Int
ern
at
i
onal
Journ
al of
El
e
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
10
,
No.
2
,
A
pr
il
2020
, p
p. 18
23
~
1832
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v10
i
2
.
pp1823
-
18
32
1823
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om/i
nd
ex
.ph
p/IJ
ECE
Onlin
e s
i
gnatu
re
verifi
cation usin
g hybrid
wa
ve
l
et t
ra
nsf
or
m
Manoj
Chava
n
1
, Ra
vish R.
Singh
2
,
V
in
ay
ak
Bh
ar
ad
i
3
1
Depa
rtment of
El
e
ct
roni
cs
and
Te
l
ec
om
m
unic
ation
Eng
ine
er
ing,
Tha
kur
Coll
ege
of
Engi
n
ee
rin
g
a
nd
Technol
og
y
,
India
2
Aca
demic
Advi
sor,
Tha
ku
r
Edu
ca
t
iona
l
Trust
,
Mum
bai
,
Indi
a
3
Inform
at
ion
T
e
chnol
og
y
Dep
artm
ent
,
Finol
ex
A
ca
dem
y
of
Mana
gement
and
T
echnolog
y
,
Ind
ia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Dec
22,
2017
Re
vised
Oct
18
,
20
19
Accepte
d
Oct
30
, 20
19
Online
signat
u
r
e
ver
if
ic
a
ti
on
i
s
a
prom
ine
nt
beha
vior
al
bio
m
et
ric
tr
ai
t
.
It
offe
rs
m
an
y
dy
nami
c
fea
tu
res
al
ong
with
static
two
dimensional
signat
ur
e
image.
In
thi
s
p
ape
r,
the
H
y
br
id
W
ave
let
Tr
ansform
(HW
T)
wa
s
gene
rated
using
Kronec
ker
produc
t
of
two
orthogona
l
tr
an
sform
such
as
DCT,
DH
T,
Haa
r,
Had
amard
and
Kekre
.
H
W
T
has
the
ability
to
ana
l
y
z
e
t
he
signal
a
t
globa
l
as
well
as
loc
al
l
eve
l
l
i
ke
wave
let
tra
n
sform
.
H
W
T
-
1
and
-
2
was
appl
i
ed
on
the
first
128
sam
p
le
s
of
the
pr
essure
par
ame
te
r
a
nd
first
16
sam
ple
s
of
the
output
were
used
as
fea
ture
ve
ct
o
r
for
signat
ure
v
eri
fi
ca
t
ion.
Thi
s
fea
tur
e
ve
ct
or
is
give
n
to
Le
ft
to
Right
HM
M
cl
assifie
r
to
ide
ntif
y
the
genui
n
e
an
d
forge
d
signature.
For
HW
T
-
1,
DCT
HA
AR
offe
rs
best
FA
R
and
FR
R.
For
H
W
T
-
2,
KEKRE
128
offe
rs
best
FA
R
and
F
RR.
HW
T
-
1
offe
rs
better
p
erf
orm
anc
e
than
HW
T
-
2
in
te
rm
s
of
FA
R
and
FR
R
.
As
the
num
ber
of
sta
te
s
in
c
rease,
t
he
perf
or
m
ance
of
th
e
syst
e
m
i
m
pr
oves.
For
HW
T
-
1,
KEKRE
128
offe
rs be
st
per
form
anc
e
at
275
s
y
m
bols
where
as
for
HW
T
-
2,
b
est
p
erf
or
m
anc
e
is
at 475 s
y
m
bols b
y
KE
KRE
128.
Ke
yw
or
d
s
:
HMM
H
WT
Left Ri
ght M
odel
On
li
ne
Sig
natu
re
Ver
ific
at
io
n
Copyright
©
202
0
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Ma
noj
C
hav
a
n,
Dep
a
rtm
ent o
f El
ect
ro
nics
and Telec
omm
un
ic
at
ion
E
ng
i
ne
erin
g,
Tha
kur
C
olleg
e of E
ng
i
neer
i
ng and
Tech
no
log
y,
Tha
kur Vil
la
ge
, K
a
ndivali
Eas
t M
um
bai
–
400101 Ma
har
a
sh
t
ra, India.
Em
a
il
: pr
of.m
a
noj@gm
ai
l.com
1.
INTROD
U
CTION
Hand
wr
it
te
n
S
ign
at
ures
ha
ve
been
us
e
d
f
or
cent
ur
ie
s
f
or
ide
ntific
at
ion
a
nd
a
uth
e
ntica
ti
on
of
a
per
s
on
as
we
ll
as
do
cum
ents
[1,
2]
.
I
n
biom
et
rics
cl
assificati
on
,
it
is
part
of
be
ha
vioral
char
act
erist
ic
s
li
ke
vo
ic
e,
gait
et
c.
w
her
eas
phy
sic
al
char
act
er
ist
ic
s
include
fin
gerpr
i
nt,
pa
l
m
pr
int,
face
,
iris,
reti
na
et
c.
[
3
]
.
Bi
om
e
tric
chara
ct
erist
ic
s
are
un
i
ver
sal
,
un
i
que
a
nd
m
easura
ble
a
nd
are
be
tt
er
than
per
s
on
al
I
D
ca
rds,
PI
N
or
pass
words
[
4
-
6
]
.
Bi
om
et
ric
syst
e
m
fo
r
sig
natu
res
can
operate
in
tw
o
ways.
First
V
erifica
ti
on,
in
wh
ic
h
the
ind
i
vidual
’s
sig
natu
re
will
be
com
par
ed
with
his
store
d
sig
nat
ur
e
i
n
a
data
base
to
ver
i
f
y
that
the
ind
ivi
du
al
is
the
sa
m
e
wh
o
he
say
s
to
be
.
Seco
nd
I
de
ntific
at
ion
,
in
whic
h
the
sign
at
ure
will
be
com
par
e
d
with the
m
any sign
at
ur
es
in
t
he
d
at
a
ba
se to
iden
ti
fy a
n
i
nd
i
vidual o
ut
of m
any un
known
s.
Au
t
om
ating
t
he
pr
ocess
of
Hand
wr
it
te
n
Sign
at
ur
e
V
erifica
ti
on
will
be
us
e
f
ul
f
or
docum
ent
ver
ific
at
io
n
in
var
i
ou
s
sect
ors
su
c
h
as
ba
nk
i
ng
,
le
gal
docu
m
entat
ion
et
c.
The
re
are
tw
o
ty
pes
of
Sig
natu
re
Ver
ifi
cat
io
n;
offli
ne
(
sta
ti
c)
or
onli
ne
(
dy
nam
ic
).
O
f
fline
sig
natu
res
of
fe
r
tw
o
di
m
ension
al
im
age
of
the
sig
natu
re
s
wh
e
reas
onli
ne
sig
natu
res
ha
ve
a
dded
ad
va
ntage
that
it
al
so
m
easur
es
press
ur
e
ap
pl
ie
d
by
the
us
e
r,
s
peed
of
w
riti
ng,
inc
li
nation
of
pe
n
al
ong
w
it
h
the
two
dim
ensio
nal
sign
at
ur
e
i
m
age
[
7
]
.
D
r.
Kekre
pro
po
se
d
Hyb
r
id
W
a
velet
T
r
ansfo
rm
(HWT)
wh
ic
h
is
f
orm
ed
by
com
bin
ing
the
tw
o
or
t
hogonal
t
ra
ns
f
or
m
s
us
in
g
K
ronec
ke
r
pro
du
ct
.
It
ha
s
the
abili
ty
t
o
analy
ze
the
sign
al
at
glo
bal
as
well
as
local
le
vel
l
ik
e
wav
el
et
trans
form
[
8
]
. H
WT
is
of tw
o t
ypes a
nd are e
xp
la
ine
d bel
ow. C
onside
r
m
at
rices X
a
nd
Y
as
sho
wn b
el
ow.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
2
,
A
pr
i
l 202
0
:
1823
-
1832
1824
X
=
[
11
12
21
22
…
1
…
2
…
…
1
2
…
…
…
]
Y
=
[
11
12
21
22
…
1
…
2
…
…
1
2
…
…
…
]
H
WT
-
1
m
at
rix
‘T
XY
’
of
siz
e
(NxN),
a
s
sho
wn
i
n
Ta
ble
1
,
can
be
f
or
m
ed
by
the
Kro
nec
ker
pro
duct
of
tw
o
or
t
hogo
nal
trans
f
or
m
m
at
rices
X
and
Y
res
pecti
vely
,
with
siz
es
(a
x
a)
a
nd
(
b
x
b),
s
uch
t
hat
N=
a
x
b.
Fo
r
H
WT
–
1,
f
irst
‘b
’
nu
m
ber
of
r
ows
of
th
e
H
W
T
m
at
rix
are
cal
culat
ed
as
the
pr
od
uct
of
each
el
em
e
nt
of
first
r
ow
of
the
ort
hogo
nal
tra
ns
f
or
m
X
with
each
of
the
c
ol
um
ns
of
t
he
ort
hogo
nal
tra
nsfo
rm
Y.
F
or
ne
xt
‘
b’
nu
m
ber
of
r
ow
s
of
H
WT
m
a
t
rix
the
sec
ond
row
of
the
or
t
hogonal
tra
ns
f
or
m
m
a
trix
X
is
sh
ift
ro
ta
te
d
after
bein
g
ap
p
e
nd
e
d
with
ze
ro
s
.
S
i
m
i
la
rly
the
oth
er
rows
of
H
WT
m
at
rix
are
gen
e
rated
as
s
et
of
b
rows
ea
ch
tim
e
for
eac
h of t
he ‘a
-
1’ rows
of
ort
hogo
nal tra
nsfo
rm
m
a
trix X
sta
rting
from
second r
ow up t
o
la
st r
ow.
Table
1.
H
WT
-
1
Ma
trix
y
11x11
y
11x12
…
y
11x1a
y
12x11
y
12x12
…
y
12x1a
…
…
y
1bx11
y
1bx12
…
y
1bx1a
y
21x11
y
21x12
…
y
21x1a
y
21x12
y
22x12
…
y
22x1a
…
…
y
2bx11
y
2bx12
…
y
2bx1a
…
…
…
…
…
…
…
…
…
…
…
…
…
…
y
b
1
x
1
1
y
b
1
x
1
2
…
y
b
1
x
1
1
y
b
2
x
1
1
y
b
2
x1
2
...
y
b
2
x1
a
...
...
y
b
1
x
1
1
y
b
1
x
1
2
...
y
b
1
x
1
2
x21
x22
…
x2a
0
0
…
0
…
…
0
0
…
0
0
0
…
0
x21
x22
…
x2a
…
…
0
0
…
0
…
…
…
…
…
…
…
…
…
…
…
…
…
…
0
0
…
0
0
0
…
0
…
…
x21
x22
…
x2a
x31
x32
…
x3a
0
0
…
0
…
…
0
0
…
0
0
0
…
0
x31
x32
…
x3a
…
…
0
0
…
0
…
…
…
…
…
…
…
…
…
…
…
…
…
…
0
0
…
0
0
0
…
0
…
…
x31
x32
…
x3a
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
…
x
a1
x
a2
…
x
aa
0
0
…
0
…
…
0
0
…
0
0
0
…
0
x
a1
x
a2
…
x
aa
…
…
0
0
…
0
…
…
…
…
…
…
…
…
…
…
…
…
…
…
0
0
…
0
0
0
…
0
…
…
x
a1
x
a2
…
x
aa
H
WT
-
2
m
at
rix
of
siz
e
(Nx
N)
is
al
so
f
or
m
ed
by
the
K
r
on
ec
ker
pro
du
ct
of
two
or
t
hogonal
trans
form
m
at
rices
X
an
d
Y
.
Fir
st
N/2
rows
of
the
m
at
rix
are
f
or
m
ed
by
pro
duct
of
each
el
em
ent
of
first
a/
2
r
ows
of
the
m
a
trix
X
w
it
h
each
of
the
colum
ns
of
the
m
at
rix
Y.
For
nex
t
‘
b’
num
ber
of
r
ow
s
of
m
at
rix,
the
‘a/
2+1’th
row
of
the
ort
hogo
na
l
tra
ns
f
or
m
m
at
rix
X
is
sh
ift
ro
ta
te
d
af
te
r
bei
ng
ap
pe
nd
e
d
with
zer
os.
Nex
t
N/2
r
ows
a
re
gen
e
rated
as se
t of
b rows
eac
h
ti
m
e fo
r
eac
h of t
he ‘a/
2’ ro
ws of
ort
hogo
na
l t
ran
s
form
m
at
rix
X
starti
ng
f
r
o
m
‘a/2+
1’
th
ro
w up to last
row.
H
WT
offe
rs
be
tt
er
perf
or
m
a
nce
in
im
age
com
pr
essio
n
than
the
or
t
hogonal
t
ra
ns
f
or
m
s
us
ed
t
o
gen
e
rate
them
[
9
,
10
]
.
H
WT
is
al
so
us
e
d
f
or
wate
r
m
ark
i
ng
[
11
]
a
nd
t
o
c
onve
rt
colo
r
im
age
to
gray
i
m
age
[1
2
]
.
V
ario
us
cl
assifi
e
rs
base
d
on
K
NN,
S
VM
an
d
NN
[1
3
,
1
4
]
hav
e
bee
n
use
d
f
or
ver
i
ficat
ion
of
sign
at
ur
es
.
In
[
1
5
]
,
KNN
cl
assifi
er
was
use
d
with,
H
WTs
of
the
pressu
re
m
ap
of
onli
ne
sign
at
ur
es
as
f
eat
ur
e
vecto
r.
It
offere
d
an
EER
of
30%.
I
n
[1
6
],
SV
M
cl
assife
r
was
us
ed
with
,
a
kernel
f
unct
ion
of
onli
ne
si
gn
at
ur
e
tim
e
series,
ba
sed
on
LCSSs
detect
ion,
as
a
featu
re
vector
.
It
offer
e
d
an
EER
of
6.84%.
Using
S
VM
i
n
conj
un
ct
io
n
wi
th
HMM
offere
d
FA
R
of
1.9
6%
an
d
FRR
of
60.43%.
In
[1
7
]
,
ne
ur
al
ne
twork
cl
assifi
e
r
wa
s
us
e
d
with,
t
he
appr
ox
im
at
ion
and
detai
l
com
pone
nt
of
D
W
T
of
t
he
pe
n
posti
on
a
nd
pen
m
ov
em
ent
ang
le
as
featur
e
vector.
Usi
ng
al
l
co
eff
ic
ints
of
D
WT,
suc
cess
r
at
e
was
100%
with
trai
ne
d
sign
at
ur
e,
90%
with
un
t
raine
d
sig
na
tures
a
nd
FR
R
of
24%.
Usi
ng
sel
ect
ed
25
coeffic
ie
nts
of
D
WT,
su
cces
s
rate
was
100%
wit
h
trai
ned sig
natu
re, 9
5% wit
h
untrai
ne
d
si
gnat
ur
es
and
FA
R
of 8
%
.
In
t
his
pa
per,
we
pr
opos
e
a
m
et
ho
d
for
onli
ne
sign
at
ur
e
ver
ific
at
io
n
us
i
ng
Hyb
rid
Wa
velet
Tra
ns
f
orm
an
d
Hidden
Ma
rko
v
Mo
del
cl
assifi
er.
T
he
pro
po
s
ed
m
et
ho
d
is
sh
own
in
Fig
ure
1
.
W
e
hav
e
us
e
d
S
V
C200
4
database
w
hic
h
is
a
la
rg
e
database
c
onta
ining
sig
natu
res
f
ro
m
4
0
in
div
id
uals.
It
ha
s
tot
al
of
1,6
00
sig
natu
res
,
ob
ta
ine
d
us
in
g
a
W
ac
om
In
tu
os
ta
blet.
It
co
ns
ist
s
of
20
ge
nu
i
ne
a
nd
20
f
orger
y
si
gn
at
ures
colle
ct
ed
for
ea
c
h
per
s
on.
Ge
nu
i
ne
si
gn
at
ur
es
a
re
c
ollec
te
d
in
two
dif
fer
e
nt
s
essions.
Forg
e
r
ie
s
for
eac
h
pe
rson
a
re
pro
vide
d
by
at
le
ast
four
ot
her
in
div
id
uals
from
the
data
base.
The
pe
r
f
or
m
ance
res
ults
of
var
i
ou
s
si
gn
at
ur
e
ve
rific
at
ion
syst
e
m
s
that
p
arti
ci
pated
in
the
SV
C
2004
c
om
petit
ion
is
avail
able.
Th
e
best
perform
a
nce
f
or
40
a
va
il
able
us
ers
is
ave
ra
ge
EER
6.9
0%
with
sta
nda
rd
dev
ia
ti
on
of
9.45%,
m
ini
m
u
m
value
of
0.0
0
a
nd
m
axim
u
m
value
of
50.
00
%
.
T
he
best
pe
rfo
rm
ance
f
or
60
othe
r
us
e
rs
is
a
verage
EER
2.89%
with
sta
nd
a
r
d
de
viati
on
of
5.69%,
m
ini
m
u
m
val
ue
of
0.0
0
a
nd
m
axi
m
u
m
va
lue
of
30
.00%
[1
8
]
.
E
ver
y
si
gn
at
ur
e
sam
ple
con
s
ist
of
X
-
c
oor
din
at
e
-
s
cal
ed
cu
rs
or
posit
ion
al
ong
the
x
-
a
xis,
Y
-
c
oor
din
at
e
-
scal
e
d
c
ur
s
or
posit
ion
al
ong
the
y
-
axi
s
,
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
On
li
ne
sig
natu
re veri
fi
cation
us
in
g hy
br
id w
avelet
tra
ns
for
m
(
Ma
noj C
ha
van
)
1825
Ti
m
e
stam
p
-
sy
stem
tim
e
at
t
he
tim
e
of
sig
ning
,
B
utton
st
at
us
-
c
urren
t
butt
on
sta
tus
(0
f
or
pe
n
-
up
a
nd
1
f
or
pen
-
dow
n)
,
A
zim
uth
-
cl
oc
kwise
ro
ta
ti
on
of
cu
rsor
a
bout
the
z
-
a
xis,
Alt
it
ud
e
-
a
ngle
up
ward
t
ow
a
rd
t
he
+
ve
z
-
axis,
Pr
es
sur
e
-
norm
al
pr
ess
ur
e
a
ppli
ed
by
hand
.
Pr
es
sur
e
app
li
e
d
by
t
he
ti
p
of
th
e
pen
on
the
pressure
sensiti
ve
pa
d
is
us
ed
f
or
generati
n
g
the
fea
ture
vect
or
.
W
e
hav
e
us
e
d
Discrete
Cosin
e
transfor
m
(D
CT)
,
Discrete
Ha
rtle
y
transfor
m
(DHT),
Disc
rete
Walsh
tra
nsfo
r
m
(DW
T
)
a
nd
Discrete
Ke
kre
trans
form
(D
KT)
t
o
form
the
H
WT
-
1
an
d H
WT
-
2
m
at
rix
.
Th
e
outpu
t
of HWT is
g
ive
n
t
o HM
M for classi
fic
at
ion
.
Figure
1.
Pro
pose
d
Syst
em
2.
RESEA
R
CH MET
HO
D
S
ign
at
ur
e
data
base
of
1600
sign
at
ur
es
,
pr
ovide
d
by
The
First
I
nter
natio
nal
Sig
natu
re
Ver
ific
at
io
n
Com
petition
(
SV
C
20
04)
is
us
e
d
.
It
has
si
gn
at
ur
es
of
4
0
us
ers
.
E
ver
y
us
er
has
40
sign
at
ur
es
,
out
o
f
w
hich,
20
a
re
ge
nuin
e
and
20
a
re
s
ki
ll
ed
fo
r
ge
ries.
The
fir
st
128
sam
ples
of
eve
ry
sign
at
ur
e
is
us
ed
t
o
fin
d
H
WT.
The
si
gn
at
ur
es
,
ha
ving
sam
ples
le
ss
tha
n
128,
will
be
pa
dded
with
ze
ro
es
.
The
first
16
s
a
m
ples
of
the
H
WT
ou
t
pu
t
are
us
e
d
as
featu
re
ve
ct
or
.
Discrete
Cosine
tran
sf
or
m
(D
CT),
Discrete
Ha
rtle
y
transfor
m
(D
HT
),
Discrete
Wals
h
tra
nsfo
rm
(DWT)
an
d
Disc
r
et
e
Kekre
t
ransform
(D
K
T)
a
re
us
ed
to
f
orm
the
H
WT
[
1
9
,
20
].
Ther
e
are
m
any
top
ol
ogie
s
of
HMM
s
uch
as
L
eft
to
Ri
ght,
E
rgo
dic
an
d
Ri
ng.
Le
ft
to
Ri
ght
topol
ogy
as
sh
ow
n
in
Fi
g
ure
2
,
is fo
und
t
o be
best s
ui
te
d for the
Sig
natu
re m
od
el
li
ng
[
2
1
, 22
]
.
Figure
2.
Left t
o
Ri
ght
HMM
m
od
el
HMM
is
rep
re
sented
by
the
transiti
on
pro
ba
bili
ty
m
at
rix
(A),
O
bs
e
rv
at
i
on
m
at
rix
(B)
and
init
ia
l
pro
bab
il
it
y
distrib
ution
m
at
rix
(π)
.
[2
3,
24
]
Con
si
der
a
sys
tem
wh
ic
h
is
in
a
disti
nct
sta
te
(S1,
S2,
..
.,
SN
)
at
any
point
of
t
i
m
e.
In
this
e
xp
e
rim
ent
the
nu
m
ber
of
st
at
es
(N)
of
th
e
m
od
el
are
va
ried
f
r
om
2
to
5.
As
the
num
ber
of
st
at
es
increase,
the
tim
e
need
e
d
f
or
trai
ning
increase
s.
T
he
nu
m
ber
of
obse
rv
at
i
ons
(M
)
corres
pondin
g
to
each
sta
te
are
va
ried
fro
m
20
0
t
o
75
0
in
the
inc
rem
e
nts
of
25.
Th
e
ou
t
pu
t
of
H
WT
is
a
m
a
trix
of
dim
ension
[
1×128
]
.
The
m
at
ri
x
el
em
ents
fro
m
1
to
16
are
c
hosen
a
s
a
fe
at
ur
e
vecto
r.
F
eat
ur
e
vecto
rs
a
re s
ca
le
d
into
M
num
ber
o
f o
bs
er
va
ti
on
s.
In
it
ia
l
Pr
oba
bili
ty
Distribu
ti
on
(π):
π
i
=
P
(
q
1
=
S
i
);
1
≤
i
≤
N.
We
ass
um
e
the
init
ia
l
pro
bab
il
it
y
of
the
fir
st
sta
te
is
1
a
nd
th
e
ot
he
rs
a
re
0
w
hic
h
im
plies
that
in
the
be
ginning
HMM
is
al
ways
in
sta
te
1.
Stat
e
transiti
on
pro
ba
bili
ty
(a
ij
):
a
ij
=
P
(S
t
=j
/
S
t
-
1
=
i).
Fo
r
the
le
ft
-
to
-
rig
ht
H
MM
,
a
ij
=0
when
i>
j
.
we
a
re
us
ing
the
HMM
of
first
or
der
s
o
that
a
ij
=0
wh
e
n
j>
i+
1.
I
niti
al
ly
,
the
sta
te
t
ran
sit
io
n
m
at
ri
x
is
ge
ne
r
at
ed
us
i
ng
the
ra
ndom
num
ber
s
su
c
h
tha
t
∑
a
ij
=
1
=
1
;
1
≤
i
≤
N.
Ob
se
r
vation
pro
bab
il
it
y
(b
j
)
:
b
j
(
k)
=
P
(
V
k
at
t
/
q
t
=
S
j
);
1 ≤
j
≤ N;
1 ≤
k ≤ M;
the
pro
bab
il
it
y of ge
ne
rati
ng a sym
bo
l V
k
in stat
e
j
.
S
ta
ti
sti
cs
and
m
achine
le
arn
i
ng
t
oo
l
box
of
t
he
MAT
L
AB
13
was
us
ed
for
i
m
ple
m
entat
i
on
of
HMM
.
In
it
ia
ll
y
a
r
an
dom
l
y
gen
e
rated
tra
ns
it
ion
pr
ob
a
bili
ty
Ma
trix
(
A)
is
ge
ne
rated
us
in
g
M
ATL
AB
.
We
assum
e
o
bse
r
vation
prob
a
bili
ty
m
at
ri
x
(B)
t
o
hav
e
equ
al
pro
ba
bili
ty
fo
r
e
ver
y
sy
m
bo
ls
and
H
MM
to
be
in
s
ta
te
1.
HMM
is
trai
ned
us
i
ng
the
f
un
ct
io
n
‘
hmm
t
rain
’
for
3
to
20
genuin
e
trai
nin
g
si
gn
at
ur
e
sa
m
ples,
nu
m
ber
of
sta
te
s
from
2
t
o
5
a
nd
sym
bols
from
200
t
o
750.
A
fter
HM
M
is
trai
ne
d,
it
is
use
d
to
te
st
2
0
ge
nu
i
ne
a
nd
20
forg
e
d si
gnat
ures of
40 u
s
ers
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
2
,
A
pr
i
l 202
0
:
1823
-
1832
1826
3.
RESU
LT
S
A
ND
DI
SCUS
S
ION
Perfo
rm
ance
of
the
syst
em
will
be
m
easur
e
d
on
t
he
basis
of
False
Re
j
ect
ion
Ra
ti
o
(F
R
R)
an
d
False
Accepta
nce
R
at
io
(F
AR
).
F
RR
ref
ers
t
o
false
re
j
ect
ion
of
genuin
e
s
ign
at
ure
a
nd
FA
R
re
fe
rs
t
o
false
acce
ptance
of
f
orge
d
sig
natu
r
e
[
2
5
]
.
FRR
is
com
pu
te
d
as
ra
ti
o
of
t
he
num
ber
of
si
gnat
ures
detect
ed
as
f
orge
d
to
the
total
nu
m
ber
of
ge
nuine
sig
natu
res
te
ste
d.
F
AR
is
com
pu
te
d
as
rati
o
of
the
num
ber
of
si
gnat
ur
e
s
detect
ed
as
ge
nu
i
ne
to
t
he
to
ta
l
fo
r
ge
d
sig
na
tures
te
ste
d.
Test
ing
has
be
en
car
ried
out
for
40
us
e
rs
a
nd
t
he
n
the
ave
rag
e
F
RR
and
F
AR
a
re
cal
culat
ed
.
I
n
FRR
-
F
AR
plo
t
sho
wn
i
n
Fi
gure
3
,
the
point
w
her
e
t
wo
gr
a
phs
cro
ss
eac
h
othe
r
is
ref
er
red
a
s
Equ
al
Er
ror
Ra
te
(EER).
A
t
this
po
int
the
va
lue
of
FRR
and
F
AR
is
m
i
nim
u
m
.
The
res
ults
ob
t
ai
ned
by
the
fi
rst
1
–
16
sam
pl
es
of
H
WT
-
1
and
2
for
DCT
,
DH
T
,
HAAR
,
HADA
M
AR
D
an
d
KEK
R
E
c
om
bi
nations i
s s
ho
wn in t
he
Ta
bl
e
s
2
-
4.
Figure
3
.
FRR
–
F
AR
plo
t
The
c
om
par
iso
n of H
WT
-
1 an
d 2 fo
r 1
–
16
bit f
or
DCT
c
ombinati
ons is
show
n
i
n
the
Ta
bl
e
s
2
-
4
.
-
Be
st
FRR
–
F
A
R:
FRR
–
FA
R
shou
l
d
be
a
s
low
as
po
s
sible.
F
or
H
WT
-
1,
DCT
HAAR
offer
s
bes
t
perform
ance
with
FRR
&
FA
R
of
0
%.
F
or
H
WT
-
2,
DCT
KEK
R
E
offer
s
best
perform
a
nce
with
FRR
&
FA
R
of
9%
.
The
perf
or
m
a
nce
offer
e
d
by
DCT
HAA
R
H
WT
f
or
H
WT
-
1
is
bet
te
r
tha
n
H
WT
-
2.
The
pe
rfor
m
ance
offe
red
by
DCT
com
bin
at
ion
s
f
or
H
WT
-
1
is
be
tt
er
than
H
W
T
-
2.
For
H
W
T
-
1,
the
pe
rfor
m
ance
offer
e
d
by
DCT
c
om
bin
at
ion
s
e
xcep
t
DCT
D
HT
is
bette
r
tha
n
Or
t
hogonal
D
CT
trans
form
.
Fo
r
H
W
T
–
2,
only
DCT
KEKR
E
of
fe
rs
bett
er
perform
ance
than
DCT
c
om
bin
at
ion
s
tha
n
Or
t
hogonal
D
CT t
ran
s
f
or
m
.
-
Be
st
Nu
m
ber
of
Trai
ning
S
a
m
ples
:
The
nu
m
ber
of
tra
ining
sam
ple
s
shou
l
d
be
as
low
as
possible
.
Fo
r
H
WT
–
1,
DCT
D
HT
off
ers
be
st
perf
orm
ance
of
12
c
om
par
ed
to
15
trai
ning
sam
ples
for
O
rth
ogon
al
DCT
tra
ns
f
orm
.
Fo
r
H
WT
–
2,
Or
t
hogonal
DCT
tra
ns
f
orm
of
fe
rs
be
st
perform
ance
of
4
tr
ai
ning
sa
m
ples
com
par
ed
to
al
l
co
m
bin
at
ion
s
of
DCT
H
WT
.
The
pe
rfor
m
ance
offe
red
by
DCT
com
bin
at
ion
s
f
or
H
WT
-
2
is bett
er
th
an
HWT
-
1.
-
Be
st
sta
te
wise
FRR
–
F
AR:
FRR
–
F
AR
shou
l
d
be
as
lo
w
as
po
s
sible
f
or
the
giv
e
n
sta
te
fr
om
2
to
5.
Fo
r H
WT
-
1,
DC
T H
A
AR o
f
f
ers
best p
e
rform
ance f
or
2,
4
and
5
sta
te
s,
D
CT KEK
RE f
or stat
e 3
an
d
D
CT
HAAR
f
or
sta
te
5
com
par
e
d
to
or
t
hogona
l
DCT
trans
f
o
rm
.
Fo
r
H
W
T
-
2,
DCT
KE
KRE
offe
rs
be
st
perform
ance
fo
r
2
to
5
sta
te
s
com
par
ed
to
or
t
hogonal
DC
T
transfo
rm
.
T
he
pe
rfor
m
ance
of
fe
re
d
by
D
C
T
com
bin
at
ion
s
f
or HWT
-
1 i
s bet
te
r
than
HW
T
-
2.
-
Be
st
Nu
m
ber
of
Sym
bo
l:
It
s
hould
be
as
lo
w
as
possible.
Test
ing
wa
s
carried
ou
t
f
or
num
ber
of
sym
bo
ls
from
20
0
to
75
0.
It
evi
den
t
th
at
the
best
per
f
or
m
ance
in
te
rm
s
of
FRR
–
F
AR,
A
AR
–
AR
R,
EER
is
of
fe
red
by
275
-
325
sy
m
bo
ls
fo
r
H
W
T
-
1
a
nd
45
0
–
500
sym
bo
l
fo
r
H
WT
-
2.
The
per
f
or
m
ance
offer
e
d
by
DC
T
com
bin
at
ion
s
f
or HWT
-
1 i
s bet
te
r
than
HW
T
-
2.
The
c
om
par
iso
n of H
WT
-
1 an
d 2 fo
r 1
–
16
bit for
DHT c
ombinati
ons is
show
n
i
n
the
Ta
bl
e
2
-
4
.
-
Be
st
Nu
m
ber
of
Traini
ng
Sam
ples
:
Fo
r
H
WT
-
1,
DHT
KE
K
RE
of
fe
rs
best
perform
ance
with
FRR
&
FA
R
of
5
%
.
For
H
W
T
-
2,
D
H
T
KEK
RE
offer
s
be
st
perf
or
m
ance
with
FRR
13
%
&
FA
R
of
19%.
The
perform
ance
offer
e
d
by
DH
T
KE
KRE
H
WT
f
or
H
WT
-
1
is
bette
r
than
H
WT
-
2.
The
perform
a
nce
offer
e
d
by DH
T co
m
bin
at
ions fo
r H
WT
-
1
is bett
er th
an
H
WT
-
2.
F
or HWT
–
1, the p
er
form
ance o
ff
e
r
ed
by
DH
T
DCT,
D
HT
KE
KRE
is
bette
r
than
O
rth
ogonal
DHT
transf
or
m
.
Fo
r
H
WT
–
2,
on
ly
DH
T
KEKR
E
offer
s
b
et
te
r
p
e
rfor
m
ance tha
n Or
t
hogo
nal DHT tra
nsfo
rm
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
On
li
ne
sig
natu
re veri
fi
cation
us
in
g hy
br
id w
avelet
tra
ns
for
m
(
Ma
noj C
ha
van
)
1827
-
Be
st
Nu
m
ber
of
Traini
ng
S
a
m
ples
:
Fo
r
H
WT
–
1,
D
HT
HAD
AMAR
D
offers
be
st
pe
rfor
m
ance
of
10
trai
ning
sam
pl
es
com
par
ed
to
13
trai
ning
sam
ples
fo
r
O
rth
ogonal
DHT
transfo
rm
.
F
o
r
H
WT
–
2,
D
H
T
HAD
AMAR
D
offer
s
best
pe
rfor
m
ance
of
5
trai
ni
ng
sa
m
ples
co
m
par
ed
to
7
trai
ni
ng
sam
ples
fo
r
Or
t
hogonal
D
HT
tra
nsfo
rm
. D
HT
c
om
bin
at
ion
s
offe
r bett
er
perform
ance
for H
WT
-
2
tha
n H
WT
-
1.
-
Be
st
sta
te
wise
FRR
–
F
AR:
Fo
r
H
WT
-
1,
D
HT
K
EKRE
offer
s
best
pe
rfo
rm
ance
fo
r
2,
3
an
d
4
sta
te
s
a
nd
DH
T
DCT
f
or
sta
te
5
c
om
par
ed
to
or
t
hogo
nal
D
HT
tran
s
form
.
Fo
r
H
W
T
-
2,
DHT
K
E
KRE
offer
s
be
st
perform
ance
fo
r
2
to
5
sta
te
s
com
par
ed
to
or
th
ogonal
D
HT
tra
ns
f
or
m
.
DH
T
c
om
bin
at
ion
s
offe
r
be
tt
er
perform
ance
for
H
WT
-
1
tha
n H
WT
-
2
.
-
Be
st
Nu
m
ber
of
Sym
bo
l:
Test
ing
wa
s
carried
ou
t
for
num
ber
of
sym
bo
ls
from
20
0
to
750.
It
evid
ent
th
at
the
best
pe
rform
ance
in
te
rm
s
of
FRR
–
FAR
,
A
AR
–
ARR
,
EER
is
offer
e
d
by
275
sym
bo
ls
for
H
WT
-
1
a
nd
45
0
–
500
sy
m
bo
ls
fo
r
H
WT
-
2.
D
HT
c
ombinati
ons
offers
bet
te
r
pe
rfo
r
m
ance
for
H
W
T
-
1
than H
WT
-
2.
The
c
om
par
iso
n of H
WT
-
1 an
d 2 fo
r 1
–
16
bit for
HAAR c
om
bin
at
ion
s is s
how
n
in
the
Ta
ble
s
2
-
4
.
-
Be
st FRR
–
FAR
:For
HWT
-
1,
H
A
AR D
CT a
nd
HAAR K
E
KRE off
e
rs
be
st per
f
or
m
ance
w
it
h
FRR
&
FA
R
of
0
%.
For
H
WT
-
2,
H
A
A
R
KEK
RE
of
fer
s
best
perf
or
m
ance
with
FRR
11
%
&
FA
R
of
12%.
The
per
form
ance
offer
e
d
by
H
AA
R
DCT
an
d
H
AA
R
KEKR
E
H
W
T
for
H
WT
-
1
is
bette
r
than
H
WT
-
2.
The
pe
rfor
m
ance
offer
e
d
by
HA
AR
com
bin
at
io
ns
f
or
H
WT
-
1
is
bette
r
than
H
WT
-
2.
For
H
WT
–
1,
the
perform
ance
of
fe
re
d
by
al
l
co
m
bin
at
i
on
s
of
HAAR
is
bette
r
than
Or
th
ogonal
HAAR
trans
f
orm
.
Fo
r
H
WT
–
2,
HAAR
D
HT
and
H
AA
R
K
EKRE
offer
s
bette
r
pe
rfo
rm
ance
tha
n
Or
t
hogonal
HAA
R
trans
form
.
-
Be
st
Nu
m
ber
of
Traini
ng
Sa
m
ples
:
Fo
r
H
WT
–
1,
H
A
AR
D
H
T
offer
s
be
st
perform
ance
of
1
2
trai
ning
sam
ples
with
FRR
,
FA
R
of
15%,
15
%
r
es
pecti
vely
com
par
e
d
to
13
tra
ining
sam
ples
with
FRR
,
F
A
R
of
10%,
30%
r
espe
ctively
for
O
rth
ogonal
DHT
trans
form
.
Fo
r
H
WT
–
2,
H
AA
R
HAD
A
MARD
off
ers
best
perform
ance
of
5
trai
ni
ng
sa
m
ples
co
m
pa
red
to
6
trai
ning
sam
ples
fo
r
Or
th
ogonal
H
AA
R
tra
ns
f
or
m
.
HAAR
c
om
bin
at
ion
s
offer bet
te
r
pe
rfor
m
anc
e
f
or
H
WT
-
2
t
han HWT
-
1.
-
Be
st
sta
te
wise
FRR
–
F
AR
:
F
or
H
WT
-
1,
H
AA
R
KE
KRE
offer
s
best
perf
or
m
ance
f
or
2
to
5
sta
te
s.
H
A
AR
DCT
offer
s
be
st
perf
or
m
ance
f
or
sta
te
5
c
om
par
ed
to
ort
hogo
nal
H
A
AR
trans
f
or
m
.
For
H
W
T
-
2,
H
A
AR
KEK
R
E
offers
best
per
f
orm
a
nce
f
or
2
to
5
sta
te
s
co
m
par
ed
to
ort
hogo
nal
HAAR
tra
ns
f
or
m
.
HAA
R
com
bin
at
ion
s
offe
r bett
er per
f
or
m
ance
f
or HWT
-
1
t
han HWT
-
2.
-
Be
st
Nu
m
ber
of
Sym
bo
l
:
Test
ing
wa
s
carried
ou
t
for
num
ber
of
sym
bo
ls
from
20
0
to
750.
It ev
ide
nt that the b
est
per
form
anc
e
in term
s o
f
FRR
–
F
AR, AAR
–
ARR, E
ER i
s o
ff
e
red
by
2
75
sym
bo
ls for
H
WT
-
1
an
d
450
–
500
sym
bo
ls
fo
r
H
WT
-
2.
HAAR
com
bin
at
ions
offe
r
be
tt
er
per
f
orm
a
nce
f
or
H
WT
-
1
than H
WT
-
2.
The
com
par
is
on
of
H
WT
-
1
and
2
f
or
1
–
16
bit
for
H
A
D
AMAR
D
com
bin
at
ion
s
is
sh
own
in
the Ta
ble
2
-
4
.
-
Be
st
FRR
–
FAR
:
Fo
r
H
WT
-
1,
H
ADAM
ARD
K
EKRE
off
ers
be
st
pe
rform
ance
with
F
RR
&
FA
R
of
0%
.
Fo
r
H
WT
-
2,
HAD
AMAR
D
KE
KRE
off
ers
be
st
pe
rfor
m
ance
with
FRR
2
5%
&
FA
R
of
22%
.
The
pe
rfo
rm
a
nce
offe
red
by
HAD
AMARD
KEK
RE
H
W
T
for
H
WT
-
1
is
be
tt
er
than
H
W
T
-
2.
The
perform
ance
offe
red
by
H
ADAMAR
D
c
om
bin
at
ion
s
f
or
H
WT
-
1
is
be
tt
er
than
H
WT
-
2.
F
or
H
WT
–
1,
the
perform
ance
of
fe
re
d
by
al
l
co
m
bin
at
ion
s
of
H
ADAM
ARD
exce
pt
H
ADA
MAR
D
DCT
is
bette
r
tha
n
Or
t
hogonal
H
ADAMAR
D
trans
f
or
m
.
Fo
r
H
W
T
–
2,
the
per
f
or
m
ance
offer
e
d
by
al
l
com
bin
at
ion
s
of
HAD
AMAR
D i
s b
et
te
r
th
an
Orth
ogonal
HA
DA
M
ARD
tra
ns
f
or
m
-
Be
st
Nu
m
ber
of
T
raini
ng
Sam
ples
:
Fo
r
H
WT
–
1,
Or
t
hogonal
HAD
AMAR
D
tra
nsfo
rm
of
fers
best
perform
ance
of
8
trai
ning
sa
m
ples
co
m
par
ed
to
al
l
com
bin
at
ion
s
of
H
A
DA
M
ARD
H
WT.
F
or
H
W
T
–
2,
Orthogonal
H
A
DA
M
ARD
tra
ns
f
or
m
offer
s
best
pe
rfor
m
ance
of
5
trai
ning
sam
ples
com
par
ed
to
al
l
com
bin
at
ion
s
of
HA
D
AMA
RD
H
WT.
H
A
DA
M
ARD
c
om
bin
at
ion
s
off
er
bette
r
perf
orm
ance
for
H
WT
-
2
than H
WT
-
1.
-
Be
st
sta
te
wise
FRR
–
F
AR:
Fo
r
H
WT
-
1,
HAD
AMAR
D
KE
KRE
of
fe
rs
be
st
pe
rform
ance
for
2
an
d
5
sta
te
s.
H
A
DA
M
ARD
D
HT
offers
be
st
perf
or
m
ance
f
or
sta
te
4
an
d
5
c
om
par
ed
to
ort
ho
gonal
H
A
D
AMAR
D
trans
form
.
For
H
WT
-
2,
H
A
DA
M
ARD
KE
KRE
offe
rs
be
st
perform
ance
for
2
to
5
sta
te
s
com
par
ed
t
o
o
rthogona
l
H
A
D
AMAR
D
tra
nsfo
rm
.
HADAM
ARD
c
om
bin
at
ions
offer
be
tt
er
pe
rfor
m
ance
for H
WT
-
1
tha
n H
WT
-
2
.
-
Be
st
Nu
m
ber
of
Sy
m
bo
l
:
Test
ing
wa
s
carrie
d
out
for
num
be
r
of
sym
bo
ls
from
20
0
to
75
0.
It
evi
den
t
th
at
the
best
perfor
m
ance
in
te
r
m
s
of
FRR
–
F
AR,
AA
R
–
ARR,
E
ER
is
of
fer
e
d
by
27
5
-
30
0
sy
m
bo
ls
f
or
HW
T
-
1
and
500
sym
b
ols
f
or
H
WT
-
2.
HAD
AMA
RD
com
bin
at
ion
s
offe
r
bette
r
pe
rfo
rm
ance
for
H
W
T
-
1
tha
n
H
WT
-
2
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
2
,
A
pr
i
l 202
0
:
1823
-
1832
1828
The
c
om
par
iso
n
of HWT
-
1 an
d 2 fo
r 1
–
16
bit for
KE
KRE c
om
bin
at
ion
s is
sh
ow
n
in
the
T
able
s
2
-
4
.
-
Be
st
FRR
–
F
A
R:
Fo
r
H
W
T
-
1,
KE
KRE
D
CT,
K
EKRE
HAAR
a
nd
K
EKRE
128
offe
rs
best
perfor
m
ance
with
FRR
&
F
AR
of
0
%
.
F
or
H
WT
-
2,
K
EKRE
128
off
ers
best
pe
rfo
r
m
ance
with
F
RR
5%
&
F
A
R
of
2%.
T
he
p
erf
o
rm
anc
e
offe
re
d
by
KE
KRE
DCT,
KE
KRE
HAAR
a
nd
KEK
R
E1
28
H
WT
for
H
WT
-
1
i
s
bette
r
than
H
WT
-
2.
T
he
pe
r
form
ance
of
fe
r
ed
by
KE
KRE
com
bin
at
ion
s
f
or
H
WT
-
1
is
be
tt
er
than
H
W
T
-
2.
Fo
r
H
WT
–
1,
the
pe
rfo
rm
ance
offer
e
d
by
K
EKRE
12
8
is
bette
r
tha
n
al
l
c
om
bin
at
ion
s
of
KE
KRE
H
WT.
Fo
r
HWT
–
2,
t
he per
form
ance off
e
red by
K
EKRE
128 i
s bet
te
r
than
all
co
m
bin
at
ion
s
of K
E
KRE
H
WT
.
-
Be
st
Nu
m
ber
of
Traini
ng
Sa
m
ples
:
Fo
r
H
WT
–
1,
K
EKR
E
D
HT
offer
s
best
pe
rfo
rm
a
nce
of
11
trai
ning
sam
ples
co
m
par
ed
t
o
20
tr
a
ini
ng
sam
ples
for
O
rth
ogona
l
KEK
RE
t
ra
ns
f
or
m
.
For
H
WT
–
2,
KE
K
RE
HAD
AMAR
D
offe
rs
best
pe
rfor
m
ance
of
6
trai
ning
sa
m
ples
co
m
pared
to
16
trai
ni
ng
sam
ples
fo
r
Or
t
hogonal
K
EKRE
trans
f
orm
.
KEK
RE
co
m
bin
at
ion
s
off
er
bette
r
perf
orm
ance
fo
r
H
WT
-
2
is
bette
r
th
an
H
WT
-
1.
-
Be
st
sta
te
wise
FRR
–
F
AR:
Fo
r
H
WT
-
1,
KE
KRE
12
8
offe
r
s
best
pe
rfo
rm
ance
f
or
2
T
O
5
sta
te
s
com
par
e
d
to
combinat
ions
of
KE
KRE
H
WT.
F
or
H
W
T
-
2,
KE
KRE
128
offe
rs
bes
t
per
f
or
m
ance
for
2
TO
5
sta
te
s
com
par
ed
to
c
om
bin
at
ion
s
of
KE
KRE
H
W
T.
KE
KRE
c
om
bin
at
ion
s
off
er
bette
r
pe
rform
anc
e
for
H
WT
-
1
is bett
er tha
n H
WT
-
2.
-
Be
st
Nu
m
ber
of
Sy
m
bo
l:
Test
ing
wa
s
carrie
d
out
for
num
ber
of
sym
bo
ls
from
20
0
to
75
0.
It
evi
den
t
th
at
the
best
per
for
m
anc
e
in
te
rm
s
of
FRR
–
F
AR,
AA
R
–
ARR,
E
ER
is
of
fe
re
d
by
275
sym
bo
ls
for
H
WT
-
1
a
nd
450
-
50
0
sy
m
bo
ls
for
H
WT
-
2
.
KEK
R
E
com
bin
at
io
ns
offe
r
bette
r
perfor
m
ance
fo
r
H
WT
-
1
is
bette
r
t
ha
n
H
WT
-
2.
Table
2
.
Be
st
F
RR
FA
R
for
H
WT
-
1
a
nd
2
HW
T
–
1
HW
T
-
2
Co
m
b
in
atio
n
s
States
Sy
m
b
o
ls
Tr
ain
in
g
san
p
les
FRR
FAR
Co
m
b
in
atio
n
s
States
Sy
m
b
o
ls
Tr
ain
in
g
san
p
les
FRR
FAR
DCT 12
8
2
300
16
5
10
DCT 12
8
5
475
11
27
32
DCT D
HT
3
300
13
10
10
DCT D
HT
5
400
10
31
34
DCT Ha
ar
5
325
18
0
0
DCT Ha
ar
4
425
9
30
33
DCT
Had
a
m
a
rd
5
300
17
0
5
DCT
Had
a
m
a
rd
5
400
10
30
42
DCT Kek
re
3
300
17
5
5
DCT Kek
re
5
500
18
9
9
DHT
1
2
8
4
350
15
5
10
DHT
1
2
8
5
500
12
24
30
DHT
DCT
5
275
18
5
5
DHT
DCT
4
500
9
31
37
DHT
Haar
5
275
17
5
10
DHT
Haar
5
500
10
29
37
DHT
Had
a
m
a
rd
4
300
14
10
15
DHT
Had
a
m
a
rd
5
500
11
31
28
DHT
Kek
re
4
275
20
0
5
DHT
Kek
re
5
475
16
13
19
Haar
12
8
2
275
13
10
30
Haar
12
8
5
475
13
24
28
Haar
DCT
5
275
20
0
0
Haar
DCT
5
500
11
31
24
Haar
DHT
5
300
15
5
15
Haar
DHT
5
450
11
26
21
Haar
Had
a
m
a
rd
2
300
14
5
10
Haar
Had
a
m
a
rd
5
425
10
30
33
Haar
Kek
re
3
275
19
0
0
Haar
Kek
re
5
500
17
11
12
Had
a
m
a
rd
128
5
350
14
5
15
Had
a
m
a
rd
128
5
475
9
32
32
Had
a
m
a
rd
DCT
5
300
17
10
10
Had
a
m
a
rd
DCT
5
500
12
27
26
Had
a
m
a
rd
DHT
4
300
14
5
5
Had
a
m
a
rd
DHT
5
500
10
29
29
Had
a
m
a
rd
Haar
4
275
16
5
5
Had
a
m
a
rd
Haar
4
500
10
31
30
Had
a
m
a
rd
Kek
re
5
275
20
0
0
Had
a
m
a
rd
Kek
re
5
500
14
25
22
Kek
re
1
2
8
2
275
20
0
0
Kek
re
1
2
8
4
475
19
5
2
Kek
re
DC
T
5
275
20
0
0
Kek
re
DC
T
5
500
13
27
27
Kek
re
D
HT
5
275
18
5
5
Kek
re
D
HT
4
350
8
30
36
Kek
re
Ha
ar
2
525
20
0
0
Kek
re
Ha
ar
5
500
14
21
21
Kek
re
Had
a
m
a
rd
2
550
19
100
0
Kek
re
Had
a
m
a
rd
5
500
13
23
21
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
On
li
ne
sig
natu
re veri
fi
cation
us
in
g hy
br
id w
avelet
tra
ns
for
m
(
Ma
noj C
ha
van
)
1829
Table
3
.
Be
st
No.
of
t
rainin
g sam
ples
for
H
WT
-
1
a
nd
2
HW
T
-
1
HW
T
-
2
Co
m
b
in
atio
n
s
States
Sy
m
b
o
ls
Tr
ain
in
g
san
p
les
FRR
FAR
Co
m
b
in
atio
n
s
States
Sy
m
b
o
ls
Tr
ain
in
g
san
p
les
FRR
FAR
DCT 12
8
3
300
15
15
15
DCT 12
8
2
275
4
35
63
DCT D
HT
2
300
12
15
25
DCT D
HT
3
325
7
33
50
DCT
Ha
ar
3
300
13
5
20
DCT Ha
ar
3
375
6
33
47
DCT
Had
a
m
a
rd
2
275
12
30
30
DCT
Had
a
m
a
rd
4
325
6
30
51
DCT Kek
re
3
300
17
5
5
DCT Kek
re
2
450
11
22
30
DHT
1
2
8
2
350
13
10
20
DHT
1
2
8
2
375
7
32
40
DHT
DCT
4
275
15
5
10
DHT
DCT
2
500
8
33
43
DHT
Haar
2
275
13
20
20
DHT
Haar
2
350
6
34
46
DHT
Had
a
m
a
rd
2
300
10
20
25
Dh
t Hada
m
a
rd
4
275
5
29
52
DHT
Kek
re
2
275
19
5
5
DHT
Kek
re
2
475
12
23
27
Haar
12
8
2
275
13
10
30
Haar
12
8
2
325
6
32
57
Haar
DCT
2
275
13
15
0
Haar
DCT
4
350
7
32
52
Haar
DHT
3
275
12
15
15
Haar
DHT
2
375
6
37
49
Haar
Had
a
m
a
rd
3
275
13
10
20
Haar
Had
a
m
a
rd
3
275
5
32
57
Haar
Kek
re
3
275
19
0
0
Haar
Kek
re
3
400
12
20
33
Had
a
m
a
rd 1
2
8
2
275
8
35
30
Had
a
m
a
rd 1
2
8
2
475
5
38
52
Had
a
m
a
rd
DCT
2
275
11
20
15
Had
a
m
a
rd
DCT
2
375
6
30
45
Had
a
m
a
rd
DHT
2
275
11
15
20
Had
a
m
a
rd
DHT
3
300
6
31
49
Had
a
m
a
rd
Haar
2
275
14
5
15
Had
a
m
a
rd
Haar
2
425
8
34
37
Had
a
m
a
rd
Kek
re
2
275
18
10
10
Had
a
m
a
rd
Kek
re
2
500
11
34
31
Kek
re
1
2
8
2
275
20
0
0
Kek
re
1
2
8
2
450
16
11
16
Kek
re
DC
T
2
275
13
10
15
Kek
re
DC
T
2
425
9
31
37
Kek
re
D
HT
2
275
11
15
25
Kek
re
D
HT
4
350
8
30
36
Kek
re
Ha
ar
2
525
20
0
0
Kek
re
Ha
ar
2
500
10
33
38
Kek
re
Had
a
m
a
rd
2
550
19
100
0
Kek
re
Had
a
m
a
rd
2
375
6
34
49
Table
4
.
Be
st
S
ta
te
wise
FRR
FA
R
for
H
WT
-
1
an
d
2
HW
T
-
1
HW
T
-
2
Co
m
b
in
atio
n
s
States
Sy
m
b
o
ls
Tr
ain
in
g
san
p
les
FRR
FAR
Co
m
b
in
atio
n
s
States
Sy
m
b
o
ls
Tr
ain
in
g
san
p
les
FRR
FAR
DCT Ha
ar
2
300
14
5
10
DCT Kek
re
2
450
11
22
30
DCT Kek
re
3
300
17
5
5
DCT Kek
re
3
450
13
19
20
DCT Ha
ar
4
275
15
10
10
DCT Kek
re
4
450
15
17
18
DCT Ha
ar
5
325
18
0
0
DCT
Kek
re
5
500
18
9
9
DHT
Kek
re
2
275
19
5
5
DHT
Kek
re
2
475
12
23
27
DHT
Kek
re
3
275
19
5
5
DHT
Kek
re
3
500
13
22
20
DHT
Kek
re
4
275
20
0
5
DHT
Kek
re
4
450
15
15
14
DHT
DCT
5
275
18
5
5
DHT
Kek
re
5
475
16
13
19
Haar
Kek
re
2
275
19
5
5
Haar
Kek
re
2
450
12
21
35
Haar
Kek
re
3
275
19
0
0
Haar
Kek
re
3
400
12
20
33
Haar
Kek
re
4
275
20
0
5
Haar
Kek
re
4
475
16
14
22
Haar
Kek
re
5
275
20
0
0
Haar
Kek
re
5
500
17
11
12
Had
a
m
a
rd
Kek
re
2
275
18
10
10
Had
a
m
a
rd
Kek
re
2
500
11
34
31
Had
a
m
a
rd
DHT
3
275
12
10
10
Had
a
m
a
rd
Kek
re
3
500
15
16
44
Had
a
m
a
rd
DHT
4
300
14
5
5
Had
a
m
a
rd
Kek
re
4
500
13
25
27
Had
a
m
a
rd
Kek
re
5
275
20
0
0
Had
a
m
a
rd
Kek
re
5
500
14
25
22
Kek
re
1
2
8
2
275
20
0
0
Kek
re
1
2
8
2
450
16
11
16
Kek
re
1
2
8
3
275
20
0
0
Kek
re
1
2
8
3
500
17
10
11
Kek
re
1
2
8
4
275
20
0
0
Kek
re
1
2
8
4
475
19
5
2
Kek
re
1
2
8
5
275
20
0
0
Kek
re
1
2
8
5
475
19
5
2
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
2
,
A
pr
i
l 202
0
:
1823
-
1832
1830
Fr
om
Table
s
2
-
4, w
e
get foll
owin
g
im
po
rtant
r
es
ults.
-
FRR
–
F
AR:
For
H
WT
-
1,
DC
T
HAAR
offe
r
s
best
perf
or
m
ance
with
FRR
and
F
AR
of
0
%.
For
H
WT
-
2,
KEK
R
E
128
offe
rs
best
pe
rfo
rm
ance
of
FRR
5
%
and
F
A
R
2
%.
H
WT
-
1
offe
rs
bette
r
perform
ance
t
ha
n
H
WT
-
2.
-
Nu
m
ber
of
trai
ning
sam
ples
:
Fo
r
H
WT
–
1,
Or
t
hogonal
H
ADAMAR
D
tr
ansfo
rm
of
fer
s
best
pe
rfo
rm
a
nce
of
8
tr
ai
n
ing
.
Fo
r
H
WT
–
2,
Or
t
hogonal
D
CT
trans
form
offer
s
best
perform
ance
of
4
tr
ai
ning
sam
ples.
H
WT
-
2
offer
s
bette
r per
form
ance tha
n H
W
T
-
1.
-
Stat
e
wise
FR
R
–
FA
R:
For
H
WT
-
1,
KE
K
RE
12
8
offer
s
best
perform
ance
f
or
2
to
5
s
ta
te
s.
For
H
W
T
-
2,
KEK
R
E
128
offe
rs
best
pe
rfor
m
ance
for
2
to
5
sta
te
s.
H
WT
-
1
offe
rs
be
tt
er
per
f
orm
a
nce
tha
n
H
WT
-
2.
As
t
he
num
ber
of
sta
te
s
inc
re
ase,
the
pe
rfo
r
m
ance
of
the
s
yst
e
m
i
m
pr
ov
e
s.
H
W
T
-
1
f
ou
nd
to
offe
r
bette
r
perform
ance f
or
3
to
5 sta
te
s a
nd HWT
-
2 f
or
5
sta
te
s.
-
Nu
m
ber
of
Sy
m
bo
l:
Fo
r
H
WT
-
1,
K
EKR
E
128
offe
rs
best
pe
r
form
a
nce
at
275
sy
m
bo
ls
wh
e
rea
s
f
or
H
WT
-
2,
best
pe
rfor
m
ance is
at
4
75
sym
bo
ls by
KEK
R
E
128.
The
pro
posed
syst
e
m
is
com
par
e
d
with
e
xi
sti
ng
syst
em
s
in
T
a
ble
5
.
T
he
pro
po
se
d
s
yst
e
m
of
fers
be
tt
e
r
perform
ance th
an
the
ex
ist
in
g
syt
e
m
s.
Table
5.
C
om
par
isi
on of
pro
pose
d
syst
em
w
it
h
existi
ng syst
e
m
s
Pap
er
FAR
(%
)
FRR
(
%)
[
1
3
]
30
30
[
1
4
]
1
.96
6
0
.93
[
1
5
]
0
8
Prop
o
sed
sy
ste
m
0
0
4.
CONCL
US
I
O
N
In
the
pro
po
se
d
syst
e
m
fo
r
on
li
ne
si
gn
at
ure
ver
ific
at
io
n
with
press
ur
e
as
featur
e
vect
or,
H
WT
-
1
offer
s
bette
r
perform
ance
than
H
WT
-
2
f
or
var
i
ous
c
om
bin
at
ion
s
of
DCT,
D
HT,
Haar
a
nd
H
adam
ard
or
t
hogonal
t
ra
ns
f
or
m
.
But
K
ekr
e
tran
sf
or
m
offe
rs
bette
r
pe
rfor
m
ance
tha
n
it
s
var
i
ou
s
co
m
bin
at
ion
of
H
WT
-
1
and
H
WT
-
2.
C
om
par
ing
KNN,
S
VM
and
NN
cl
assifi
er
with
va
rio
us
dy
nam
ic
par
am
et
ers
as
feature
vector,
HMM
offer
s
be
tt
er
per
f
or
m
ance.
T
his
fin
din
gs
s
how
that
the
H
WT
with
HMM
has
be
en
a
feasible
m
et
ho
d
for feat
ur
e
v
e
ct
or extract
on of
on
li
ne
sig
natu
r
e v
ect
or
base
d biom
et
ric syst
e
m
s.
REFERE
NCE
S
[1]
R.
Plamondon,
"
A
kine
m
at
i
c
th
e
or
y
o
f
rap
id
hu
m
an
m
ovements:
Part
I:
Movem
ent
r
epr
ese
n
ta
t
io
n
and
g
ene
r
at
ion
,
"
Bi
ological
Cyber
net
ic
s
,
vol
.
72
,
no.
4
,
pp
.
295
-
3
07,
1995
.
[2]
R.
Plamondon,
"A
kine
m
at
ic
theor
y
of
rap
id
hu
m
an
m
ove
m
ent
s:
Part
II:
Movem
ent
ti
m
e
and
cont
rol,
"
Bi
o
logi
c
al
Cybe
rnet
ic
s
,
vol. 72, no. 4, pp. 30
9
-
320,
1995
.
[3]
A.
K.
Jain
,
A.
R
oss
and
S.
Prabha
kar
,
"A
n
Introd
uct
ion
to
Biom
etric
Re
cogni
t
ion,"
IEE
E
Tr
ansact
ions
on
Circui
t
s
and
Syste
ms
for
Vi
deo
Technol
og
y
, v
o
l. 14, no. 1,
p
4
-
20,
Jan
2004
.
[4]
K.
Vee
ramac
he
neni
,
L
.
A.
Os
adc
iw
and
P.
K.
Varshne
y
,
"A
n
ada
pti
ve
m
ult
imodal
biometric
m
ana
geme
nt
al
gorit
hm
,
"
IEEE
Tr
ans.
Syst
ems,
Man
and
Cyb
e
rneti
cs
.
Part
C,
v
ol.
35
,
no
.
3
,
p
.
3
44
–
356,
Aug
20
05
.
[5]
K.
Huang
and
H
.
Yan,
"S
ignatur
e
ver
if
icati
on
us
ing
fra
c
ta
l
tra
nsf
orm
at
ion,
"
Proc
.
15th
Int.
Con
f.
Pat
te
rn
Re
cog
.
(
ICPR
-
15)
,
Barcel
ona
,
Spa
in, Se
p
t
2000
.
[6]
S.
Nana
vat
i
,
M.
Thi
eme,
and
R.
Nana
vati,
“
B
iometri
cs:
Ide
nt
ity
Ve
rifi
cation
in
a
Networke
d
W
orld,
”
New
York:
W
il
e
y
,
p
123
–
13
1
,
2002
.
[7]
T.
Ohishi,
Y.
Kom
iy
a,
H.
Mo
rit
a
,
and
T
.
Ma
tsum
oto,
“
Pen
-
input
onli
ne
sig
nat
ure
v
eri
fi
ca
t
i
on
with
positi
o
n,
pre
ss
ure
,
in
cl
in
a
ti
on
tr
aj
e
ct
ori
es,
”
in
Proc
.
15th
Int.
Par
al
le
l
Distrib.
Proc
ess.
Sy
mp.
(
IPDP
S
-
15)
,
San
Franc
isco,
CA,
p.
170
,
Apr.
2001
.
[8]
H.
B.
Kek
re,
T
.
K.
Sarode
and
S
.
D.
The
p
ade,
"I
nce
pt
ion
of
HW
T
using
Two
Orthogonal
Tra
nsf
orm
s
and
It’s
use
for
Im
age
Com
pre
ss
ion,
"
(
IJCSI
S)
Inte
rnational
Journal
of
Co
mputer
S
ci
en
ce
and
Information
Sec
urit
y
,
vol
.
9
,
no.
6
,
p
80
-
87,
Jun 2011.
[9]
H.
Kekre
,
T
.
Sarode
and
P.
Nat
u,
"Colour
Im
age
Com
pre
ss
ion
using
DK
T
-
DC
T
H
y
brid
W
avele
t
Tr
ansform
in
Vari
ous
Colour
Space
s,"
Int
ernati
onal
Journal
o
f
Signal
Proce
ss
ing,
Image
Proc
essing
an
d
Patte
rn
Re
cognition
,
vol.
7
,
no
.
5
,
pp
.
105
-
124,
2014
.
[10]
H.
Kekre
,
T
.
Sa
rode
and
N.
Pra
chi
,
"Im
age
Co
m
pre
ss
ion
Us
in
g
Real
Fourier
Tra
nsform
,
Its
W
ave
le
t
Tr
ansform
and
H
y
br
id
W
a
vel
e
t
with
DCT
,
"
Inte
rnationa
l
Journal
of
Ad
va
nce
Computer
S
ci
en
ce
and
Applicat
ion
,
vol
.
4,
no.
5
,
2013
.
[11]
H.
Kekre
,
T
.
Sa
rode
and
S.
Na
t
u,
"Robus
t
W
at
e
rm
ark
ing
Techni
que
using
H
y
br
i
d
W
ave
le
t
Tra
ns
form
Gene
rat
ed
from
Kekre
Tra
nsform
and
D
iscre
t
e
Cosine
Tra
nsform
,
"
Inte
rnational
Jour
nal
of
Sci
entific
and
Re
searc
h
Publ
ic
a
ti
ons
,
vo
l.
4
,
no
.
2
,
Feb
2
014.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
On
li
ne
sig
natu
re veri
fi
cation
us
in
g hy
br
id w
avelet
tra
ns
for
m
(
Ma
noj C
ha
van
)
1831
[12]
G.
B.
Atkar
an
d
S.
Gore,
"En
hanc
e
d
Perfo
rm
anc
e
of
Co
lor
To
Gra
y
a
nd
B
ac
k
b
y
Us
ing
H
y
brid
W
avele
t
Tra
nsform
s,"
Int
ernati
onal
Journ
al
of
Eme
rging
Technol
og
y
and
A
dvanc
ed
Engine
ering
,
vo
l. 3, no. 4, April
2013
.
[13]
M
.
Singhal,
M
.
Tri
kh
a,
M
.
D
utt
a
,
“
Ti
m
e
In
depe
nden
t
Sign
at
ure
Veri
fi
cati
on
using
Norm
al
i
ze
d
W
ei
gh
ted
Coeff
icient
s
,
”
In
te
rnational
Jour
nal
of
E
le
c
tric
al
and
Computer
Engi
ne
ering
(
IJE
CE)
,
vol.
6
,
n
o
.
6,
p
2658
-
2664
,
Dec
ember
2016
.
[14]
M.
Singhal
,
M.
Tri
kha
,
M.
Dut
t
a,
“
Signat
ur
e
V
eri
fi
ca
t
ion
using
Norm
al
iz
ed
St
a
ti
c
Fe
at
ur
es
and
Neura
l
Ne
twork
Cla
ss
ifi
c
at
ion
,
”
I
nte
rnational
Jou
rnal
of
El
ec
tri
ca
l
and
Computer
Engi
ne
ering
(
IJE
CE)
,
vol.
6,
no.
6,
p
2665
-
2673
,
Dec
ember
2016
.
[15]
V.
A.
Bhara
d
i
,
V.
I.
Singh
and
Bhushan
Nema
d
e,
“
H
y
b
rid
W
ave
le
ts
base
d
Feat
ure
Ve
ct
or
Gene
ration
fro
m
Multi
dimension
al
Dat
a
set
for
On
-
li
ne
Hand
writt
en
Sign
at
ur
e
Rec
ogn
it
ion
”
Proc.
5
th
Int
.
Conf.
Con
fl
ue
nce
The
next
Gen
erati
on
In
formation
Technol
og
y
Su
mit
(
Confl
uence)
Noid
a
,
Indi
a, Nov
2014
.
[16]
C.
Gruber,
T
.
Gruber,
S.
Krinni
nger
,
and
B.
Sic
k,
“
Online
Sign
at
ure
V
eri
fi
catio
n
W
it
h
Support
Vec
tor
Ma
chi
ne
s
Based
on
LCSS
Kerne
l
Func
ti
o
ns,”
IE
EE
Tr
ansacti
ons
on
Syst
e
ms
,
MAN.
And
Cybe
rnati
cs
–
Pa
rt
B:
Cyb
erne
tic
s
,
vol.
40
,
no
.
4
,
p
10
88
-
1100,
Aug
2010.
[17]
Mage
d
M.M.
F
ahm
y
,
”
Online
handwri
tten
sig
nat
ure
v
eri
f
ic
a
tion
s
y
stem
base
d
on
DW
T
fea
t
ure
s
ext
ra
ction
and
neur
al ne
twork
c
la
ss
ifi
c
at
ion
,
”
Ain Shams
Engi
ne
ering
Journal
,
v
ol.
1
,
no
.
1
,
p
59
-
70,
Sept
2010.
[18]
SV
C
2004
H
om
e.
ht
t
p:
/
/
w
w
w
.
cse
.
us
t
.
hk/
svc2004/
.
A
c
cess
ed
02
D
ec 2
0
11
.
[19]
M.
Chava
n,
R.
R.
Singh
and
V
.
A.
Bhara
di,
“
Online
Signat
ur
e
Veri
ficati
on
u
sing
HW
T
with
Hidden
Markov
Model”
in
Pro
c
4th
Inte
rnati
onal
Confe
renc
e
on
Computing,
Comm
unic
at
ion,
Control
A
nd
Aut
omation
(
ICC
UBEA
)
,
Pu
ne,
Ind
ia Aug 20
17.
[20]
M.
Chava
n,
R
.
R.
Singh
and
V.
A.
Bhara
di
,
“
Handwrit
te
n
Signa
t
ure
Veri
fi
ca
t
ion
using
Hidden
Markov
Model
wit
h
HW
T”
in
Pro
c
4th
In
te
rnati
onal
Confe
ren
c
e
on
Computi
ng,
Comm
unic
a
ti
on,
Con
trol
And
Aut
omat
io
n
(
ICCUBEA
)
,
Pu
ne,
Ind
ia A
ug
20
17.
[21]
S.
Garc
i
a
-
Salice
tt
i
and
B.
Dori
z
zi
,
“
On
using
th
e
Vite
rb
i
pa
th
along
with
HM
M
li
kelihood
infor
m
at
ion
for
onli
n
e
signat
ure
ver
i
ficati
on
,
”
IEEE
Tr
ans.
Syst
.
,
Man,
C
ybe
rn. B
,
vol
.
37
,
no
.
5
,
p
1237
–
1
247,
Oct
2007
.
[22]
L.
Rab
ine
r
,
"A
tut
oria
l
on
h
idden
Markov
m
odel
s
and
sele
c
te
d
a
ppli
c
at
ions
in
sp
ee
ch
r
ec
ogn
it
ion
,
"
Proceedi
ngs
o
f
IEE
E
,
vo
l. 77, n
o.
2
,
p
257
-
286,
Feb
1989.
[23]
M.
M.
Shafiei
a
nd
H.
R.
Rab
ie
e
,
"
A
New
On
-
li
ne
Veri
fi
cation
Al
gorit
hm
Us
ing
Vari
able
le
ng
th
Segm
ent
at
ion
an
d
Hidden
Markov
Models,
"
in
Proc
7
th
Int
e
rnational
Confer
enc
e
on
Docu
ment
Analysis
and
Re
cogn
it
io
n
(
ICDAR’
03)
,
Singapore
,
2003.
[24]
H.
S.
Yoon,
J.
Y.
Le
e
and
H.
S.
Yang,
"A
n
on
-
li
ne
signa
ture
ver
ifica
t
ion
s
y
st
em
using
hidde
n
Markov
m
odel
in
pola
r
spa
ce,"
in
Proc.
8
th
In
t. W
orkshop F
ront.
Handwr
it
ing
R
e
cogni
t
.
(
IWF
HR
-
8)
,
Ontar
io, Ca
n
ada
,
Aug.
2002.
[25]
D.
Im
pedovo
an
d
G.
Pirlo
,
"A
ut
om
at
ic
S
ign
at
ure
Veri
fi
ca
t
ion:
Th
e
State
of
the
Art,
"
IE
EE
Tr
ansacti
on
on
Syst
ems,
MAN
and
C
ybe
r
nati
cs
Part
C:
A
ppli
cation
and
R
ev
i
ews
,
vo
l. 28,
no.
5,
Sept
2008
.
BIB
LIOGR
A
PH
IES
OF
A
UTHO
RS
Manoj
Ch
avan
rec
e
ive
d
th
e
B.
E.
d
egr
e
e
from
the
Vid
y
ava
rd
hini
’s
Coll
eg
e
of
Engi
neering
and
Te
chno
log
y
,
Univer
sit
y
of
Mum
bai
,
Maha
rashtr
a,
Indi
a,
in
2001,
the
M.E
.
degr
e
e
from
the
Shivaj
i
Uni
ver
sit
y
,
Kolhap
ur,
India
,
in
20
12,
and
cur
ren
t
l
y
he
is
doing
Ph.D.
fr
om
Pac
ifi
c
Univer
sit
y
,
Udai
pur
,
India.
From
2001
to
2005,
He
worked
as
le
cturer
in
the
dep
ar
tment
of
El
e
ct
ro
nic
s
and
Te
l
ec
om
m
unic
a
ti
on
Engi
n
eering
at
St.
Franc
i
s
Instit
ute
of
Tec
hnolog
y
,
Borival
i,
Mum
bai
-
400103,
India
.
He
has
bee
n
working
as
A
ss
ista
nt
Profess
o
r
in
Tha
kur
Coll
ege
of
Engi
neeri
ng
and
Te
chnol
o
g
y
,
Mum
bai
,
since
2
005.
He
is
li
fe
m
ember
of
ISTE,
India.
His
are
a
of
int
ere
st
is
Patt
ern
re
cogni
t
ion
and
Mac
hine L
ea
rni
ng.
Dr.
Ravish
R.
S
ingh
rec
ei
ved
B
.
E.
from
K.J.
Som
aiy
a
Coll
eg
e
o
f
Engi
ne
eri
ng,
Mum
bai
,
Univer
sit
y
of
Mum
bai
,
Jul
y
1
991
and
M.
Tec
h.
(Com
m
unic
at
i
on
Engi
neering),
I.
I.
T
.
Bom
ba
y
,
Janua
r
y
2001
.
H
e
did
his
Ph.D.
(Tech
nolog
y
)
,
from
Sarda
r
Pat
el
Co
llege
of
Engi
ne
ering,
Mum
bai
,
Un
ive
rs
ity
of
Mum
bai
,
Novem
ber
2013.
He
worked
as
As
socia
te
Profes
sor
in
Elec
troni
c
s
and
T
el
e
communic
a
ti
on
in
Th
a
kur
Coll
ege
of
Eng
i
nee
ring
and
Te
c
hnolog
y
from
1st
Janua
r
y
,
2006
to
4th
Jul
y
,
201
4.
He
a
lso
worke
d
as
As
sistant
Profes
sor
in
El
ec
tronic
s
and
Te
le
comm
un
ic
a
ti
on
Engi
n
ee
ring
from
20th
Jul
y
,
2002
to
31st
Dec
ember
2005
and
as
Senior
Le
ct
ur
er
in
El
e
ct
r
onic
s
and
Te
lec
om
m
unic
at
ion
Engi
nee
r
ing
from
8th
Augus
t,
2001
to
19th
Jul
y
2002.
Now
he
is
holdi
ng
cha
ir
of
Aca
d
emic
Advisor
at
Tha
kur
Educat
io
nal
Trust,
Mum
bai.
He
has
aut
hor
ed
and
co
-
aut
h
ore
d
seve
r
al
bo
oks
inc
ludi
ng
E
le
c
tri
c
al
Engi
n
e
eri
ng
(McGraw
Hill
Educ
a
ti
on,
India,
2016),
Engi
n
e
eri
ng
Math
emat
ic
s
(McGraw
Hill
Educ
a
ti
on
,
I
ndia
,
2016)
Num
eri
cal
and
Statis
ti
c
al
Methods
for
Co
m
pute
r
Engi
n
ee
r
ing
(McGraw
H
il
l
Educa
ti
on
,
In
dia,
2016),
Com
ple
x
Vari
ables
and
Num
eri
ca
l
Metho
ds
(McGraw
Hill
Educ
ation,
Indi
a,
2015),
and
Linea
r
Algebr
a
and
Ve
ct
or
Calculus
(McGraw
Hill
Educ
ation,
Indi
a,
2014).
He
is
li
fe
m
ember
of
IET
E
and
ISTE,
Ind
ia.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
2
,
A
pr
i
l 202
0
:
1823
-
1832
1832
Dr.
Vinay
ak
As
hok
B
haradi
r
ec
e
ive
d
th
e
B.
E
.
degr
e
e
in
Ele
ct
roni
cs
from
F
inol
ex
Aca
d
em
y
of
Mana
gement
an
d
Technol
og
y
,
Univer
sit
y
of
Mum
bai
,
in
20
02.
M.E
.
d
egr
e
e
in
Elec
tron
ic
s
and
Te
l
ec
om
m
unic
ation
in
2007
and P
hD i
n
engi
nee
ri
ng
in
2011
from
NM
IMS
Univer
sit
y
,
Mum
bai
,
Ind
ia
.
Present
l
y
he
is worki
ng
as
As
so
ci
a
te
Profess
or
with
Depa
rtment
of
Inform
at
ion
Te
chno
log
y
of
F
inol
ex
Aca
dem
y
of
Ma
nage
m
ent
and
T
ec
hnolog
y
H
e
is
li
fe
m
ember
of
ISTE
and
area
o
f
int
e
rest
are
cl
o
ud
computing,
b
ig
d
at
a
an
aly
t
ic
s,
bio
m
et
ric
s a
nd
DS
P.
Evaluation Warning : The document was created with Spire.PDF for Python.