mixed conjunct consonants recognition by using soft ...like as fuzzy neural hybrid system, fuzzy...
TRANSCRIPT
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 23 (2017) pp. 13405-13415
© Research India Publications. http://www.ripublication.com
13405
Mixed Conjunct Consonants Recognition by using Soft Computing based
FNHS through NI-Vision Assistant
Santosh Kumar Henge1
Department of Computer Science, University Campus College, Kakatiya University, Hanamkonda, Warangal District, Telangana State, India.
Dr B.Rama2
Assistant Professor, Department of Computer Science, University Campus College, Kakatiya University, Hanamkonda, Warangal District, Telangana State, India.
1Orcid: 0000-0003-1884-9945, IDsScopus Author ID: 57193696079
Abstract
Soft Computing technology represents the improved versions
of intellectual architectures such as artificial intelligence
based neural networks, fuzzy control system, genetic
algorithms and these individual approached are blended
together to fabricate the influential intellectual hybrid systems
like as fuzzy neural hybrid system, fuzzy support genetic
algorithm and genetically support neural approach to convert
their individual drawbacks into features. In OCR approach,
many researchers are proposed various traditional cum
advanced approaches to recognize the individual characters,
number, consonants. Some researchers are designed
methodologies to recognize the conjunctives with their
consonants. But in real time scenario, machines can face the
difficulty to classify the input characters into consonants and
conjunctives, and especially identify the mixed connective
consonants with their exact correlated connectives especially
when they mixed with pre cum post located consonants.
This paper is proposing layered methodology to solve the
complex problems in mixed conjunct consonant OCR by
using the intellectual fuzzy based neural hybrid systems.
Performing the accurate categorization of input text into
concern individual consonant and conjunctives or vowel
composed consonant by measuring the correlated connectives
of each consonant to form the full text and identifications of
end-points where the connectives of vowels are mixed with
consonants. The input text smoothing, erosion, interpolation,
normalization, various stroke recognition variations,
classification and identification of mixed conjunct consonant
based recognition problems has been reduced through neural–
fuzzy–neural hybrid system through the NI-Vision Assistant
technologies.
Keywords: Soft Computing (SC), Text-Stressing-Blows
(TSB), Composite-Text (CT), NI-Vision Assistant (NIVA),
Input-Text-Normalization (ITN), Input-Text-Smoothing
(ITS), Input-Text-Interpolation (ITI).
INTRODUCTION
The In present days the traditional computing system based
approaches are successfully replaced with the hybrid soft
computing (SC) approaches by holding the complementary
merits and best practices to stimulate their appliance in each
environment [2]. SC hybrid based systems are sparkling
explore innovation of modern computational aptitude with the
growth of future invention of intellectual systems [1] and it is
an aggregation of growing methodologies intend at employing
the tolerance-ability for vagueness, partial truth towards to
attain heftiness, suppleness and low cost [10]. Most of many
Indian languages based text has composed with the similar
kind character set flow by holding the various key strokes.
Designing optical character recognition (OCR) engine
approached framework for one Indian language serves as a
supporting framework for other Indian language [8]. Brahmi
script is the main base for all the Indian languages. The
Brahmi script is a phonographic style of writing system. In
this system almost of all symbol forms are directly or
indirectly correlated to the phonemes of the language in which
it is written and also the characters processed with their
descendants and conjunctives. This type of mixed form of text
variations has been called as composite-text (CT). And also,
sometimes the individual consonants can mix with not only
with each of their correlated vowels of the handwritten system
but also with other consonants to outline the language
ligatures.
Researchers are proposed and designed the various OCR
approaches for handwritten and printed characters recognition.
The handwritten character recognition (HCR) is more
complex than the printed character recognition (PCR) because
of their internal stroke rate variations, flow and placement of
characters. Some of the methods have been proposed for the
handwritten character recognition, such as stochastic based
representation [22], structure oriented representation [19-20],
learning cum training environmental based approaches [21],
motor based method [23], support vector machine (SVM) and
neural network based approaches (NNbA). The learning cum
training environmental based approaches is inward extensive
consideration for HCR and pattern recognition (PC) based
problems. The NNbA has attained the high success rate
comparatively with other supporting survive models in many
recognition areas. The SVM has attained the rational
simplification based accuracy for recognizing the handwritten
based numeric recognition [26], Arabic character recognition
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 23 (2017) pp. 13405-13415
© Research India Publications. http://www.ripublication.com
13406
[24], Roman character recognition [25] and Thai character
recognitions [27]. Every approach has its own techniques,
achievements, success rates and lagging issues. In this paper is
on the designing the system for off-mode printed and
handwritten character recognition of Telugu south Indian
language scripts by using the fuzzy neural hybrid system
through the NI-Vision Assistant (NIVA). An artificial based
neural network was pioneer by the McCulloch and Pitt based
on the surveillance that the human-brain consists of plentiful
interrelated neurons encapsulating the rarest thing in this
humankind - human based intelligence [12]. The major benefit
of NN is over conventional computers lies are in its high level
parallelism [11]. It is attractive extensively conventional
technology that the advent of NN will open new sympathetic
into how to make things easier programming and algorithm
invent for a given end of problems [5]. NN can also deriving
its own limitations like as the designing neural based
networks, extensive training periods, and opportunity of over-
fitting ratios, but finally it will effectively executes all
operations and forms the excellent outcomes by taking a
more care in right mode.
Figure 1: The computational intelligence based systems
derives many application areas in complex domains and
concern areas.
A blended environment of fuzzy logic controller (FLC) and
neural based network (NbN) practices are measured as single
amongst the list of methods [6]. By taking into consideration
of these two advancements independently, each of NbN based
and FLC inference based systems are holding its own self
qualities. The learning ability of NbN is best fit for structures
based function approximations. FLC inference rule based
systems are utilized for enhancing the decision making of
knowledge representations along with the NbN justification
potentials. The blended hybridization of soft computing based
fuzzy neural hybrid system (FNHS) is used to defeat their
individual limitations and it is imitated cum implicates the
human based thoughts and actions [7][14]. This combinational
hybridization intelligent techniques are attained the high
success rate in complex domains of applications in modern
and traditional literature [8] either when fuzzy set theory is the
heart of such a system, or when the neural mechanism is the
dominant component in the architecture. This blended
hybridization system will less evident and significant, relating
to dissimilar FLC or NbN based arrangements such as self-
systematize-maps or radial-based-operations [4].
OBSERVED COMPLEX PROBLEMS IN MIXED
CONJUNCT CONSONANTS OCR In OCR approach many researchers are proposed various
traditional cum advanced approaches to recognize the
individual characters, number, consonants. Some of them are
designed some methodologies to recognize the conjunctives
with their consonants. Some of the researchers are proposed
an overlapping-bounding-box based approach with the
implementation of convinced choice of edging margins for
segmentation of Telugu touching conjunctive consonants with
their vowels. This overlapping-bounding-box based method
was derived for recognizing the single conjunctives and it is
text-size based box advance. The detection of two fonts with
having touchy connection-form has expressed [15-16] and
tested with only touching of two basic characters [14]. But in
real time scenario, machines can face the difficulty to classify
the input characters into consonants and conjunctives, and
especially identify the mixed connective consonants with their
exact correlated connectives especially when they mixed with
pre cum post located consonants. The important major
complexity is to in Indian language HCR for differentiate
among the writing variations with the different text stressing
blows of single writer and same type of text stressing blows
by the different writers along with the negligible deviation in
comparable texts in the scripted language.
1 2 3 4 5 6 7
Figure 2: Telugu conjunctive consonant: first two consonants
formed with left-sided conjunctives, third consonant formed
with top cum bottom positioned conjunctives, fourth and fifth
consonants formed bottom positioned conjunctives with sub-
conjunctives, sixth and seventh consonant formed with left-
sided space-allocated conjunctives.
CIbS application areas in complex domains
Decision-making-process
Diagnosis-retrieval
Supervision-un-supervision
Computing-modelling
Data-analysis
Control based engineering,
Data analysis and Function approximation
Monitoring and diagnosis of complex dynamic
systems, chaotic domains and time-series data,
with a special emphasis on economic or
financial problems and electromechanical
based devices and systems
Numerous medical diagnosis problems
Managerial decisions and strategic decision-making
CIbS techniques Applied Areas
Computational Intelligence based Systems
(CIbS)
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 23 (2017) pp. 13405-13415
© Research India Publications. http://www.ripublication.com
13407
Figure 3: Same conjunctive consonant formed with different handwritten key-strokes by the different writers
Major Objectives
This paper is proposing layered methodology to solve the
complex problems in mixed conjunct consonant OCR by
using the intellectual fuzzy based neural hybrid systems.
Performing the accurate categorization of input text into
concern individual consonant and conjunctives or vowel
composed consonant by measuring the correlated connectives
of each consonant to form the full text and identifications of
end-points where the connectives of vowels are mixed with
consonants. The input text smoothing, erosion, interpolation,
normalization, various key-stroke recognition variations,
classification and identification of mixed conjunct consonant
based recognition problems has been reduced through neural–
fuzzy–neural hybrid system through the Lab View
technologies.
METHODOLOGY IMPLICATIONS THROUGH
FUZZY NEURAL HYBRID SYSTEM (FNHS)
The combinational framework of FNHS will implicate the
working mechanism to learn all the parameters from FLC
system. The FNHS system can executes the inference based
operations based on the FLC fuzzification rules by taking the
help of prior-predefined existing knowledge.
The layered environmental FNHS, The human-like reasoning
style of FLC will compose linguistic replica and fuzzy-sets
include a hug set of If-Then implicated fuzzy based rules.
FNHS derives the trained data-set has been formed from the
n-dimensions of functions and also it derives the inaccuracy-
computing-module to progress the learning statements when
the faults has been considered. The membership functions are
primarily defined, activate the generated parameters from the
membership functions and the training-learning process is
performed when it has required for the further execution [6].
Under the training process, the conditional based if-then-else
fuzzy inference rules are applied for constructing the domain
based expert process and for the future extraction. The NbN is
re-applied when the execution procedure has identifies the
poor training data-set, its re-initiated for another modified
training data-set. The classifications cum identification
process will starts by sort out the total text into two different
types of recognition forms such as normal consonants and
conjunct consonants.
The transformation futures of FNHS represents the inputs as
neuron based layers, converts it into fuzzified based set inputs,
and then apply the fuzzification sequence practices with the
wanted fuzzy rule based knowledge to the concern fuzzy input
sets and finally to produce the required output through
concern responses [7]. The derived combinational mixed-
version of FNHS based inference rules invents the technical
source to solve complex problems connective conjunct
consonants and numerals [14]. FNHS handle all divergence of
rational, numeric and linguistic data set values. Expert based
knowledge is easily adaptable by using the NbN and with
advancement of supplementary rules of fuzzification
membership functions to accomplish the convinced
conditions. It generates the realistic, accurate outcomes and
saves the times. As shown in the fig. 4.
ʃ (P(X))
Best outcomes
Normalization-
Input samples produced
based on the scaling factors
Fuzzification
Defuzzification
Rule-based
Inference Method
Operations are
executed under the
controlled of physical
device
Y1..n
ʃ (Q(Y))
X1...n
Basic input
values to the
system
Fuzzy System – estimated calculation
Component feeler used
for inaccuracy
calculations
Normalization-
Re-sampling the un-
recognized output
with new scaling
factors
Fuzzy-based data
samples generation Rules and group generations based
on the Fuzzification Regulation
Figure 4: Architecture of Fuzzy Neural Hybrid System with their concern internal Inference Rules [14].
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 23 (2017) pp. 13405-13415
© Research India Publications. http://www.ripublication.com
13408
The rough input values will converted into layered
approached neuron base inputs such as X(i1-in), then these
inputs altered into fuzzy based input values to generate the
fuzzy sets. Then the fuzzification method has implemented
with fuzzy based inference rules such as ʃ (P(X)) for
previously collected data units, then the hug fuzzy set has
been formed with concern data sets A(i1-in) with help of
fuzzification membership functions cum derived rules and it
generates the values of ʃ (Q(Y)). In next stage, the linguistic
values such as Y1-Yn has been generated based set-based-
fuzzy-inference-rules. Finally the output responses (B) will be
produced once the linguistic-values satisfy the required-input-
linguistic values. If it is not satisfied or when it detects any
unrelated output symbol or identification-error then the
normalization progression will be implicated to concern
output linguistic values to avoid the redundancy based errors,
this process repetitive until the identifier detects any
overlapping recognition or slot correction based errors [14].
Figure 5: FNHS based methodology with fault calculation
mechanism implicating with passing through the concern
layers approach [7].
Pre-processing of input knock data
The text-stressing-blows (TSB) is pre-processed by using
three major steps, such as input-text-normalization, input-text-
smoothing and input-text- Interpolation as explained in the
following subsections [8].
1) Input-Text-Normalization (ITN): Most of characters
will be formed by various text-stressing-blows and
some TSB enhanced with the placement of top and
bottom of the original consonant. The original
consonant has been placed with one-line-space based
on the size of text has been formed by the composer.
Here the entire text has been classified by the three
line-spaces or layers such as the middle, top and bottom
line-spaces. The main and important line-space called
as middle-line-space which consist the main consonant,
the top-line-space and bottom-line-spaces will consists
the conjunctives and vowels. The middle-line-spaced
main consonants will derive largest TSB that crop-up in
the main part of the text and scaled down by its height
value. The pre-text, post-text and size of the text is
worn for normalization process if is very difficult to
recognize the TSB.
2) Input-Text-Smoothing (ITS): The input text
smoothing process can be applied once finished the
generalized ITN process when the TSB is motionless
piecewise linear with the line based fragments
connecting the points along its concern curve motion.
The Gaussian membership based filters has been best
applicable for providing the smoothing the input based
characters. Here different parameters has been
considered and applied in Gaussian membership based
filters depends on the size of the text, size the function
has been retrieved. In FLC system, the fuzzy based sets
can be build with help of the membership function for
representing each input value with the fuzzy
implicated real-numbers based set intervals 0’s and 1’s
such as [0, 1]. These fuzzy implicated set-intervals are
mentioned as membership based degrees or grades.
This function of memberships formed as Gaussian,
triangular and trapezoidal. A triangular membership is
described by a triplet (x, p, y), where “p” is the modal
value, x represents right boundary and y represents left
boundary. The Gaussian implicated membership
functions and trapezoidal implications are shown in
Fig. 6.
With 0 for any k ϵ { 1, 2,…….n}
Figure 6: The Gaussian implicated membership functions
represented for and trapezoidal implications
Bell Curve sets of Gaussian give fuzzy system with higher
quality by basic learning rules. The variables of bell curve can
tune by using these learning rules. The following equation
represents the function of Gaussian.
Figure 7: Sample data representation with various cost driver
implicated by using Membership Function of Gaussian. Here
Ci represents as center and σi represents width of the ith
fuzzy set respectively.
Layer-1
Layer-2 Layer-3 Layer-4 Layer-5
Membership degree
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 23 (2017) pp. 13405-13415
© Research India Publications. http://www.ripublication.com
13409
The fuzzy logic based sets implicated with linguistic values
with concern membership function μ Gaussian shaped has
shown in Fig. 6. The fuzzy sets can be represented by
linguistic values, for each factor of interest it may be character
intensity, orientation, layout and alignment.
3) Input-Text-Interpolation (ITI): The input text
interpolation process can be applied after the ITS where
the smoothing TSB based text is interpolated based on
the starting cum ending points of the concern
consonants and their correlated conjunctives along with
mentioned space gap which implicated by the curve
variations. Each consonant has derived two end-points,
one as starting-point and ending-point. Sometimes
consonant derives four end-points if it formed with
concern conjunctive either in top-line-spaced or
bottom-line-spaced, two end-point belongs to main-text
considered as middle-line-spaced text and another two
end-point belongs concern conjunctive. Sometimes
consonant derives eight end-points once it has formed
with top-line-spaced and bottom-line-spaced
conjunctives along with the middle-line-spaced text.
The granules of ‘x’ play the task of fuzzy actions and
the Pi is their granular probabilities [29]. It should be
pointed out that in the case of probability distributions
it is necessary to reinitiate between ‘I’ granular
probability distributions and granule-valued probability
distributions [35][30] fig. 9 and fig 10. It should be
noted that there is a connection between the concept of
a granular probability distribution and the notion of
perfilieva transform [31]. An instance of a granule-
valued distribution is a random set. There is a close
connection between granule valued distributions and
the Dempster–Shafer theory of evidence [32] [33] [34].
Figure 8: Function represented with granulation based
implications such as S as small, M as medium and L as large
for generating the fuzzy sets
The granuland of f, f viewed as a synopsis of f.
Figure 9: Probability distribution based on Interpolation and
granulation
Figure 10: General-granular versus granule-based-valued
distributions
Decision Making State (DMS) of FNHS
The DMS of input text can be executed with help of four
various implications by taking the support of AI based neural
networks. Originally the input-text will classified into three
line-spaces like as Middle-line-spaced text (MLST), Top-line-
spaced text (TLST) and Bottom-line-spaced text (BLST). The
neural networks based layer approach can be applied for
distributing the considered input text. The MLST can be
considered as neural networks based layer1, TLST considered
as neural networks based layer2, BLST considered as neural
networks based layer3 and also there some other hidden
neural networks layers can be considered if there is sub
sequences of main consonants and their conjunctives. Here the
classification process can be done with help of four methods
such as Middle-line-spaced text (MLST); Middle-line-spaced
text (MLST) with Top-line-spaced text (TLST); Middle-line-
spaced text (MLST) with Bottom-line-spaced text (BLST);
Middle-line-spaced text (MLST) with Top-line-spaced text
(TLST) and Bottom-line-spaced text (BLST).
In this evolutionary research, the hybrid fuzzy neural
invention model is proposed by using Gaussian based
functions to deal with basic linguistic data values and text
image with layered approach analysis, and to generate
inference rules of fuzzy membership functions and
corresponding responses for further processing in neural
hybrid fuzzy system, the membership functions primitives
have been added to find form a character, which is part of the
layered based dictionary. The finally the character is to be
recognized cum tested with layered based dictionary
containing origin characters by using proficient methodology.
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 23 (2017) pp. 13405-13415
© Research India Publications. http://www.ripublication.com
13410
Table 1: Input text implicated through AI based neural networks with their layer implication, origin of the text and fuzzy based
implicated inference rules.
S.No. AI based Neural
Networks
Layer Implication Base of Text Fuzzy based
Function
1 Layer1 (L1) Middle-line-spaced text
(MLST)
Original base Consonants (ObC) ʃ (1, 0.0, 0.0)
2 Layer2 (L2) Top-line-spaced text (TLST) Vowels Type TLST placed
Conjunctives (VTTLSTC)
ʃ (0, 1.0, 0.0)
3 Layer3 (L3) Bottom-line-spaced text (BLST) Vowels Type BLST placed
Conjunctives (VTBLSTC)
ʃ (0, 0.0, 1.0)
4 Hidden Layer1 (HL1) Only MLST without TLST and
BLST
MLST without TLST & BLST ʃ (1, 0.0, 0.0)
5 Hidden Layer2 (HL2) MLST with TLST ObC with VTTLSTC ʃ (1, 1.0, 0.0)
6 Hidden Layer3 (HL3) MLST with BLST ObC with VTBLSTC ʃ (1, 0.0, 1.0)
7 Hidden Layer4 (HL4) MLST with TLST and BLST ObC with VTTLSTC &&
VTBLSTC
ʃ (1, 1.0, 1.0)
8 Hidden Layer5 (HL5) MLST with TLST+Sub- TLST
and BLST
ObC with VTTLSTC && Sub-
VTTLSTC && VTBLSTC
ʃ (1, 1.1, 1.0)
9 Hidden Layer6 (HL6) MLST with TLST and
BLST+Sub-BLST
ObC with VTTLSTC &&
VTBLSTC && Sub-VTBLSTC
ʃ (1, 1.0, 1.1)
10 Hidden Layer7 (HL7) MLST with BLST+Sub-
BLST+Sub.Sub-BLST
ObC with && VTBLSTC &&
Sub-VTBLSTC && Sub-Sub-
VTBLSTC
ʃ (1, 1.0, 1.1.1)
Experimental Design and Simulation Results
OCR is the process by which a NI machine vision application
interprets the characters and symbols in assessment images.
The OCR Training based interface has implicated for training
and learning process to classify the text based on their text-
orientation, text-alignment, text-normalization based on the
neural networks layers with help of fuzzy based inference
rules to recognize the concern characters.
Figure 11: Telugu consonants and conjunctives implicated
with interface based learning set
Figure 12: Various involved stages in pre-processing input
text-stressing-blows (TSB) of Telugu handwritten characters,
the stages are Input Text Normalization (ITN), Erosion, Input
-Text- Smoothing (ITS) and Input Text Interpolation (ITI)
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 23 (2017) pp. 13405-13415
© Research India Publications. http://www.ripublication.com
13411
Figure 13: Telugu consonants and conjunctives implicated with concern converted another form of codes
Figure 14: Telugu consonants and conjunctives implicated with concern converted another form of codes
Figure 15: Final results of Telugu consonants and conjunctives along with their converted recognized symbol code
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 23 (2017) pp. 13405-13415
© Research India Publications. http://www.ripublication.com
13412
Table2. Input text implicated with another form of text with concern classification score, fuzzy logic inference set along with
recognition success rate.
Symbol
Formation
Classification
Score
Fuzzy Logic
Inference Set Left Top Width Height
Recognition
Success Rate in %
n 710 ʃ (1, 0.0, 0.0) 55 39 28 16 71 %
€ 809 ʃ (1, 0.0, 0.0) 100 39 28 16 80.9 %
‚ 976 ʃ (1, 0.0, 0.0) 145 39 31 19 97 %
‡ 992 ʃ (1, 1.0, 0.0) 193 29 39 26 99 %
– 785 ʃ (1, 1.0, 0.0) 249 33 25 22 78.5 %
} 989 ʃ (1, 1.0, 0.0) 291 33 41 22 98.9 %
TT 985 ʃ (1, 0.0, 0.0) 350 39 47 16 98.8 %
TÖ 996 ʃ (1, 1.0, 0.0) 414 39 60 16 99 %
m 924 ʃ (1, 0.0, 0.0) 491 33 23 22 92.6 %
@ 803 ʃ (1, 1.0, 0.0) 531 30 23 25 80 %
× 937 ʃ (1, 0.0, 0.0) 55 99 23 16 93 %
ÿ 915 ʃ (1, 0.0, 0.0) 95 99 24 16 91 %
z 966 ʃ (1, 1.0, 0.0) 136 92 24 23 96 %
W 970 ʃ (1, 1.0, 0.0) 177 94 29 21 97 %
ng 995 ʃ (1, 0.0, 0.0) 223 99 49 16 99 %
n^g 988 ʃ (1, 0.0, 0.0) 297 99 34 16 98.5 %
£ 797 ʃ (1, 1.0, 0.0) 55 161 24 26 79.7 %
K 980 ʃ (1, 0.0, 1.0) 96 171 25 21 98.2 %
· 480 ʃ (1, 1.0, 0.0) 138 161 23 26 48.6 %
· 990 ʃ (1, 1.0, 1.0) 176 160 35 33 99.2 %
¤ 566 ʃ (1, 0.0, 0.0) 228 170 31 21 57 %
á 871 ʃ (1, 1.0, 0.0) 276 161 26 26 87 %
ó 992 ʃ (1, 1.0, 1.0) 319 161 26 31 92 %
È 839 ʃ (1, 0.0, 0.0) 362 171 24 16 84 %
Á 992 ʃ (1, 1.0, 1.0) 403 160 41 33 99 %
v 958 ʃ (1, 0.0, 0.0) 462 171 24 17 96 %
³ 889 ʃ (1, 1.0, 0.0) 503 165 25 22 89 %
s 638 ʃ (1, 1.0, 0.0) 545 161 23 26 64 %
& 565 ʃ (1, 1.0, 0.0) 55 220 23 26 57.5 %
ó 750 ʃ (1, 1.0, 1.0) 96 220 23 31 75 %
D 980 ʃ (1, 0.0, 0.0) 137 230 24 16 98 %
Ô 832 ʃ (1, 1.0, 0.0) 178 220 25 26 84 %
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 23 (2017) pp. 13405-13415
© Research India Publications. http://www.ripublication.com
13413
ó 771 ʃ (1, 1.0, 1.0) 220 220 24 31 78 %
< 890 ʃ (1, 1.0, 0.0) 261 220 24 26 89 %
ó 863 ʃ (1, 1.0, 1.0) 302 220 24 31 87 %
ƒ 831 ʃ (1, 1.0, 0.0) 343 220 23 26 84 %
| 789 ʃ (1, 1.0, 0.0) 383 219 23 27 79 %
/ 890 ʃ (1, 1.0, 1.0) 424 219 23 32 89 %
‹ 459 ʃ (1, 0.0, 0.0) 465 230 25 16 47 %
“: 465 ʃ (1, 1.0, 1.0) 507 222 28 29 46 %
T 961 ʃ (1, 1.0, 0.0) 549 220 34 26 96 %
j 993 ʃ (1, 1.0, 0.0) 63 280 44 26 99.3 %
s 868 ʃ (1, 1.0, 0.0) 124 280 23 26 86.8 %
\ 868 ʃ (1, 0.0, 0.0) 162 290 24 16 87 %
e 383 ʃ (1, 1.0, 0.0) 203 280 23 26 39 %
ø 994 ʃ (1, 1.0, 0.0) 243 280 27 26 99.4
d 216 ʃ (1, 1.0, 0.0) 286 279 23 27 22 %
Ÿ 984 ʃ (1, 1.0, 0.0) 326 279 27 31 98 %
Ÿ 878 ʃ (1, 1.0, 0.0) 367 279 36 27 89 %
ø 997 ʃ (1, 1.0, 0.0) 420 280 25 26 99.7 %
£ 998 ʃ (1, 1.0, 1.0) 461 280 25 37 99 %
n 66 ʃ (1, 0.0, 0.0) 502 290 27 16 66 %
CONCLUSION
The individual approaches of neural networks and fuzzy logic
control system are blended together to fabricate the influential
intellectual hybrid systems like as fuzzy neural hybrid system,
fuzzy support genetic algorithm and genetically support neural
approach to convert their individual drawbacks into features.
OCR is the process of moderating the human written cum
printed text into machine readable digitalized form.
This paper is proposing layered methodology to solve the
complex problems in mixed conjunct consonant OCR by
using the intellectual fuzzy based neural hybrid systems. The
Input Text Smoothing, Erosion, Interpolation, Normalization,
various Stroke Recognition variations, classification and
identification of mixed conjunct consonant based recognition
problems has been reduced through neural–fuzzy–neural
hybrid system through the Lab View technologies.
REFERENCES
[1] Santosh Kumar Henge, Dr B.Rama, “Comparative
Analysis of Soft Computing Hybrid Intellectual System
Implications for Salvation of Machine Learning based
Complex Problems”, in International Journal of
Advanced Scientific Technologies, Engineering and
Management Sciences (IJASTEMS-ISSN: 2454-356X),
Page No: 15-27, Volume.3,Special Issue.1,March.2017.
[2] Rahul Kala, Anupam Shukla, Ritu Tiwari, “Comparative
analysis of intelligent hybrid systems for detection of
PIMA indian diabetes”, Published in: Nature &
Biologically Inspired Computing, World Congress
onDOI: 10.1109/NABIC.2009.5393877, IEEE Xplore:
22 January 2010.
[3] A. Conn, “The AI wars: The battle of the human minds
to keep artificial intelligence safe,”
http://futureoflife.org/2015/12/17/the-ai-warsthe-battle-
of-the-human-minds-to-keep-artificial-intelligence-safe,
Dec. 2015.
[4] Athanasios Tsakonas and George Dounias, “Hybrid
Computational Intelligence Schemes in Complex
Domains An Extended Review”, University of the
Aegean, Business School, Dept. of Business
Administration, 8 Michalon St., 82100 Chios, Greece.
[5] Machine Learning-What it is and why it matters
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 23 (2017) pp. 13405-13415
© Research India Publications. http://www.ripublication.com
13414
[6] http://www.sas.com/en_us/insights/analytics/machine-
learning.html
[7] Santosh Kumar Henge, Dr B.Rama, “Neural Fuzzy
Closed Loop Hybrid System for Classification,
Identification of Mixed Connective Consonants and
Symbols with Layered Methodology”, IEEE
International Conference on Power Electronics,
Intelligent Control and Energy Systems (ICPEICES-
2016) 978-1-4673-8586-2/16/$31.00 ©2016 IEEE, Pp.
2880-2887, July 2016,
[8] Santosh Kumar Henge, B Rama, “Five Layered-Neural
Fuzzy Closed Loop Hybrid Control System with
Compound Bayesian Decision Making Process for
Classification cum Identification of Mixed Connective
Conjunct Consonants and Numerals” in Springer Nature
Singapore Pte Ltd. 2017, S.K. Bhatia et al. (eds.),
Advances in Computer and Computational Sciences,
Advances in Intelligent Systems and Computing 553 ,
pp.619-629, , DOI 10.1007/978-981-10-3770-2_58.
[9] H. Swethalakshmi, Anitha Jayaraman, V. Srinivasa
Chakravarthy, C. Chandra Sekhar, Department of
Computer Science and Engineering, Department of
Biotechnology, Indian Institute of Technology Madras,
Chennai - 600 036, India
[10] Chen Z. Computational Intelligence for Decision
Support. CRC Press, 2000
[11] S. B. Kotsiantis, Supervised Machine Learning: A
Review of Classification Techniques, Department of
Computer Science and Technology University of
Peloponnese, Greece End of Karaiskaki, 22100 , Tripolis
GR, Informatica 31 (2007) 2.
[12] Jason Brownlee, Applied Machine Learning Process -
The Systematic Process For Working Through
Predictive Modeling Problems That Delivers Above
Average Results, By on February 12, 2014 in Machine
Learning Process.
[13] http://machinelearningmastery.com/process-for-
working-through-machine-learning-problems/
[14] Sumathi, S., Paneerselvam, S., (2010), Computational
Intelligence Paradigms: Theory & Applications using
MATLAB.
[15] Special Issue on Solving Complex Machine Learning
Problems with Ensemble Methods.
https://www.journals.elsevier.com/neurocomputing/call-
for-papers/special-issue-on-solving-complex-machine-
learning-problems
[16] Santosh Kumar Henge, Dr B.Rama, “Compound
Bayesian Decision Making Approach with Negative
Input Image based on Neural Fuzzy Logic Inference
Rules” in “1st International Conference on Computer,
Communication and Management Technologies
ICCCMT-2016”, pp.128-134, ISBN No: 978-93-86176-
12-7, held on 22-24 September 2016, Organized by
IEEE-UP Section incorporated with S.P.Memorial
Institute of Technology, Allahabad, India.
[17] J. Bharathi, Dr. P. Chandrasekar Reddy, “Segmentation
of Telugu Touching Conjunct Consonants Using
Overlapping Bounding Boxes”, International Journal on
Computer Science and Engineering (IJCSE), vol. 5, 06
Jun 2013.
[18] J. Bharathi, P. Chandrasekhar Reddy, “Improvement of
Telugu OCR by Segmentation of Touching Characters”,
IJRET: International Journal of Research in Engineering
and Technology , vol. 03 Issue: 10th Oct-2014.
[19] H. A. Murthy. C. Chandra Sekhar, C. S. Ramalingam, V.
S.Chakravarthy, ”A multimodal Indian language
interface to the computer”, Conference on Sharing
Capability in Localisation and Human Language
Technologies - 2004, Kathmandu, Nepal, Jan. 5-7, 2004.
[20] H. Rogers, Writing Systems : A Linguistic Approach,
Blackwell Publishers, Australia, 2005.
[21] K. H. Aparna, Vidhya Subramanian, M. Kasirajan, G.
Vijay Prakash, V. S. Chakravarthy, Sriganesh
Madhvanath, ”Online handwriting recognition for
Tamil”, Proceedings of the Ninth International
Workshop on Frontiers in Handwriting Recognition
(IWFHR' 04), Tokyo, Japan, 2004, pp 438-443.
[22] K. F. Chan, D. Y. Yeung, ”Elastic structural mapping
for online handwritten alphanumeric character
recognition”, Proceedings of 14th International
Conference on Pattern Recognition, Brisbane, Australia,
August, pp 1508-1511, 1998.
[23] S. Manke, U. Bodenhausen, ”A connectionist recognizer
for online cursive handwriting recognition”, Proceedings
of ICASSP 94, Vol. 2, 1994, pp 633-636.
[24] X. Li, R. Plamondon, M. Parizeau, ”Model-based online
handwritten digit recognition”, Proceedings of 14th
International Conference On Pattern Recognition,
Brisbane, Australia, August, 1998, pp 1134-1136.
[25] L. R. B. Schomaker, H. L. Teulings, ”A handwriting
recognition system based on the properties and
architectures of the human motor system”, Proceedings
of the IWFHR, CENPARMI, Concordia, Montreal,
1990, pp 195-211.
[26] H. Bentounsi, M. Batouche, ”Incremental support vector
machines for handwritten Arabic character recognition”,
Proceedings of the International Conference on
Information and Communication Technologies, 2004, pp
1764-1767.
[27] C. Bahlmann, B. Haasdonk, H. Burkhardt, ”Online
handwriting recognition with support vector machines -
a kernel approach”, IEEE Transactions on Pattern
Analysis and Machine Intelligence, Vol. 26, No. 3, pp
299-310, March 2004.
[28] Z. Bin, L. Yong, X. Shao-Wei, ”Support vector
machine and its application in handwritten numeral
recognition”, Proceedings of the 15th International
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 23 (2017) pp. 13405-13415
© Research India Publications. http://www.ripublication.com
13415
Conference on Pattern Recognition, 2000, pp 720-723.
[29] P. Sanguansat, W. Asdornwised, S. Jitapunkul, ”Online
Thai handwritten character recognition using hidden
Markov models and support vector machines”,
International Symposium on Communications and
Information Technology , 2004, Japan, October 26-29,
2004, pp 492-497
[30] RAZALI BIN ABU BAKAR,: Development of Online
Unconstrained Handwritten Character Recognition
Using Fuzzy Logic, Universiti Teknologi MARA.
[31] L.A. Zadeh,: From imprecise to granular probabilities,
Fuzzy Sets and Systems 154 (2005) 370–374.
[32] L.A. Zadeh,: Toward a perception-based theory of
probabilistic reasoning with imprecise probabilities,
Journal of Statistical Planning and Inference 105 (2002)
233–264.
[33] I. Perfilieva, Fuzzy transforms: a challenge to
conventional transforms, in: P.W. Hawkes (Ed.),
Advances in Images and Electron Physics, vol.147,
Elsevier Academic Press, San Diego 2007, pp.137-196.
[34] A.P. Dempster,: Upper and lower probabilities induced
by a multivalued mapping, Annals of Mathematical
Statistics 38 (1967) 325-329.
[35] G. Shafer,: A Mathematical Theory of Evidence,
Princeton University Press, Princeton, NJ, 1976
[36] D. Schum,: Evidential Foundations of Probabilistic
Reasoning, Wiley & Sons, 1.
[37] L.A. Zadeh,: Generalized theory of uncertainty(GTU)–
principal concepts and ideas, Computational
Statistics & Data Analysis 51 (2006) 15–46.