Recognition 58 (2016) 172–189 177

Fig. 4. Schematic representation of the integrated system developed under present work.

section, a NSGA-II based region sampling methodology is intro-
duced for this purpose.
(1) Initialization: The algorithm is initialized with an empty set ß;
upon termination ß contains the non-dominated points
extracted from the pareto-optimal front. The population size
is fixed at the number of possible non-trivial subset cardinal-
ities of the local region set. In our present work, the popula-
tion size of NSGA-II based region sampling methodology is 15.
The parameters crossover-probability (pc) is set to 0.9 and
mutation-probability (pm) is set to (1/(total number of local
regions)) or 0.0625 in our present work, as suggested by Deb
et al. [18].
(2) Identification of the pareto-optimal points in search-space:
Through reproduction and natural selection, the algorithm
heuristically tries to identify the set of local regions having the
most class-separability power, while incurring minimal recog-
nition cost. The region selection method is described in details
in Appendix A as Algorithm 1.
(3) Objective functions: The algorithm has two objectives:
(1) minimization of average recognition cost for handwritten
character/digit and (2) maximization of recognition accuracy
of the test dataset, using a SVM-based classifier. The recogni-
tion cost is expressed by, (i) average per character recognition
time and (ii) number of local regions used to represent each
handwritten character/digit.
178 R. Sarkhel et al. / Pattern Recognition 58 (2016) 172–189

featureset F R extracted from V R (as discussed in Section 4.2) , f T

denotes the average per character recognition time achieved by
the proposed system and f R denotes the number of local regions
used to represent each handwritten character/digit.
(4) Termination criteria: The algorithm is terminated when the
algorithm has successfully reproduced 25 generations of the
initial population. Hence, the maximum number of iterations
(NI) of the algorithm is 25.

Pseudocode of the proposed NSGA-II based region sampling

algorithm and a detailed analysis of the algorithm, is described in
Appendix A as Algorithm 1.

5.2.2. A Non-dominated Sorting Harmony-search (NSHA) based

Fig. 5. Binary operators used in a genetic algorithm.
region sampling methodology
Harmony search is a derivative-free nature-inspired meta-
heuristics algorithm, proposed by Geem et al. [22], in the year
2001. It mimics a musician’s journey towards a better state of har-
mony. Since then, it has been used extensively by researchers for
various applications [45,46]. Several variations of multi-objective
harmony search algorithm have also been proposed by researchers
[28], but the performance of a multi-objective harmony search
algorithm for region sampling based handwritten character recog-
nition system is yet to be explored. To address this, a non-dominated
sorting harmony-search based region sampling method is proposed
in our present work. The proposed method uses a non-dominated
sorting and crowding-distance based ranking technique, introduced
by Deb et al. in [18], to extract the pareto-optimal solutions from the
solution space. The exploration (harmony memory consideration rate
or HMCR) and exploitation (pitch adjustment rate or PAR) parameters
used by this algorithm are dynamically adjusted with respect to the
current generation of the population.
In our experimental setup, a comparative study between the
proposed algorithm and a basic harmony search based region
sampling was performed on the dataset of randomly mixed
handwritten Bangla Basic and Compound characters [39,12]. Our
present work have provided a significant 14.2965% increase in
recognition accuracy and 6.25% decrease in the associated recog-
nition cost, compared to basic harmony search based region
sampling method [47].

(1) Initialization: The algorithm is initialized with an empty set ß;

upon termination ß contains the non-dominated points
extracted from the pareto-optimal front. Harmony memory size
(HMS) is fixed at the number of possible non-trivial subset
Fig. 6. Crowding distance of point P is the perimeter of the cuboid shown in cardinalities of the set of local regions. In our present work, HMS
dashed line. is set to 15. The parameters harmony memory consideration rate
(HMCR) and pitch adjustment rate (PAR) are self-adaptive. Value
To formalize, the optimization problem can be described as of HMCR at the tth population generation is denoted by HMCRt
following: and the value of PAR at the tth population generation is denoted
8 by PARt , defined as following:
< Maximize f A ðO1Þ
>
Minimize f T ðO2Þ HMCRt ¼ HMCRmin þ ðHMCRmax  HMCRmin Þ=t  NI ð5Þ
>
: Minimize f
R ðO3Þ
PARt ¼ PARmax –ðPARmax PARmin Þ=t  NI ð6Þ

with respect to the following constraints:

8 PARmax and PARmin denote the maximum and minimum values of
>
> f Z0 ðC1Þ the parameter pitch adjustment rate or PAR, whereas HMCRmax
> A
>
< f 40 ðC2Þ and HMCRmin denote the maximum and minimum values of the
T
>
> f R ¼ jV R j ðC3Þ parameter harmony memory consideration rate or HMCR. The
>
>
: 1 r f o 16 ðC4Þ values of HMCRmax and HMCRmin are set as 1.0 and 0.9 respec-
R
tively, whereas the values of PARmax and PARmin are 1.0 and
where V R denotes the region-vector representing the encoding 0.0 respectively, as suggested by Pan et al. in [48].
of local regions returned by the algorithm, f A denotes the (2) Identification of the pareto-optimal points in search-space: The
recognition accuracy achieved by the SVM based classifier using algorithm heuristically searches for the most informative set
the features extracted from the global featureset G and the local of local regions that will incur minimal recognition cost. The
 R. Sarkhel et al. / Pattern Recognition 58 (2016) 172–189
Fig. 7. Analogy of musical improvisation and optimization technique in Harmony Search Method.
Fig. 8. Directional vector model of a region in residual regions-space.
region selection method employed under present work is
described in details in Appendix A as Algorithm 2.
(3) Objective functions: The algorithm has two objectives:
(1) minimization of average recognition cost for handwritten
character/digit and (2) maximization of recognition accuracy
of the test dataset, using a SVM-based classifier. The recogni-
tion cost is expressed by, (i) average per character recognition
time and (ii) number of local regions used to represent each
handwritten character/digit, same as the NSGA-II based region
sampling method described before. Therefore, the optimiza-
tion problem can also be formalized same as the NSGA-II
based region sampling algorithm, described before.
(4) Termination criteria: The algorithm is terminated when the
algorithm has successfully reproduced 25 generations of the
initial population. Hence, the maximum number of iterations
(NI) of the algorithm is 25. Pseudocode and a detailed analysis
of the proposed NSHA based region sampling algorithm is
described in Appendix A as Algorithm 2.
Both of the algorithms proposed in our experimental setup,
have been extensively tested for a dataset of randomly mixed
handwritten Bangla Basic and Compound characters and a dataset
of Bangla numerals separately. The results of the experiments are
shown in Section 5. It is to be noted that both of the algorithms
discussed above are developed independent of any specific dataset
and can be successfully applied for any region sampling based
pattern recognition task.
179
5.3. Integrating the pareto-optimal solutions using AFS theory based
fuzzy logic
Applying NSHA and NSGA-II on a test dataset of handwritten
Bangla characters/numerals (discussed in 5.2.1 and 5.2.2) produce
two pareto-optimal sets of local region subsets. Let, the pareto-
optimal set returned by NSHA is denoted by ß1 and the pareto-
optimal set returned by NSGA-II is denoted by ß2. The present
work uses a novel region sampling method to objectively choose
the most discriminative set of regions, using the information
gathered from ß1 and ß2. This is one of the major contributions of
the present work.
It has been proved in [21] that elitism helps accelerate the
convergence rate of a multi-objective evolutionary algorithm. To
choose a solution from set of candidate solutions in both ß1 and ß2,
a quality consensus is performed. The regions with majority con-
sensus are therefore excluded from the region-space from further
considerations and maintained separately henceforth. Let, this set
of regions is denoted by E. The resultant region space VR forms the
vocabulary for fuzzy concepts and semantics; it is called residual
region-space. Clearly, VR ¼Ec.
As previously discussed in Section 3, one of the main motiva-
tions of using fuzzy logic is better interpretability of complex
concepts to model the outstanding human ability of decision
making with imprecise understanding or lack of background
knowledge. In the final phase of the proposed method, the residual
180 R. Sarkhel et al. / Pattern Recognition 58 (2016) 172–189

region-space is reinterpreted as a fuzzy feature set. The steps are
discussed in the following section.
5.3.1. Scalarization of the residual feature set
As discussed earlier in Section 4.2.2, longest run features are
extracted along 4 axes from each local region. Considering a unit
vector along the direction of each of the four axes, h(horizontal), v
(vertical), d1(principal diagonal 1) and d2(principal diagonal 2), each
region can be represented as a vector of the longest run of black
pixels (shown in Fig. 3b) along each of the four directions. The first
step of final phase of the proposed method is scalarization of the
local feature set extracted from each region. Let the scalarized feature
set is denoted by S. The proposed directional vector model of longest
run based features, used in the present work is shown in Fig. 8.
In our present work, scalarization of local regions is done by
representing it as a vector of directional components.
Example:
Suppose, the feature set extracted from a local region R is given
below:
Vertical Horizontal Principal Diagonal 1 Principal Diagonal 2
A B C D
Using the proposed vector model the region R can be repre-
sented as
R ¼ Av þ Bh þ Cd1 þ Dd2
Using vector decomposition, (7) can be rewritten as
R ¼ Av þ Bh þ Cðhsin45ο –vcos45ο Þ þ Dðhsin45ο þ vcos45ο Þ
yields
⟹ ð8Þ
R ¼ ðA–Ccos45 ο þDcos45ο Þv þ ðB þ Csin45 ο þ Dsin45ο Þh
Hence, the local region R can
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðA  C cos 451 þ D cos 451Þ2 þ ðB þ C sin 451 þ D sin 451Þ2 : In the fea-
ture set S, there will be a scalarized feature against each of the
regions of the residual region-space.
5.3.2. Defining the fuzzy feature set
The fuzzy features are defined such that it can reflect the
proximity of a sample character image to its true class with respect
to all the other possible classes. For a feature f A S, it is represented
as fuzzy feature uf . Value of uf for a handwritten character/digit
image, x A X is denoted by uxf . It is defined as following:
 
spreadðclassx Þ
xf  meanclass
f
x
x
uf ¼ P    vf ð9Þ
spreadðiÞ
iAC xf  meanif
where xf is the value of the feature f for element x, classx is the true
class of x and C denotes the set of all possible classes, meanif is
mean of the values of the feature f in class i, vf is the vote share of
the local region corresponding to the feature f in the previously
performed quality consensus on the pareto-optimal set extracted
from both NSGA-II and NSHA and spread(i) denotes the spread of
the elements in class i (standard deviation is used for this measure
in the present work).
It has been mentioned in the literature [49] that associating too
fuzzy feature set UF and the scalarized feature set S are same, as
the mapping between UF and S is one to one.
In a system with more than two elements, understanding the
overall interaction between a set of elements is best indicated by
investigating the interaction between a pair of elements and then
their combination with other elements. In a similar way, in our
present work, a subset of features is selected by investigating the
interaction between a pair of features first and then combining
them with other features such that the resultant subset of features
with best overall interaction between them is returned.
5.3.3. Computing the class-separation power of a set of fuzzy
features
Using AFS theory [43], the membership function of a pair of
fuzzy features u1 ; u2 A UF can be calculated by computing θu1 u2 (x)
(as shown in Section 3). Membership function denotes the
apparent class-separation power of the combination of the two
features u1 ; u2 . Another metric for class-separation is introduced in
the present work, denoted by normðu1 ; u2 Þ. It is defined in the
following way:
P
xAX θu1 u2 ðxÞ
normðu1 ; u2 Þ ¼ ð10Þ
j Xj
ð7Þ 5.3.4. Selecting the set of regions with the most class-separating
power
To consider all possible combination of local regions, a 1616
matrix M is formed (number of local regions in our experimental
setup is 16), where each row and column corresponds to a local region
from the residual region-space. M[i,j] denotes the combined class-
separation power of the i-th and j-th feature from the fuzzy feature set
UF, represented by the metric defined in (10). The diagonal elements
be scalarized
to:
are made to zero to nullify the effect of combining i-th feature with
itself. The task of selecting the most informative set of local regions
now reduces to finding the maximum sum k  k non-trivial sub-
matrix of M. The sub-matrix is formed such that, if the i-th row is
selected, then the i-th column also has to be selected [50], which
essentially means that the i-th region is being considered. The k  k
maximum sum sub-matrix returns the best k features from the fuzzy
featureset. Now, as there is a one to one mapping between the original
featureset and the fuzzy featureset, this helps us to identify the k best
features from the residual featureset. Each feature in the residual fea-
tureset has a one to one mapping with a local region in the residual
region-space, therefore we get the k best local regions from the resi-
dual region-space at the end of this method. A pseudocode of the
algorithm described above, is described in Appendix A as Algorithm 3.
Finally, the union of the set of k selected regions and the elitist
set of regions E, which was previously separated and maintained is
returned as the best candidate solution.
Table 2
Values of the parameters used in our experimental framework.
Region sampling Parameter values used in the present work
methodology

many fuzzy sets and rules can make the fuzzy model hard to NSHA [17] based region HMCRmax HMCRmin PARmax PARmin POP NI
interpret and thus defeating the purpose of fuzzifying the original sampling 1.0 0.9 1.0 0.0 15 25
featureset in the first place. In our present work, a one to one methodology pm pc POP NI
mapping is maintained between a feature and its corresponding NSGA-II [18] based 0.0625 0.9 15 25
region sampling
fuzzy feature, therefore the dimensionality of fuzzy featureset
methodology
does not increase than the original featureset. Cardinalities of the
 R. Sarkhel et al. / Pattern Recognition 58 (2016) 172–189
Fig. 9. Pareto-optimal surface for the dataset (DB4) of randomly mixed handwritten Bangla Basic and Compound characters.
Fig. 10. Pareto-optimal surface for the dataset (DB2) of handwritten Bangla digits.
6. Experimental results
Objectives of the present work are: (a) increasing the recog-
nition accuracy and (b) decreasing the corresponding recognition
cost for the recognition of isolated Bangla handwritten characters.
The recognition cost is represented in our experimental setup as,
(i) average per character recognition time and (ii) number of local
regions used to represent each handwritten digit/character. The
integrated system shown in Fig. 4 is implemented for the experi-
mental setup. As discussed in Table 1 of Section 4.1, performance
of the system is evaluated on four separate datasets, (a) a dataset
of handwritten Bangla numerals (DB1), (b) a dataset of hand-
written Bangla numerals (DB2), (c) a dataset of handwritten
Bangla Basic characters (DB3), and (d) a randomly mixed dataset of
handwritten Bangla Basic and Compound characters (DB4). In our
experimental setup, LIBSVM [51], an open-source tool based on
SVM [52] is used for classification tasks. Values of the different
parameters used in our experimental setup is described in Table 2.
The experiments are performed on an Intel(R) Core(TM) i5 pro-
cessor, with 3.10 GHz clock-frequency and 8 GB RAM.
As shown in Fig. 4, in the first phase of our experimental setup,
pareto-optimal solutions are extracted from the final population of
the evolutionary algorithms (NSHA and NSGA-II). These solutions
are plotted as an ordered triplet of [recognition accuracy, average
per character recognition time, number of local regions used to
describe each handwritten character/digit] over the solution-space.
Figs. 9 and 10 demonstrate the resultant representative pareto-
181
optimal surfaces for two of the larger datasets (DB2 and DB4 in
Table 1) used in our experimental setup.
Finally, in the second phase, a quality consensus is performed
among the regions extracted from the pareto-surfaces and the
regions with majority consensus are separated. The minority
regions form the residual region-space. The local regions can now
be selected from the residual region-space, based on their class-
separability power, using AFS theory based fuzzy logic (as dis-
cussed in Section 5.3), such that the recognition accuracy achieved
by the proposed system is maximized.
A comparative analysis of the maximum recognition accuracies
achieved by employing NSHA, NSGA-II separately on all of the
datasets used in the present work and their fuzzy recombination is
shown in Fig. 11.
The final results returned by the system are shown in Tables 3
and 4. Dataset indexes are same as mentioned in Table 1. From the
results provided in Tables 3 and 4, it is clear that the proposed
method provides a substantial increase in recognition accuracy
and decrease in recognition cost for all of the datasets. For the
dataset of randomly mixed Bangla Basic and Compound char-
acters, per character average recognition time taken by the clas-
sifier has slightly increased. This may be attributed to greater
number of support vectors needed to be generated to define the
Maximal Marginal Hyperplane [52] for the dataset. This apparent
contradiction serves the purpose of highlighting the difficulties
involved in handwritten character recognition. However, overall
performance of the proposed system on the datasets used in our
present experimental setup is proved to be good.
182 R. Sarkhel et al. / Pattern Recognition 58 (2016) 172–189

Comparison of recognition accuracy(%) achieved for

the dataset (DB1) of handwritten Bangla digits

Recognition Accuracy (%)

97.8
97.6
97.4
97.2
97
Handwritten Bangla digits
NSHA 97.3
NSGA-II 97.3
Present work 97.8

NSHA NSGA-II Present work

Comparison of recognition accuracy(%) achieved for the

dataset (DB2) of handwritten Bangla digits
Recognition Accuracy (%)

98.3
98.2
98.1
98
Handwritten Bangla
digits
NSHA 98.1
NSGA-II 98.15
Present work 98.23
NSHA NSGA-II Present work

Comparison of recognition accuracy(%) achieved for the

Recognition Accuracy (%)

dataset (DB3) of hanwritten Bangla Basic characters

87.5
87
86.5
86
85.5
Handwritten Bangla
characters
NSHA 86.3095
NSGA-II 86.6363
Present work 87.28

NSHA NSGA-II Present work

Comparison of recognition accuracy(%) achieved for the

dataset (DB4) of randomly mixed hanwritten Bangla Basic
and Compound characters
Recognition Accuracy (%)

86.8
86.6
86.4
86.2
86
Handwritten Bangla characters
NSHA 86.3095
NSGA-II 86.6363
Present work 86.6478

NSHA NSGA-II Present work

Fig. 11. A comparative analysis of the maximum recognition accuracy achieved the proposed system.
 R. Sarkhel et al. / Pattern Recognition 58 (2016) 172–189 183

Table 3
Final results on the dataset of handwritten Bangla numerals.

Table 3.1

Dataset Index Dataset type Methodology Recognition Accuracy (%) by Average per character recogni- Number of local regions
SVM classifier tion time (ms) used
(A) (B) (C)

DB1 Isolated handwritten Bangla The Present work 97.8 16.531 11

numerals Without region sampling 97.15 17.57 16
Improvement using present 0.67 5.913 31.25
work (%)

Table 3.2

Dataset Index Dataset type Methodology Recognition Accuracy (%) by Average per character recogni- Number of local regions
SVM classifier tion time (ms) used
(A) (B) (C)

DB2 Isolated handwritten Bangla The Present work 98.23 30.6872 14

numerals Without region sampling 97.5 30.72 16
Improvement using present 0.73 0.107 12.5
work (%)

Table 4
Final results on the dataset of handwritten Bangla characters.

Table 4.1

Dataset Index Dataset type Methodology Recognition Accuracy (%) by Average per character Number of local
SVM classifier recognition time (ms) regions used
(A) (B) (C)

DB3 Isolated handwritten Bangla Basic The present work 87.28 28.96 14
characters Without region sampling 84.5750 30.02 16
Improvement using present 3.1983 3.531 12.5
work (%)

Table 4.2

Dataset Index Dataset type Methodology Recognition Accuracy (%) by Average per character Number of local
SVM classifier recognition time (ms) regions used
(A) (B) (C)

DB4 Randomly mixed isolated handwritten The present work 86.6478 90.6675 15
Bangla Basic and Compound characters Without region sampling 85.33 85.52 16
Improvement using present 1.3178  6.019 6.25
work (%)

A brief comparative analysis of the proposed method with
some of the contemporary methods present in the literature is
presented in Table 5. Competing methods are tested on the same
datasets [12,38], used in our experimental setup and their per-
formance is compared based on two evaluation metrics: (a) the
maximum recognition accuracy achieved by the method after
repeating the experiment for 25 times and (b) the reduction in
recognition cost using the method. In our experimental setup,
recognition accuracy of a competing method is computed by using
a SVM based classifier whereas the reduction in recognition cost is
computed as the sum of average reduction in per character
recognition time and the average reduction in the number of local
regions needed to represent each character. The local regions used
to represent each character are produced by a two-level CG based
quad-tree partitioning [41]. The proposed method achieved
(1) 97.8% recognition accuracy and 37.163% reduction in recogni-
tion cost, (2) 98.23% recognition accuracy and 12.61% reduction in
recognition cost, (3) 87.28% recognition accuracy and 16.031%
reduction in recognition accuracy & (4) 86.6478% recognition
accuracy and 0.231% reduction in recognition cost for the datasets
DB1, DB2, DB3 and DB4 respectively.
To establish the superiority of the method proposed in the present
work, statistical significance tests are performed on the results shown
in Table 5. In our experimental setup, paired t-tests at 5% significance
level have been performed to establish statistical significance of the
superior performance shown by the proposed method. Results of the
paired t-tests are shown in Table 6. Indexes are same as in Table 5. The
results shown in Table 6 emphatically prove that superior perfor-
mance provided by the present work is statistically significant for
9 out of 10 competing methods in case of Bangla characters and 10 out
of 11 competing methods in case of Bangla digits.
In spite of the various precautions taken to disambiguate
similarly shaped handwritten Bangla characters, due to various
writing styles and/or complex shape of different Bangla characters
and digits, some of the characters and digits in the test dataset
were misclassified by the proposed methodology, which highlights
the challenges inherent in handwritten character recognition.
Fig. 12 shows some of the correctly classified and misclassified
characters by the proposed method.
184 R. Sarkhel et al. / Pattern Recognition 58 (2016) 172–189

Table 5
Comparative analysis of the proposed method with contemporaries.

Table 5.1

index Work reference Database Type of data Classification Scheme Recognition accuracy Reduction in recognition
(%) cost (%)

A1 Hassan et al. [53] DB1 Isolated handwritten Bangla numerals SRC 94.00 0
A2 Hassan et al. [54] KNN 96.70 0
A3 Das et al. [55] SVM 97.70 0
A4 Das et al. [15] SVM 97.70 3.9
A5 Roy et al. [56] SVM 93.30 0
A6 Xu et al. [57] Bayesian Network 87.50 0
A7 Present work SVM 97.80 37.163
B1 Wen et al. [58] DB2 Isolated handwritten Bangla numerals SVM 75.05 0
B2 Bhattacharya et al. SVM 84.50  170.11
[34]
B3 Basu et al. [42] MLP 67.70 0
B4 Das et al. [15] SVM 80.58  100
B5 Roy et al. [56] SVM 87.26  846.59
B6 Present work SVM 98.23 12.61

Table 5.2

Index Work reference Database Type of data Classification Scheme Recognition accuracy Reduction in recognition
(%) cost (%)

C1 Basu et al. [59] DB3 Isolated handwritten Bangla Basic characters SVM 80.58  200
C2 Roy et al. [60] SVM 86.40 13.52
C3 Das et al. [61] SVM 82.27 0
C4 Bhowmick et al. [62] MLP 84.23 0
C5 Sarkhel et al. [47] SVM 86.533 8.583
C6 Present work SVM 87.28 16.031
D1 Das et al. [61] DB4 Randomly mixed, isolated handwritten Bangla SVM 75.05 0
D2 Roy et al. [60] Basic and Compound characters SVM 84.50  170.11
D3 Basu et al. [14] MLP 67.70 0
D4 Basu et al. [59] SVM 80.58  100
D5 Das et al. [10] SVM 86.96  846.59
D6 Present work SVM 86.6478 0.231

Table 6
Results of paired t-test at 5% significance level.

Table 6.1

Index Database Type of data p-Value t-Value H r H0

A1 DB1 Bangla digits 0.0348 5.2197 Yes

A2 0.0472 4.4388 Yes
A3 0.4597 0.7454 No
A4 0.0009 3.5355 Yes
A5 0.0001 18.6852 Yes
A6 0.0012 3.4333 Yes
B1 DB2 Bangla digits 0.0002 66.65 Yes
B2 0.0001 24104 Yes
B3 0.0004 47.44 Yes
B4 0.0448 2.212 Yes
B5 0.0114 9.297 Yes

Table 6.2

Index Database Type of data p-Value t-Value H r H0

C1 DB3 Bangla Basic characters 0.0310 2.2222 Yes

C2 0.0077 2.7828 Yes
C3 0.0168 2.4773 Yes
C4 0.0001 6.1000 Yes
C5 0.0492 2.0182 Yes
D1 DB4 Randomly mixed Bangla 0.0112 9.361 Yes
D2 Basic and Compound 0.0015 25.52 Yes
D3 characters 0.0122 8.954 Yes
D4 0.0801 3.209 No
D5 0.0001 139.4 Yes
 R. Sarkhel et al. / Pattern Recognition 58 (2016) 172–189 185

Fig. 12. Samples of correctly classified and misclassified characters by the present work.

dataset of handwritten Bangla digits [38] respectively. Both of the

Fig. 13. Samples of broken and degraded Bangla handwritten characters and digits
which were correctly classified by the present work.
It is to be noted that our present work can successfully handle
broken and degraded characters. Fig. 13 shows some of such char-
acters present in the datasets that have been used in our experiment.
All these samples were successfully recognized by our proposed
method, despite the apparent discontinuities present in the shapes of
the characters. The reason behind this may be attributed to the
similarities in the sub-images generated in the first and second level
of depths of quad-tree partitioning of the handwritten character/digit
image. Interested readers can find more details regarding this in [10].
multi-objective evolutionary region sampling algorithms used in
the experimental setup, have been developed independent of any
particular dataset. The results extracted from the pareto-optimal
surfaces of both the algorithms are fuzzily combined to return the
best compromise solution. The multi-objective region sampling
algorithms and the AFS theory based fuzzy region selection
methodology, proposed under the current work, mark one of the
contributions of this paper. Also to the best of our knowledge, this
is the first work which tries to find a trade-off between max-
imizing the achieved recognition accuracy and minimizing the
associated recognition cost for a handwritten character recogni-
tion system. The experimental results have shown great promise
in this approach. It opens up a new frontier of developing efficient
optical character recognition systems, which can be successfully
used in various real-life applications.

7. Conclusion
A novel region sampling method based on multi-objective
evolutionary algorithms is presented here. The proposed method
tries to identify the set of most discriminative local regions while
incurring minimal recognition cost. A non-dominated sorting
harmony search based region sampling method and a non-
dominated sorting genetic algorithm based region sampling
methodology have been employed in our experimental setup. It
provided 1.3178% and 0.73% increase in recognition accuracy,
while decreasing the associated recognition cost by 12.607% and
0.234% for a dataset of handwritten Bangla characters [35] and a
Conflict of interest
None declared.
Acknowledgment
The authors are thankful to CVPR unit, ISI, Kolkata for providing
the dataset for the experiment. The authors are also thankful to
“Centre for Microprocessor Application for Training Education and
Research” and Department of Computer Science & Engineering,
Jadavpur University, Kolkata for kindly providing the resources and
infrastructural facilities that helped to complete this work.

Appendix A

To describe the algorithms in more details, pseudocodes of the proposed algorithms developed under our present work is described here. In
this section, the pseudocodes of proposed NSGA-II [18] based region sampling methodology, the NSHA [17] based region sampling methodology
and the AFS theory based local region selection methodology is described under Algorithm 1, Algorithm 2 and Algorithm 3 respectively.
Algorithm 1. A non-dominated sorting genetic algorithm based
region sampling methodology.

Input: Initial feature set extracted from the Bangla handwritten character/digit database.
Output: A set of non-dominated points extracted from the pareto-optimal front.
Initialize the parameters: population size (POP), crossover-probability (pc), mutation-probability (pm), maximum number of iterations (NI)

Begin
1. ß ¼ Φ /* ß denotes the pareto-optimal set */
2. for (i ¼ 1 to NI){
186 R. Sarkhel et al. / Pattern Recognition 58 (2016) 172–189

3. C ¼β /* C denotes population of the current generation, represented by a set of ordered pairs (recognition accuracy,
recognition cost) */
4. for ( j ¼ 1 to POP-1){ /*POP denotes the maximum possible population size*/
5. if (population is empty){
6. Initialize the population with a set S containing 2 sets ofjrandomly selected local regions.
7. C ¼ C[S
8. }
9. Generate a uniform random number r1 ϵ (0, 1).
10. if (r1 r pc ){ // pc denotes the crossover probability
11. Randomly select two parents from the current generation.
12. /* These two parents constitute the mating pool. */
13. Randomly select a crossover point and perform single-point crossover operation. If the two offspring generated by this
operation are denoted by O1 and O2.
14. Update population: C ¼ C [ O1 [ O2
15. }
16. Generate a uniform random number r2 ϵ (0, 1).
17. if (r2 r pm ){ // pm denotes the mutation probability
18. Randomly select a parent from the current generation. Make a copy of this parent for the mutation operation. Perform
mutation on this parent (shown in Fig. 5).
19. Let, the mutated parent is denoted by MP.
20. C ¼ C [ MP
21. }
22. }
23. for the population in C{
24. Perform fast non-dominated sort and assign non-domination rank, as suggested by Deb et al. [16]. For a parent p A C, non-
domination rank of p is the number of individuals in C that dominates p.
25. To preserve diversity among the pareto-optimal points, a metric called the crowding- distance (shown in Fig. 6) is also
integrated, when trying to find the non-dominated fronts.
26. Using these two metrics i.e. non-domination ranks and crowding-distance, non-dominated fronts are computed based on a
partial order relation denoted by ! n . For any p, q A C, p ! n q implies [(prank o qrank) or ((prank ¼ qrank) and (pdistance 4qdistance))]. The
points of the pareto-optimal front are represented by the set ß.
27. }
28.
29. return ß. /* ß contains the pareto-optimal set of points */
End

Analysis of the algorithm

Using the meta-heuristics searching ability of the algorithm, the pareto-optimal front produces several alternate solutions to the user
to choose from. Time complexity analysis of the algorithm for each iteration, is as following:
 
1) Time complexity for non-dominated sorting is: O M ð2N Þ2 ;
2) Time required for crowding-distance assignment is:
OðM ð2N Þlog ð2N ÞÞ;
3) Sorting on ! n requires time of the order of: Oð2Nlog ð2NÞÞ:
 
In aggregate, total time required for each iteration of the algorithm is of the order of: O MN2 , where N is the total number of local
regions in the local feature set and M is the number of objectives to be optimized. This is a significant reduction from experimenting with
all the possible combinations of local regions, which would have taken time of the order of Oð2N Þ.

Algorithm 2. A self-adaptive non-dominated sorting harmony-

search region sampling methodology.

Input: Initial feature set extracted from the Bangla handwritten character/digit database.
Output: A set of non-dominated points extracted from the pareto-optimal front.
Initialize the parameters: harmony memory size (HMS), minimum value of HMCR (HMCRmin), maximum value of HMCR (HMCRmax),
minimum value of PAR (PARmin), maximum value of PAR (PARmax), maximum number of iterations (NI).
Begin
1. ß ¼ Φ /* ß denotes the pareto-optimal set */
2. for (i ¼ 1 to NI ){
3.C ¼ ß /* C denotes population of the current generation, represented by a set of ordered pairs (recognition
accuracy, recognition cost) */
4. for ( j ¼ 1 to HMS-1){ /* HMS denotes the maximum possible population size */
5. if (population is empty){
6. Initialize the population with a set S containing 2 sets of j randomly selected local regions.
 R. Sarkhel et al. / Pattern Recognition 58 (2016) 172–189 187

7. C ¼ C[S
8. }
9. Generate a uniform random number r1 ϵ (0, 1).
10. if (r1 r HMCRi ){ /* The value of HMCRi is set by (5) */
11. Randomly select a parent from the current generation. Make a copy of this parent for pitch adjustment.
12. Generate a uniform random number r2 ϵ (0, 1).
13. if (r2 r PARi ){ /* The value of PARi is set by (6) */
14. Randomly select a region from the previously chosen parent i.e. a note from a harmony.
15. Replace the selected region with a region selected randomly from the set of local regions that are not included in the
chosen parent.
16. Let the pitch adjusted parent is denoted by P.
17. C ¼ C⋃P
18. }
19. }
20. else {
21. Initialize a parent T with j randomly selected regions such that, T 2= C.
22. C ¼ C⋃T
23. }
24. }
25. for the population in C {
26. Perform fast non-dominated sort and assign non-domination rank, as suggested by Deb et al. [16]. For a parent p A C, non-
domination rank of p is the number of individuals in C that dominates p.
27. To preserve diversity among the pareto-optimal points, a metric called the crowding-distance (shown in Fig. 6) is also inte-
grated, when trying to find the non-dominated fronts.
28. Using these two metrics i.e. non-domination ranks and crowding-distance, non-dominated fronts are computed based on a partial
order relation denoted by ! n . For any p, q A C, p ! n q implies [(prank o qrank) or ((prank ¼ qrank) and (pdistance 4qdistance))]. The points of the
pareto-optimal front are represented by the set ß.
29. }
30. }
31. return ß. /* ß contains the pareto-optimal set of points */
End

Analysis of the algorithm

The proposed method searches for the most informative set of local regions in the search space of (recognition accuracy, recognition
cost) and returns the pareto-optimal set. Time complexity analysis of the algorithm for each iteration, is given as following:
 
1) Time complexity for non-dominated sorting is: O M ð2N Þ2 ;
2) Time required for crowding-distance assignment is:
OðM ð2N Þlog ð2N ÞÞ;
3) Sorting on ! n requires time of the order of: Oð2Nlog ð2NÞÞ:
 
In aggregate, total time required for each iteration of the algorithm is of the order of: O MN2 , where N is the total number of local
regions in the local feature set and M is the number of objectives to be optimized. This is a significant reduction from experimenting with
all the possible combinations of local regions, which would have taken time of the order of Oð2N  1 Þ.

Algorithm 3. An AFS theory based feature selection technique

from the residual featureset.

Input: The set of training samples X, original featureset F.

Output: The most discriminating subset of features from the residual featureset.
1 for x A X // X is the set of training samples
2 Calculate uxf according toxf ; // f A F(the original set of features) and
// uxf denotes the fuzzy features for the
// element x, as shown in Eq. (9)
3 for ui ; uj A UF // UF denotes the fuzzy featureset
4 Calculateθu1 u2 ðxÞ; // As shown in Eq. (7)
5 for ui ; uj A UF
6 Calculatenormðu1 ; u2 Þ; // As shown in Eq. (10)
7 Create matrix M;
8 Find maximum sum k  k sub-matrix from M; // k  k sub-matrix is chosen from M
// that if i-th row is selected, i-th column
// will also have to be selected.
188 R. Sarkhel et al. / Pattern Recognition 58 (2016) 172–189
References
[1] A. Ul-Hasan, S. Bin Ahmed, F. Rashid, F. Shafait, T.M. Breuel, Offline printed
Urdu Nastaleeq script recognition with bidirectional LSTM networks, in: Pro-
ceedings of the 2013 12th International Conference on Document Analysis and
Recognition, 2013, pp. 1061–1065.
[2] A. Ray, S. Rajeswar, S. Chaudhury, Text recognition using deep BLSTM net-
works, in: Proceedings of the 2015 Eighth International Conference on
Advances in Pattern Recognition (ICAPR), 2015, pp. 1–6.
[3] A. Park, Offline text recognition without intraword character segmentation
based on two-dimensional low frequency discrete Fourier transforms, 24-
May-1994.
[4] A. Vinciarelli, S. Bengio, H. Bunke, Offline recognition of unconstrained
handwritten texts using HMMs and statistical language models, IEEE Trans.
Pattern Anal. Mach. Intell. 26 (6) (2004) 709–720.
[5] H. Fujisawa, Forty years of research in character and document recognition—
an industrial perspective, Pattern Recognit. 41 (8) (2008) 2435–2446.
[6] R.M. Bozinovic, S.N. Srihari, Off-line cursive script word recognition, IEEE
Trans. Pattern Anal. Mach. Intell. 11 (1) (1989) 68–83.
[7] P.-K. Wong, C. Chan, Off-line handwritten Chinese character recognition as a
compound Bayes decision problem, IEEE Trans. Pattern Anal. Mach. Intell. 20
(9) (1998) 1016–1023.
[8] J. Cao, M. Ahmadi, M. Shridhar, Recognition of handwritten numerals with
multiple feature and multistage classifier, Pattern Recognit. 28 (2) (1995)
153–160.
[9] S.V. Rajashekararadhya, P.V. Ranjan, Efficient zone based feature extraction
algorithm for handwritten numeral recognition of four popular south indian
scripts, J. Theor. Appl. Inf. Technol. 4 (12) (2008) 1171–1181.
[10] N. Das, R. Sarkar, S. Basu, P.K. Saha, M. Kundu, M. Nasipuri, Handwritten Bangla
character recognition using a soft computing paradigm embedded in two pass
approach, Pattern Recognit. 48 (6) (2014) 2054–2071.
[11] U. Pal, B.B. Chaudhuri, Indian script character recognition: a survey, Pattern
Recognit. 37 (9) (2004) 1887–1899.
[12] R. Sarkar, N. Das, S. Basu, M. Kundu, M. Nasipuri, D.K. Basu, CMATERdb1: a
database of unconstrained handwritten Bangla and Bangla–English mixed
script document image, Int. J. Doc. Anal. Recognit. 15 (1) (2011) 71–83.
[13] D. Impedovo, G. Pirlo, Zoning methods for handwritten character recognition:
a survey, Pattern Recognit 47 (3) (2014) 969–981.
[14] S. Basu, N. Das, R. Sarkar, M. Kundu, M. Nasipuri, D.K. Basu, Handwritten
‘Bangla’ alphabet recognition using an MLP based classfier, in: Proceedings of
the 2nd National Conference on Computer Processing of Bangla—2005, 2005,
pp. 285–291.
[15] N. Das, R. Sarkar, S. Basu, M. Kundu, M. Nasipuri, D.K. Basu, A genetic algorithm
based region sampling for selection of local features in handwritten digit
recognition application, Appl. Soft Comput. 12 (5) (2012) 1592–1606.
[16] K. Deb, Multi-objective optimization using evolutionary algorithms: an
introduction, 2011, pp. 1–24.
[17] S. Sivasubramani, K.S.S. Swarup, Multi-objective harmony search algorithm for
optimal power flow problem, Int. J. Electr. Power Energy Syst. 33 (3) (2011)
745–752.
[18] K. Deb, A. Member, A. Pratap, S. Agarwal, T. Meyarivan, A fast and elitist
multiobjective genetic algorithm, IEEE Trans. Evol. Comput. 6 (2) (2002)
182–197.
[19] L. Xiaodong, The fuzzy theory based on AFS algebras and AFS structure, J.
Math. Anal. Appl. 217 (2) (1998) 459–478.
[20] V. Chankong, Y.Y. Haimes, Multiobjective Decision Making: Theory and
Methodology, North Holland, New York, 1983.
[21] E. Zitzler, K. Deb, L. Thiele, Comparison of multiobjective evolutionary algo-
rithms: empirical results, Evol. Comput. 8 (2) (2000) 173–195.
[22] M. Atiquzzaman, S.-Y. Liong, X. Yu, Alternative decision making in water dis-
tribution network with NSGA-II, J. Water Resour. Plan. Manag. 132 (2) (2006)
122–126.
[23] E. Fallah-Mehdipour, O. Bozorg Haddad, M.M. Rezapour Tabari, M.A. Mariño,
Extraction of decision alternatives in construction management projects:
application and adaptation of NSGA-II and MOPSO, Expert Syst. Appl. 39 (3)
(2012) 2794–2803.
[24] S. Dhanalakshmi, S. Kannan, K. Mahadevan, S. Baskar, Application of modified
NSGA-II algorithm to combined economic and emission dispatch problem, Int.
J. Electr. Power Energy Syst. 33 (4) (2011) 992–1002.
[25] S. Kannan, S. Baskar, J.D. McCalley, P. Murugan, Application of NSGA-II algo-
rithm to generation expansion planning, IEEE Trans. Power Syst. 24 (1) (2009)
454–461.
[26] K.S. Lee, Z.W. Geem, A new meta-heuristic algorithm for continuous engi-
neering optimization: harmony search theory and practice, Comput. Methods
Appl. Mech. Eng. 194 (36–38) (2005) 3902–3933.
[27] X.S. Yang, Harmony search as a metaheuristic algorithm, Stud. Comput. Intell.
191 (2009) 1–14.
[28] J.R. Germ, Multiobjective Harmony Search Algorithm Proposals, vol. 281, 2011,
pp. 51–67.
[29] S. Sivasubramani, K.S. Swarup, Environmental/economic dispatch using multi-
objective harmony search algorithm, Electr. Power Syst. Res. 81 (9) (2011)
1778–1785.
[30] I. Tsamardinos, C.F. Aliferis. Towards principled feature selection: relevancy,
filters, and wrappers, in: Proceedings of the Ninth International Workshop on
Artificial Intelligence and Statistics, 2003.
[31] M. Hanmandlu, A.V. Nath, A.C. Mishra, V.K. Madasu, Fuzzy Model Based
Recognition of Handwritten Hindi Numerals using Bacterial Foraging, in:
Proceedings of the 6th IEEE/ACIS International Conference on Computer and
Information Science (ICIS 2007), 2007, pp. 309–314.
[32] I. Guyon, A. Elisseeff, An introduction to variable and feature selection, J. Mach.
Learn. Res. 3 (2003) 1157–1182.
[33] M. Ramze Rezaee, B. Goedhart, B.P.F. Lelieveldt, J.H.C. Reiber, Fuzzy feature
selection, Pattern Recognit. 32 (12) (1999) 2011–2019.
[34] U. Bhattacharya, B.B. Chaudhuri, Handwritten numeral databases of Indian
scripts and multistage recognition of mixed numerals, IEEE Trans. Pattern
Anal. Mach. Intell. 31 (3) (2009) 444–457.
[35] cmaterdb – CMATERdb: The pattern recognition database repository – Google
Project Hosting. [Online]. Available: https://code.google.com/p/cmaterdb/.
(accessed: 31.01.15).
[36] R. Sarkar, N. Das, S. Basu, M. Kundu, M. Nasipuri, CMATERdb1 : a database of
unconstrained handwritten Bangla and Bangla – English mixed script docu-
ment image, 2012, pp. 71–83.
[37] S. K. Parui, K. Guin, U. Bhattacharya, B.B. Chaudhuri, Online Handwritten
Bangla Character Recognition Using HMM 3. Analysis of strokes in hand-
written, IEEE, 2008, pp. 1–4.
[38] ISI Image Databases of Handwritten Isolated Bangla Numerals. [Online].
Available: 〈http://www.isical.ac.in/  ujjwal/download/BanglaNumeral.html〉.
(Accessed: 20.04.15).
[39] CMATERdb3.1.3.3 – cmaterdb – Handwritten Bangla Compound character
image database – CMATERdb: The pattern recognition database repository –
Google Project Hosting.
[40] N. Das, S. Basu, R. Sarkar, M. Kundu, M. Nasipuri, D. Kumar Basu, An improved
feature descriptor for recognition of handwritten Bangla alphabet, in: Pro-
ceedings of the ICSIP 2009, 2009, pp. 451–454.
[41] S. Basu, N. Das, R. Sarkar, M. Kundu, M. Nasipuri, D. Kumar Basu, A novel
framework for automatic sorting of postal documents with multi-script
address blocks, Pattern Recognit. 43 (10) (2010) 3507–3521.
[42] S. Basu, N. Das, R. Sarkar, M. Kundu, M. Nasipuri, D. K. Basu, Recognition of
numeric postal codes from multi-script postal address blocks, in: Pattern
Recognition and Machine Intelligence, Springer, 2009, pp. 381–386.
[43] X. Liu, W. Pedrycz, Axiomatic Fuzzy Set Theory and Its Applications, Springer,
Berlin, Heidelberg, 2009.
[44] G.V. Geem, Loganathan, A new heuristic optimization algorithm: harmony
search, Simulation 76 (2) (2001) 60–68.
[45] S. Salcedo-Sanz, A. Pastor-Sánchez, J. Del Ser, L. Prieto, Z.W. Geem, A coral reefs
optimization algorithm with harmony search operators for accurate wind
speed prediction, Renew. Energy 75 (2015) 93–101.
[46] Z.W. Geem, J.-H. Kim, Wastewater treatment optimization for fish migration
using harmony search, Math. Probl. Eng. 2014 (2014) 1–5.
[47] R. Sarkhel, A. Saha, N. Das, An enhanced harmony search method for bangla
handwritten character recognition using region sampling, in: Proceedings of
the 2nd IEEE International Conference on Recent Trends in Information Sys-
tems (ReTIS-15), 2016, p. (in press).
[48] Q.-K. Pan, P.N. Suganthan, J.J. Liang, M.F. Tasgetiren, A local-best harmony
search algorithm with dynamic subpopulations, Eng. Optim. 42 (2) (2010)
101–117.
[49] Y. Li, Z.-F. Wu, Fuzzy feature selection based on min–max learning rule and
extension matrix, Pattern Recognit. 41 (1) (2008) 217–226.
[50] A. Roy, N. Das, R. Sarkar, S. Basu, M. Kundu, An Axiomatic Fuzzy Set Theory
Based Feature Selection Methodology for Handwritten Numeral Recognition.
[51] C. Chang, C. Lin, LIBSVM: A Library for Support Vector Machines, vol. 2 (3),
2011.
[52] V.N. Vapnik, An overview of statistical learning theory, IEEE Trans. Neural
Netw. 10 (5) (1999) 988–999.
[53] H.A. Khan, A. Al Helal, K.I. Ahmed, Handwritten Bangla digit recognition using
Sparse Representation Classifier, in: Proceedings of the 2014 International
Conference on Informatics, Electronics & Vision (ICIEV), 2014, pp. 1–6.
[54] T. Hassan, H.A. Khan, Handwritten Bangla numeral recognition using Local
Binary Pattern, in: Proceedings of the 2015 International Conference on
Electrical Engineering and Information Communication Technology (ICEEICT),
2015, pp. 1–4.
[55] N. Das, J.M. Reddy, R. Sarkar, S. Basu, M. Kundu, M. Nasipuri, D.K. Basu, A
statistical–topological feature combination for recognition of handwritten
numerals, Appl. Soft Comput. 12 (8) (2012) 2486–2495.
[56] A. Roy, N. Mazumder, N. Das, R. Sarkar, S. Basu, M. Nasipuri, A new quad tree
based feature set for recognition of handwritten bangla numerals, in: AICERA
2012 – Annual International Conference on Emerging Research Areas: Inno-
vative Practices and Future Trends, 2012, pp. 1–6.
[57] J. Xu, J. Xu, Y. Lu, Handwritten Bangla digit recognition using hierarchical
Bayesian network, in: Proceedings of the 2008 3rd International Conference
on Intelligent System and Knowledge Engineering, 2008, vol. 1, pp. 1096–1099.
[58] Y. Wen, Y. Lu, P. Shi, Handwritten Bangla numeral recognition system and its
application to postal automation, Pattern Recognit. 40 (1) (2007) 99–107.
[59] S. Basu, N. Das, R. Sarkar, M. Kundu, M. Nasipuri, D.K. Basu, A hierarchical
approach to recognition of handwritten Bangla characters, Pattern Recognit 42
(7) (2009) 1467–1484.
[60] A. Roy, N. Das, R. Sarkar, S. Basu, M. Kundu, M. Nasipuri, Region selection in
handwritten character recognition using artificial bee colony optimization, in:
 R. Sarkhel et al. / Pattern Recognition 58 (2016) 172–189 189

Proceedings of the Third International Conference on Emerging Applications
of Information Technology (EAIT– 2012), 2012, pp. 183–186.
[61] N. Das, B. Das, R. Sarkar, S. Basu, M. Kundu, Handwritten Bangla basic and
compound character recognition using MLP and SVM classifier, J. Comput. 2
(2) (2010) 109–115.
[62] S.K.P.T.K. Bhowmik, U. Bhattacharya, Recognition of Bangla Handwritten
Characters Using an MLP Classifier Based on Stroke Features, Springer, Berlin,
Heidelberg, 2004.

Ritesh Sarkhel received his B.C.S.E degree from Jadavpur University in 2012. He worked as an R&D Engineer in Samsung Research Institute, Noida from 2012 to 2014. He is
currently pursuing M.C.S.E degree from Jadavpur University. His areas of current research interest are OCR of handwritten text, optimization techniques and computer vision.

Nibaran Das received his B.Tech degree in Computer Science and Technology from Kalyani Govt. Engineering College under Kalyani University, in 2003. He received his M.C.
S.E. degree from Jadavpur University, in 2005. He received his Ph.D. (Engg.) degree thereafter from Jadavpur University, in 2012. He joined J.U. as a lecturer in 2006. His areas
of current research interest are OCR of handwritten text, Bengali fonts, optimization techniques and image processing. He has been an editor of Bengali monthly magazine
“Computer Jagat” since 2005.

Amit K. Saha received his B.Tech degree in Information Technology from WBUT, in 2011. He received his M.T.C.T. degree from Jadavpur University, in 2015. His areas of
current research interest are OCR of handwritten text, Nature Inspired Computing and Multi-Objective Optimization.

Mita Nasipuri received her B.E.Tel.E., M.E.Tel.E., and Ph.D. (Engg.) degrees from Jadavpur University, in 1979, 1981 and 1990, respectively. Prof. Nasipuri has been a faculty
member of J.U. since 1987. Her current research interest includes image processing, pattern recognition, and multimedia systems. She is a senior member of the IEEE, U.S.A.,
Fellow of I.E. (India) and W.B.A.S.T., Kolkata, India.