Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
9
, No
.
5
,
Octo
ber
201
9
, pp.
4372
~4
381
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v9
i
5
.
pp4372
-
43
81
4372
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Real
-
tim
e
onlin
e f
ingerpri
nt image
classifi
cation
usin
g
ad
aptive h
ybrid t
echn
iqu
es
An
n
apurn
a M
ishra
1
,
Satchi
da
n
anda
D
e
h
uri
2
1
Depa
rtment
of Electronics a
nd
Com
m
unic
at
ion Engi
ne
eri
ng,
Sil
i
con
Insti
tut
e
of Tec
hno
log
y
,
Ind
ia
2
Depa
rtment
of
I
nform
at
ion
and
Com
m
unic
at
ion Te
chno
log
y
Faki
r
Mohan
Univ
er
sit
y
,
Ind
ia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Fe
b 2
1
, 2
01
9
Re
vised
A
pr 18
, 2
01
9
Accepte
d
Apr 30
, 201
9
Thi
s
pape
r
pre
s
ent
s
thre
e
di
ffe
r
ent
h
y
brid
class
ifi
c
at
ion
t
ec
hniq
ues
appl
ied
for
the
first
t
ime
in
rea
l
-
ti
m
e
onl
ine
fingerprint
c
la
ss
ifi
c
at
ion
.
Clas
sific
at
io
n
of
onli
ne
re
al
tim
e
finge
rprin
ts is
a
complex
ta
sk
as
it invol
v
es
ad
apt
a
ti
on
and
tuni
ng
of
c
las
sifie
r
par
amet
ers
for
bet
t
er
cl
assifi
ca
t
ion
ac
cur
acy
.
To
a
cc
om
pli
sh
the
op
ti
m
al
a
dapt
a
ti
on
of
pa
ramet
ers
of
fun
ct
ion
al
li
nk
art
if
ic
i
al
n
eur
a
l
net
work
(FL
AN
N)
for
real
-
ti
m
e
onl
ine
finge
rprin
t
cl
assifi
ca
t
ion,
p
rove
n
and
established
opt
imiz
ers,
such
as
Bi
ogeogr
ap
h
y
base
d
opti
m
ize
r
(BBO),
Gene
t
ic
a
lgori
thm
(
GA
),
and
Particle
sw
arm
opti
m
iz
er
(PS
O)
are
intelligentl
y
infused
with
it
to
form
h
y
br
id
cl
assifi
ers.
The
global
fea
tu
res
of
the
rea
l
-
t
i
m
e
finge
rprint
s
are
ext
r
acte
d
usi
ng
a
Gabor
fil
ter
-
bank
and
t
hen
passed
int
o
ada
pti
v
e
h
y
bri
d
cl
assifie
rs
for
the
desired
cl
assifi
ca
t
ion
as
per
the
Henr
y
s
ystem.
Thre
e
h
y
br
id
cl
assifi
ers,
th
e
opti
m
iz
ed
weight
ada
pt
ed
Bioge
ogr
aph
y
base
d
op
ti
m
ized
func
ti
ona
l
link
artificia
l
neur
al
ne
twork
(BBO
-
FLANN
),
Gene
t
ic
al
go
rit
hm
base
d
func
ti
on
al
lin
k
art
if
ic
i
al
n
eur
al
net
work
(GA
-
FLAN
N)
and
Parti
cle
sw
arm
opti
m
iz
e
d
func
ti
on
al
li
nk
art
if
ic
i
al
neur
al
net
work
(PS
O
-
FLAN
N),
are
expl
ore
d
for
rea
l
-
ti
m
e
onl
ine
finge
rprin
t
c
la
ss
i
fic
a
ti
on,
where
t
he
PS
O
-
FLAN
N
technique
is
show
ing
superior
p
erf
orm
anc
e
as
comp
are
d
to
GA
-
FLAN
N
and
BBO
-
FLAN
N
te
chni
ques.
Th
e
b
est
ac
cur
acy
obs
erv
ed
b
y
the
ap
pli
c
at
ion
of
PSO
-
FLAN
N,
is 98%
for
re
al
-
t
i
m
e
onli
ne
fing
er
print
cl
assifi
cati
on.
Ke
yw
or
d
s
:
BBO
Cl
assifi
cat
ion
Fing
e
r
pr
int
FLANN
Henry
S
yst
em
Copyright
©
201
9
Instit
ute of
Ad
v
ance
d
Engi
ne
eri
ng
and
Sc
ie
n
ce
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Anna
pur
na
Mi
sh
ra
,
Dep
a
rt
m
ent
of
Ele
ct
ro
nics
and C
omm
un
ic
ation
En
gin
ee
rin
g
,
Sil
ic
on
In
sti
tut
e of Tec
hnolog
y, Sil
ic
on
Hill
s,
Pati
a
,
Bh
ub
a
ne
swar
-
75
1024,
Od
is
ha, I
nd
ia
.
Em
a
il
:
ann
ap
urnam
ishra1
2@
gm
ai
l.co
m
1.
INTROD
U
CTION
Fing
e
r
pr
int
is
the
uniq
ue
patte
rn
of
each
a
nd
ever
y
in
div
i
du
al
hum
an
being
an
d
is
the
m
os
t
widely
us
e
d
bi
om
et
ric
authe
ntica
ti
on
subj
ect
.
For
r
ecognit
ion
as
well
as
identif
ic
at
ion
of
fi
ngerprints,
it
s
hould
be
cl
assifi
ed
into
diff
e
ren
t
patt
ern
s
cal
le
d
fi
nger
pr
i
nt
cl
ass
[1
-
2]
.
As
pe
r
Henry
syst
e
m
of
fin
ge
rprint
s,
it
is
div
ide
d
into
five
ty
pes
of
uniqu
e
patte
r
ns
c
al
le
d
Glob
al
cl
ass
patte
rn
s
[
3
]
,
li
ke
two
loop
patte
r
ns
(Le
f
t
loo
p
and
Ri
gh
t
lo
op
),
on
e
w
horl,
a
nd
t
wo
a
rc
h
pa
tt
ern
s
(
Ar
c
h
a
nd
Te
nted
Ar
c
h)
as
gi
ven
i
n
F
igure
1
[
4
].
T
he
two
arch
patte
rn
s
c
an
be
c
om
bin
ed
to
be
repres
ented
as
on
e
c
la
ss
.
W
it
h
this
com
bin
at
ion
the
total
nu
m
be
r
of
cl
asses co
ns
ide
red is 4
classe
s
li
ke
Left
lo
op,
Ri
gh
t L
oop,
Who
rl and
Ar
c
h
[
5
-
6
]
.
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
Real
-
ti
me o
nline fi
nger
pr
int i
mage
cl
as
sif
ic
ation u
sin
g adap
ti
ve
hybri
d
t
echn
i
qu
e
s
(
An
napurna
Mi
sh
r
a
)
4373
Figure
1.
Five
fin
gerpr
i
nt clas
ses of
he
nr
y sy
stem
Fo
r
cl
assifi
cat
i
on
m
any
research
e
rs
ha
ve
dev
el
op
e
d
ef
fi
ci
ent
cl
assifi
ers,
an
d
ha
ve
te
ste
d
on
the
avail
able
Fin
ge
rprint
Data
ba
se
i
m
ages
li
ke
NI
S
T,
DB,
and
FV
C
et
c.
S.D
e
huri
et
.a
l
[7
]
has
disc
us
se
d
reg
a
rd
i
ng
the
cl
assifi
cat
ion
a
ccur
acy
of
the
com
bin
ed
ef
f
or
ts
of
FL
A
N
N
a
nd
I
PS
O,
wh
ic
h
giv
es
ri
se
to
a
rob
us
t
cl
assifi
e
r
interm
s
of
it
s
arc
hitec
tural
c
om
plexity
as
com
par
ed
t
o
M
LP,
S
VM,
RB
F
an
d
FS
N
m
e
thods.
B.
Naik
et
.al
[
8]
has
e
xp
la
ine
d
the
c
om
petence
of
HMBO
al
gorithm
fo
r
m
at
ing
process
and
s
el
ect
ion
of
best
weig
hts for
F
L
ANN
cl
assifi
er
s.
T.
D
ash
et
.al
[
9] h
as
discu
s
sed
re
gardi
ng
Fu
zzy
-
ML
P approac
h
f
or
non
-
li
near
patte
rn
cl
assifi
cat
ion
of
fi
ngerprints.
A.
K
.
Jai
n
et
.al
[
10]
has
disc
us
s
ed
re
gardin
g
t
he
cl
assifi
cat
ion
of
fin
gerpr
i
nt
patte
rn
s
i
nto
5
cl
asses
an
d
cl
assi
fied
the
fin
gerpr
i
nts
avail
abl
e
in
N
IS
T
-
4
da
ta
base.
He
ac
hieve
d
about
90
%
acc
ur
acy
for
five
cl
ass
cl
assifi
cat
ion
a
nd
94.
8%
accu
racy
f
or
f
our
cl
ass
cl
assifi
cat
ion
for
offli
ne
database
im
ages.
D.
Sim
on
[
11
]
has
disc
usse
d
re
gardin
g
t
he
m
at
he
m
at
ics
f
or
m
ulati
on
of
BB
O
a
nd
ha
s
see
n
that
it
has
si
m
il
ar
featur
es
li
ke
GA
a
nd
PS
O
a
nd
ca
n
be
a
ppli
cabl
e
to
sim
il
ar
optim
iz
at
ion
pr
ob
le
m
s.
The
acc
uracy
of
the
cl
assi
ficat
ion
ca
n
be
i
m
pr
ov
e
d
f
urt
he
r
with
real
-
tim
e
database
i
m
ages
us
i
ng
di
ff
ere
nt
hybri
d
cl
assifi
e
rs.
In
t
his work
w
e h
av
e cla
ssifie
d
fi
ng
e
r
pr
ints
of
real t
i
m
e create
d
data
base a
nd
he
re w
e
b
as
ic
al
ly
f
ocu
s
on
Hyb
rid
cl
as
sifie
rs,
c
om
bin
ing
di
ff
e
ren
t
optim
iz
ers
al
ong
with
FL
A
N
N
cl
assifi
er.
Fi
rst
we
ha
ve
co
ll
ect
ed
the
res
ults
of
cl
assifi
cat
ion
al
gorithm
s
li
ke
G
A
-
F
LA
N
N
,
the
n
BB
O
-
F
LANN
a
nd
P
SO
-
FL
ANN
a
nd
the
n
analy
zed
the
re
su
lt
s.
A
fter
c
om
par
ison
,
it
is
cl
ear
that
PSO
-
FL
ANN
is
pr
oducin
g
best
r
esults
as
com
par
ed
t
o
the o
t
her tw
o
a
lgorit
hm
s
2.
PROP
OSE
D
METHO
D
2.1
.
In
trod
uct
ion
to FLA
NN a
s
a cl
as
sifie
r
Her
e
we
ha
ve
us
ed
t
he
trig
onom
et
ric
expansio
n
m
od
el
,
wh
e
re
eac
h
el
e
m
ent
of
the
input
featu
re
vecto
r
befo
re
exp
a
ns
i
on
ca
n
be
represe
nted
as,
I
i
i
r
1
),
(
w
her
e
ea
ch
el
e
m
ent
r(
i)
can
be
re
pr
es
ented
as
,
1
,
N
n
i
r
n
wh
e
re
N
=
nu
m
ber
of
e
xp
a
nded
points
for
each
in
put
el
em
ent
[`
12]
[
13]
.
I
n
our
case,
N=11
and
I=
re
pr
ese
nts
the
total
num
ber
of
featu
r
es
in
the
featu
r
e
vector
[14
].
T
he
ex
pa
ns
io
n
can
be
re
present
ed
as
[15
-
16
]
,
(1)
wh
e
re,
d
i
i
r
1
),
(
,
d
is t
he
set o
f feat
ur
e
s
in the dat
a set.
The
n
the
ra
ndom
we
igh
ts
ch
ose
n
from
the
ra
ng
e
[
1
1]
are
m
ulti
plied
to
the
outp
ut
a
nd
the
n
a
dded
t
o
pro
du
ce
the
ac
tual
outp
ut
of
the
net
wor
k
as
giv
e
n
in
F
ig
ure
2
[
12]
.
F
or
com
par
ison
th
e
sp
eci
fie
d
de
sired
ou
t
pu
t
is
ta
ke
n
into
co
ns
i
de
rati
on
a
nd
the
cor
res
po
nd
i
ng
diff
e
ren
ce
is
the
cal
culat
ed
error
a
nd
is
use
d
to
m
od
ify
the w
ei
gh
t i
n
eac
h pat
h q, w
hich
ca
n be e
xpresse
d
a
s
[
17
]
,
)
(
)
(
k
e
k
xf
k
W
j
j
(2)
wh
e
re,
)
(
k
xf
j
is t
he
f
un
ct
io
nally
expan
de
d
in
put a
t k
th
it
erati
on.
Fo
r
q
num
ber
of p
at
te
r
ns
,
the
ch
a
ng
e
in wei
gh
t i
s
k
W
q
k
W
q
i
i
j
j
1
1
(3)
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.
9
, N
o.
5
,
Oct
ober
20
19
:
4
3
7
2
-
4
3
8
1
4374
The wei
ght
updation i
s
done by,
k
W
k
W
k
W
j
j
j
1
(4)
Wh
e
re,
W
j
(
k)
i
s the
j
th
weig
ht
at
the k
th
it
erati
on.
By
ta
kin
g
y(
k)
as
the
desi
red
ou
t
pu
t
of
t
he
ne
twork
,
an
d
ŷ(
k)
as
t
he
act
ual
ou
t
pu
t
of
t
he
netw
ork,
t
he
error e(
k) can
be
cal
culat
e
d
a
s,
e(
k
)
=y(k)
-
ŷ(
k
)
(5)
Wh
e
re
k
w
k
xf
k
y
j
J
j
j
.
ˆ
1
(6)
and
xf
j
re
pr
ese
nt
s the e
xp
a
ns
io
n of i
nput.
Th
ough
FL
A
N
N
cl
assifi
er
is
pro
du
ci
ng
a
ve
ry
good
acc
ura
cy
of
100%
f
or
offli
ne
data
ba
se,
but
for
on
-
li
ne
real
ti
m
e
cl
assifi
cati
on
the
fo
ll
owi
ng
dr
a
w
back
s
are
pr
e
sent.
T
he
sta
bili
zi
ng
factor
(µ)
ne
e
ds
to
be
tun
e
d
by
the
use
r
by
hit
an
d
tria
l
m
et
ho
d
to
achie
ve
acc
uracy
.
D
ue
t
o
th
e
m
anu
al
tun
i
ng
of
the
pa
ram
et
er,
i
t
fail
s to
cl
assify
the
real
-
ti
m
e o
nline
fin
gerpr
i
nt classi
ficat
io
n.
Figure
2.
T
he
st
ru
ct
ur
e
of FL
A
NN cla
ssifie
r
2.2.
Pr
opose
d adaptiv
e h
ybri
d
cl
as
si
fica
tion tech
nique
s
In
this
pa
per
we
are
us
in
g
three
hy
br
i
d
cl
assifi
ers
f
or
re
al
-
tim
e
on
li
ne
cl
assifi
cat
ion
and
c
om
par
ed
their
res
ults
in
te
rm
s
of
accur
acy
and
e
xecu
t
ion
ti
m
e.
The
cl
assifi
ers
are:
(
i)
BB
O
-
FL
A
N
N
,
(ii)
G
A
-
F
L
ANN
,
(iii
)
PS
O
-
FLANN
.
2.2.1.
BBO
-
F
LAN
N
This
is
a
hy
br
i
d
te
ch
nique
co
ns
ist
ing
of
FLAN
N
cl
as
sifie
r
al
on
g
with
Bi
oge
ogr
aph
y
base
d
op
ti
m
iz
ation
to
opti
m
iz
e
the
diff
e
re
nt
pa
ram
et
ers
of
F
LANN
durin
g
on
li
ne
real
-
t
i
m
e
pr
ocessi
ng
of
fin
gerpr
i
nts.
B
BO has a
pro
pe
rty
o
f
esti
m
a
tin
g t
he pa
ram
eter
s in
g
e
ogra
phic
al
r
e
gions.
1.
BB
O
as
Op
ti
m
iz
ed
wei
gh
t
ad
apter
Bi
og
e
ography
Ba
sed
O
ptim
i
zat
ion
basical
l
y
fo
cu
ses
on
sp
eci
es
distri
buti
on
within
t
he
nei
ghbor
isl
and
s
[
18
-
19
],
that
m
eans
m
at
hem
atical
l
y
i
t
is
the
descr
ipti
on
of
sp
eci
es
m
igrati
on
f
ro
m
isl
and
to
isl
an
d
an
d
dev
el
op
m
ent
of
ne
w
sp
eci
es
.
Her
e
S
IV
s
(
Su
it
abili
ty
ind
ex
va
riables)
i
nd
ic
at
e
the
ha
bitabil
it
y
par
am
et
er,
wh
e
re
as
HS
I
s
(H
a
bitat
Sui
ta
bili
ty
In
dex)
in
dicat
es
the
well
su
it
ed
reside
nce
f
or
bio
lo
gical
spe
ci
es
,
consi
der
i
ng
t
he
featur
es
li
ke
weather
c
onditi
on,
la
nd
area
tem
per
at
ur
e
et
c
[
19
].
In
BB
O,
S
I
Vs
are
ind
e
pende
nt
va
riables
of
the
hab
it
at
an
d
HSI
is
the
de
pe
ndent
va
riable.
F
or
opti
m
iz
at
io
n
ta
sk,
it
f
ollo
ws
t
he
f
ollo
wing ste
ps:
a.
Mi
gr
atio
n
:
He
re
the
colle
ct
ion
of
ca
nd
i
da
te
so
luti
ons
th
at
is
po
pula
ti
on
is
represe
nted
in
te
rm
s
of
vecto
rs
of
i
nteger
s
or
S
IV
s
.
The
best
fit
s
ol
ution
s
a
re
ta
ke
n
as
ha
bitat
s
with
high
H
SI
(
S
2
)
an
d
le
ast
f
it
so
luti
ons
are
c
al
le
d
low
H
SI
(
S
1
)
.Th
e
par
a
m
et
ers
1
and
1
r
epr
ese
nt
the
ra
te
of
i
m
m
igrat
ion
a
nd
e
m
igrati
on
f
or
S
1
and
2
an
d
2
rep
rese
nt
the
rat
e
of
im
m
igrati
on
a
nd
em
igra
ti
on
f
or
S
2
res
pecti
vely
.
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
Real
-
ti
me o
nline fi
nger
pr
int i
mage
cl
as
sif
ic
ation u
sin
g adap
ti
ve
hybri
d
t
echn
i
qu
e
s
(
An
napurna
Mi
sh
r
a
)
4375
The
and
values
of
eac
h
s
olu
ti
on
pr
ob
a
bili
sti
cal
l
y
sh
are
inf
orm
ati
on
betwee
n
hab
it
at
s
.
W
it
h
a
pro
ba
bili
ty
m
o
d
P
, each sol
ution i
s fu
rther m
od
i
fied.
b.
Mutatio
n:
The
SI
V
m
utati
on
process
in
vo
l
ve
s
the
su
dde
n
change
in
ha
bi
ta
t’s
HS
I
due
to
rand
om
even
ts.
The
s
pecies
c
ount
pr
ob
a
bili
ti
es
are
us
e
d
t
o
dete
rm
ine
m
utati
on
rates
[
18
].
T
he
m
utati
on
rate
‘
m’
i
s
inv
e
rsely
prop
or
ti
onal
to
the
so
luti
on
pro
ba
bili
ty
.
2.
BBO
-
F
LA
N
N
te
chn
i
qu
e
f
or fi
ng
e
rprint cla
ss
ific
at
ion
This
w
ork
us
e
s
BB
O
as
an
op
ti
m
iz
er
to
up
date
the
weig
ht
param
et
ers
of
F
LA
N
N
cl
assifi
er.
T
he
ste
ps
a
re:In
it
ia
ll
y
a
fixed
nu
m
ber
of
ha
bitat
s
are
ge
ner
at
e
d,
w
her
e
each
hab
it
at
car
ries
the
res
pecti
ve
weig
hts
and
bias
of
t
he
netw
ork.
T
he
n
the
best
fit
va
lue
in
te
rm
s
of
MSE
is
cal
c
ul
at
ed.
He
re
t
he
goal
is
to
m
ini
m
ize
the
er
ror
with
resp
ect
t
o
the
desire
d
a
nd
th
e
est
i
m
at
ed
outpu
t
of
t
he
cl
as
sifie
r.
S
o
t
o
sa
ti
sfy
the
opti
m
iz
at
ion
crit
eria,
va
rio
us
operati
ons
li
ke
I
niti
al
iz
at
io
n
of
ha
bitat
,
m
igrati
on,
a
nd
m
uta
ti
on
are
pe
rfor
m
ed
an
d
on
ce
the
conditi
on
is
sat
isfie
d
it
is
te
rm
inate
d
t
o
fi
nd
t
he
best
s
olu
ti
on
in
te
rm
s
of
optim
iz
at
ion
.Then
th
e
netw
ork
with
high
fitness
(s
olu
ti
on
param
et
ers)
are
pass
ed
to
the
nex
t
gen
erati
on
an
d
rep
ea
te
d
unti
l
the
desire
d
go
al
i
s
achieve
d
as
g
i
ven in
F
ig
ure
3.
Figure
3.
Str
uct
ur
e
of the
pr
opos
e
d
BB
O
-
FL
ANN hyb
rid
m
et
hod
2.2.2.
GA
-
FL
ANN
This
is
a
hybri
d
te
ch
nique
c
onsist
ing
of
FL
ANN
cl
assifi
e
r
al
ong
with
Ge
netic
al
gorithm
to
opti
m
iz
e
the
diff
e
re
nt
pa
ram
et
ers
of
FLA
N
N
du
rin
g
on
li
ne
real
-
ti
m
e
pr
oc
essin
g
of
fin
gerpr
i
nts.
GA
has
a
property
of
est
i
m
ating
the
par
am
et
ers
with
gen
et
ic
pro
gra
m
m
ing
th
r
ough
heurist
ic
sear
ch.
1.
GA as
Op
ti
m
ized
wei
ght ada
pt
er
Gen
et
ic
Algori
thm
s
(G
As)
is
a
he
ur
ist
ic
sear
ch
process
a
nd
an
a
dap
ti
ve
al
gorithm
based
on
nat
ur
al
sel
ect
ion
of
ge
netic
s.
It
is
a
n
intel
li
ge
nt
al
gorithm
capabl
e
of
s
olv
in
g
op
ti
m
iz
ation
pro
blem
.
It
pro
ves
a
n
intel
li
gen
t
ra
ndom
searc
h
to
so
lve
optim
izati
on
pr
oble
m
[1
3,
20
].
G
As
are
design
e
d
to
so
lve
pro
bl
e
m
s
in
natu
ral
syst
e
m
s
tho
se
fo
ll
ow
the
pr
i
nciple
of
“su
rv
i
val
of
f
it
te
st”
.
It
is
a
ro
bust
sear
ch
al
gorithm
.
Ba
sicall
y
i
t
fo
ll
ows t
he foll
ow
i
ng steps:
a.
Select
ion
Op
er
ato
r
:
It
is
the
f
irst
ste
p
in
the
search
process
wh
ic
h
giv
es
prefe
ren
ce
to
be
tt
er
cand
idate
s
and
passe
s
the
ir
res
pecti
ve
ge
nes
f
or
the
ne
xt
ge
ner
at
io
n.
The
goodnes
s
crit
eria
m
os
tly
dep
e
nd
on
th
e
fitness
value
a
nd are
d
eci
ded
by the
obj
ect
i
ve
fun
ct
io
n
[21,
22
].
b.
Cross
over
O
pe
ra
tor
:
I
n
this
process
tw
o
in
div
id
ual
so
l
ution
s
a
re
sel
ect
ed
from
the
whole
popula
ti
on.
The
stri
ng
va
lues
of
t
he
in
div
id
ual
s
olu
ti
on
s
are
exc
ha
ng
e
d
by
ra
nd
om
l
y
sel
ect
ing
the
point
of
cro
ss
over
.
Thi
s
m
at
ing
proce
ss
create
s
tw
o
off
-
s
pr
i
ngs
w
hi
c
h
are
f
ur
t
her
us
ed
f
or
the
ne
xt
ge
ner
at
io
n
[13, 2
2
].
c.
Mutatio
n
O
per
ato
r
:
T
his
proc
ess
us
es
the
l
ow
pro
ba
bili
ty
i
nd
i
vidual
so
l
ution
s
wh
e
re
the
bits
are
flip
pe
d
for
m
ai
ntaining
div
er
sit
y.
S
o
m
utati
on
al
ong
with
sel
e
ct
ion
process
com
bin
ed
to
ge
ther
m
akes
th
e
al
gorithm
r
obust
and
no
ise
t
oleran
t
[17, 2
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.
9
, N
o.
5
,
Oct
ober
20
19
:
4
3
7
2
-
4
3
8
1
4376
2.
GA
-
F
LA
N
N
t
echn
i
qu
e
f
or f
i
ng
e
r
pr
int cl
assi
ficat
ion
This
w
ork
use
s
GA
as
an
op
ti
m
iz
er
to
update
the
weig
ht
par
am
et
ers
of
FL
A
NN
cl
assifi
er
.
The
ste
ps
are:
In
it
ia
ll
y
a
fix
ed
nu
m
ber
of
popula
ti
on
s
a
re
ge
ne
rated,
wh
e
re
eac
h
popula
ti
on
car
ries
the
resp
ect
ive
w
ei
gh
ts a
nd
bias
of the
netw
ork
.
The best fit
value
as M
SE is c
al
culat
ed.
Her
e
our
i
ntensi
on is erro
r
m
ini
m
iz
at
ion
with
res
pect
to
the
desir
ed
and
the
est
im
at
ed
ou
tp
ut
of
the
cl
assifi
er.
S
o
to
sat
isfy
the
op
ti
m
iz
ation
crit
eria,
var
i
ou
s
op
er
at
ion
s
li
ke
In
it
ia
li
zat
ion
of
weig
hts,
el
itis
m
con
diti
on
,
Muta
ti
on
an
d
Cros
s
over
are
perform
ed
and
on
ce
the
co
nd
it
ion
is
sat
isfied
it
is
te
r
m
inated
to
fin
d
the
best
so
luti
on
in
te
rm
s
of
op
ti
m
iz
a
ti
on
.T
he
n
the
net
work
with
hi
gh
fi
tness
(
so
l
ution
pa
ram
et
ers)
are
pass
ed
to
the
nex
t
ge
ne
rati
on
and re
peated
unti
l t
he
de
sire
d g
oal is ac
hiev
ed
as
g
i
ven in
F
ig
ure
4.
Figure
4.
Str
uc
ture of
the
pro
po
s
ed
GA
-
FL
ANN hyb
rid
m
et
hod
2.2.3.
PSO
-
FL
ANN
This
is
a
hy
br
i
d
te
c
hn
i
qu
e
co
ns
ist
ing
of
FL
ANN
cl
assi
fier
al
ong
with
Pa
rtic
le
Sw
a
rm
op
tim
iz
at
ion
to
op
ti
m
iz
e
the
diff
ere
nt
pa
ra
m
et
ers
of
FLANN
durin
g
onli
ne
real
-
tim
e
pr
oces
sin
g
of
fi
ng
e
r
pr
ints
.
PS
O
has
a
pro
per
ty
of esti
m
at
ing
the
para
m
et
ers
base
d on bi
rd
s
f
l
ock
i
ng or t
he natur
al
p
he
nom
ena.
PSO
as
Op
timi
z
ed wei
gh
t
ad
ap
t
er
Partic
le
swar
m
op
ti
m
iz
a
ti
on
fo
ll
ows
the
po
pu
la
ti
on
ba
sed
al
go
rit
hm
that
op
ti
m
iz
es
the
obj
ect
ive
functi
on
[12
-
13]
.
Her
e
the
s
olu
ti
on
is
bas
ed
on
par
ti
cl
es
[
13,
2
3
],
w
hich
im
it
a
te
s
bir
d’
s
fl
ock
i
ng
a
nd
ar
e
al
lowed
to
fly
fr
eel
y
in
the
search
s
pace.
I
n
this
process
each
an
d
ever
y
pa
rtic
le
are
al
lo
wed
to
up
date
thei
r
resp
ect
ive
po
sit
ion
an
d
ve
locit
y
fo
r
the
whole
popu
la
ti
on
.
PS
O
-
F
LANN
te
ch
ni
qu
e
for
fin
ge
rprint
cl
assifi
cat
ion
.
This
work
use
s
PS
O
as
a
n
optim
iz
er
to
update
t
he
weig
ht
pa
ram
et
ers
of
F
LA
NN
cl
assifi
er
.
The
ste
ps
in
volved
duri
ng
th
is
pr
oces
s
are:
init
ia
l
ly
a
fixed
num
ber
of
hab
it
at
s
are
ge
ner
at
e
d,
w
her
e
eac
h
hab
it
at
car
ries
the
res
pecti
ve
weig
hts
a
nd
bias
of
t
he
netw
ork.
T
he
be
st
fit
value
i
n
te
rm
s
of
M
SE
is
cal
culat
ed.
He
r
e
the
goal
is
t
o
m
ini
m
iz
e
the
error
with
re
spe
ct
to
the
desir
ed
a
nd
the
est
im
at
ed
outp
ut
of
th
e
cl
assifi
er.T
o
sat
isfy
the
op
ti
m
iz
at
ion
crit
eria,
var
io
us
ope
rati
on
s
li
ke
I
ni
ti
al
iz
ation
of
weig
hts,
posit
ion
a
nd
velocit
y
updat
e,
m
e
m
or
y
up
da
te
are
pe
rfo
r
m
ed
an
d
once
the
co
ndit
ion
i
s
sat
isfie
d
it
is
te
rm
inate
d
to
f
ind
t
he
best
so
l
utio
n
in
te
rm
s
of
op
ti
m
iz
at
ion
.Th
e
n
the
net
work
wi
th
hi
gh
fitness
(so
l
ution
pa
ra
m
et
ers)
are
pas
sed
to
the n
e
xt
genera
ti
on
a
nd r
e
peat
ed un
ti
l t
he
d
es
ired
goal
is ac
hi
eved
a
s
giv
e
n i
n
Fig
ure
5
.
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
Real
-
ti
me o
nline fi
nger
pr
int i
mage
cl
as
sif
ic
ation u
sin
g adap
ti
ve
hybri
d
t
echn
i
qu
e
s
(
An
napurna
Mi
sh
r
a
)
4377
Figure
5.
Str
uc
ture of
the
pro
po
s
ed
PSO
-
F
LANN
hybri
d
m
et
ho
d
3.
RESEA
R
CH MET
HO
D
3.1.
G
ener
at
i
on
of
re
al
-
time
dat
abase
Cl
assifi
cat
ion
of
fi
ng
e
r
pr
ints
is
basical
ly
tested
on
sta
nd
a
rd
data
base
e.
g.
N
IS
T
9.
But
in
this
wor
k
we
hav
e
f
oc
use
d
on g
ene
rati
ng
a real
ti
m
e
database
of f
in
gerpr
i
nt
sam
ples. H
ere
we
ha
ve
c
ollec
te
d
a
g
r
oup
of
50
fi
nger
pr
i
nt
sam
ple
i
m
ages
fr
om
10
stude
nts
of
Sil
ic
on
I
ns
ti
tute
of
Tec
hnology,
Bh
ub
anes
war
.
T
he
im
ages
are cap
tu
red
th
rou
gh
f
i
ng
e
rpri
nt sen
s
or
s and
store
d
in a
m
e
m
or
y. So
a t
ota
l of
5
0
fi
ng
e
rpr
ints con
sist
in
g
of
all
4
cl
asses
of
fin
gerpr
i
nts
or
igi
nates the
d
at
a
ba
se for
fur
t
her
processi
ng task
.
3.2.
Feature e
xt
r
ac
tion a
nd
cl
assi
ficat
i
on
usin
g
pr
oposed
h
ybri
d t
ec
hniqu
e
The
gro
ups
of
fin
gerpr
i
nts
ar
e
assigne
d
to
a
par
ti
cular
cl
as
s
if
the
m
ajo
rity
detai
ls
are
si
m
il
ar
in
bo
th
the
sam
ples.
In
orde
r
to
assi
gn
a
par
ti
cula
r
cl
ass
to
each
on
e
of
t
he
fin
ge
rprints,
her
e
t
he
featu
re
ext
r
act
ion
ste
p
is
done
[
24]
.
D
uri
ng
featur
e
ex
tract
i
on the
fo
ll
owin
g
st
eps
a
re ca
rr
ie
d ou
t.
a.
Feat
ur
e
v
ect
or
creati
on
b.
Norm
al
iz
a
ti
on
and Se
gm
entation
of f
i
ng
e
r
pr
i
nt im
age
c.
Or
ie
ntati
on
fiel
d
est
im
ation
d.
Core p
oin
t est
i
m
at
ion
e.
Ci
rcu
la
r regi
on
for
m
at
ion
f.
Me
an
a
nd V
a
riance calc
ulati
on
for
eac
h of t
he
sector
s
g.
Gabo
r
filt
erin
g
by
2
-
D
c
on
vo
l
ution
f
or
0,
45,
90,
135
degrees
a
n
gles
kee
ping
a
c
onsta
nt
fr
e
qu
e
nc
y
separ
at
es
the
de
ci
sion
boun
da
ry to i
nd
ic
at
e t
he
in
put cl
ass
4.
RESU
LT
S
A
ND
DI
SCUS
S
ION
Sele
ct
ed
de
gr
a
ded
fin
gerpr
i
nt
i
m
ages
fr
om
t
he
create
d
a
nd
extracte
d
re
al
tim
e
database
are
featu
re
extracte
d
us
i
ng
filt
er
ba
nk
ap
proac
h
an
d
t
he
featu
re
e
xcel
sh
eet
is
create
d
c
on
sist
in
g
of
152
featu
res
of
eac
h
fin
gerpr
i
nt
im
a
ge
[1
4
]
.
T
he
e
xtracted
feat
ures
of
50
real
-
ti
m
e
on
li
ne
fin
ge
rprint
im
ages
in
e
xcel
f
orm
at
are
us
e
d
f
or
trai
ni
ng
an
d
te
sti
ng
of
the
netw
ork.
T
he
set
s
of
featur
e
vect
or
s
are
of
al
l
the
cl
asses
sta
rting
from
cl
ass1
to
cl
ass
4.Here
cl
ass
4
(
arch)
an
d
cl
ass
5
(tente
d
arc
h)
j
oin
tl
y
represe
nt
cl
a
ss4
.
Eac
h
cl
ass
is
rep
res
ented
in
the
excel
she
et
in
te
r
m
s
of
fo
ur
valu
es
i.e.(
-
1,
-
0.33,0.3
3,1)
to
re
pr
es
e
nt
the
four
cl
asses
resp
ect
ive
ly
.
It
is
te
ste
d
f
or
bette
r
pe
rfo
rm
ance
for
diff
e
re
nt
it
erati
on
le
ve
ls
li
ke
500,
1000,
20
00,
30
00,
50
00
an
d
10,
00
0
resp
e
ct
ively
.
T
he
te
st c
onfusion m
at
rix
and the
best c
os
t g
r
aph are
co
ll
ect
ed fo
r
a
naly
sis pur
po
se
.
Fr
om
Table
1
it
is
cl
ear
that
FLANN
cl
assifi
er
is
pr
oduci
ng
100%
accu
racy
as
sho
wn
in
Fig
ur
e
6,
con
i
der
i
ng
the
ang
le
featu
re
s
and
al
so
f
or
total
featur
e
vecto
r
in
64.
16
seco
nds
by
ta
kin
g
the
sta
bili
zi
ng
factor(µ)
w
it
h a value
of 0.0
005.T
he
sta
bili
zi
ng
facto
r(
µ
)
is a par
am
et
er which
is t
un
e
d
i
n
hit a
nd
t
rail
m
et
hod
in
our
cl
assifi
c
at
ion
process
.
So
her
e
the
t
unin
g
of
sta
bili
zi
ng
facto
r
(µ
)
is
done
ra
ndom
ly
and
m
an
ually
to
i
m
pr
ove
the
a
ccur
acy
w
hich
will
create
prob
le
m
in
real
tim
e
on
-
li
ne
a
naly
sis.For
sel
ect
ing
t
he
sta
bi
li
zi
ng
par
am
et
er in
a
n
a
dap
ti
ve
pr
oc
ess, the
h
y
br
i
d
cl
assifi
er
s are
u
se
d.
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.
9
, N
o.
5
,
Oct
ober
20
19
:
4
3
7
2
-
4
3
8
1
4378
Table
1.
Cl
assi
ficat
ion
acc
ura
cy
an
d t
i
m
e d
urat
ion o
f
F
LA
NN (
for of
fli
ne
i
m
ages)
Sl.
No
.
Stab
ilizin
g
Para
m
eter
(µ)
Selecte
d
Fe
atu
re
V
ector
Maxi
m
u
m
I
te
ratio
n
Clas
sif
icatio
n
Accurac
y
(
%)
Ti
m
e
Duratio
n
(in Seco
n
d
s)
1
0
.5
Total f
eatu
re
v
ecto
r
500
22
8
1
.3
1000
26
1
7
6
.08
2
0
.05
Total f
eatu
re
v
ecto
r
500
30
6
7
.68
1000
30
1
3
3
.48
3
0
.00
5
Total f
eatu
re
v
ecto
r
500
66
6
8
.09
1000
62
1
3
3
.11
4
0
.00
0
5
0
deg
ree
500
100
2
1
.12
4
5
deg
ree
500
100
2
1
.25
9
0
deg
ree
500
100
2
1
.34
1
3
5
deg
ree
500
100
1
9
.3
Total f
eatu
re
v
ecto
r
500
100
6
4
.16
Figure
6.
Th
e
C
onfusion
m
at
rix for
total
feature
vec
tor
in
FLAN
N (for
offl
ine databa
se)
T
a
b
l
e
2
s
h
o
w
s
t
h
e
r
e
s
ul
t
s
o
f
B
B
O
-
F
L
A
N
N
c
l
a
s
s
i
f
i
c
a
t
i
o
n
.
I
t
i
s
p
r
o
d
u
c
i
n
g
a
n
m
a
xi
m
um
a
c
c
u
r
a
c
y
o
f
8
2
%
f
o
r
9
0
d
e
g
r
e
e
f
e
a
t
u
r
e
v
e
c
t
o
r
w
i
t
h
a
b
e
s
t
c
o
s
t
v
a
l
u
e
o
f
0
.
0
7
2
7
i
n
8
9
1
.
7
9
s
e
c
o
n
d
s
a
n
d
6
8
%
a
c
c
u
r
a
c
y
c
o
n
s
i
d
e
r
i
n
g
t
h
e
t
o
t
a
l
f
e
a
t
u
r
e
v
e
c
t
o
r
w
i
t
h
a
b
e
s
t
c
o
s
t
v
a
l
u
e
o
f
0
.
1
3
9
5
5
i
n
3
.
5
6
5
7
e
+
0
4
s
e
c
o
n
d
s
.
A
l
t
h
o
u
g
h
t
h
i
s
h
y
b
r
i
d
a
l
g
o
r
i
t
hm
i
s
p
r
o
d
u
c
i
n
g
g
o
o
d
r
e
s
u
l
t
s
o
f
8
2
%
f
o
r
o
n
l
i
n
e
f
i
n
g
e
r
p
r
i
n
t
s
i
n
r
e
a
l
t
i
m
e
p
l
a
t
f
o
r
m
,
b
u
t
t
h
e
e
x
e
c
u
t
i
o
n
t
i
m
e
i
s
v
e
r
y
h
i
g
h
.
Table
3
sho
w
s
the
resu
lt
s
of
GA
-
FL
A
NN
cl
assifi
cat
ion
.
It
is
pr
oduci
ng
a
n
m
axi
m
u
m
acc
ur
acy
of
10
0%
fo
r
90
de
gr
ee
f
e
a
t
u
r
e
vector
with
a
bestcos
t
value
of
0.0
087
i
n
143.9
se
cond
s
and
94% acc
uracy
co
ns
i
der
i
ng the total fe
at
ur
e
vecto
r wit
h a bestcost
val
ue
of
0.098
15
i
n
1
.20
7e+03 se
conds
.
As
pe
r
the
re
s
ult
,
the
al
gorithm
is
ro
bu
st
e
nough
for
cl
as
sific
at
ion
of
di
ff
e
ren
t
an
gula
r
featur
es
as
we
ll
as
fo
r
total
featu
re
ve
ct
or
.
Table
2
.
Cl
assi
ficat
ion
acc
ura
cy
, b
est
c
os
t a
nd ti
m
e d
ur
at
io
n o
f
BB
O
-
FL
A
NN (
for on
li
ne
i
m
ages)
Sl.
No
.
Selecte
d
Fe
atu
re
Vector
Maxi
m
u
m
Gen
eration
Bes
t Co
st
v
alu
e
Clas
sif
icatio
n
Accurac
y
(%)
Ti
m
e
Duratio
n
(in Seco
n
d
s)
1
0
deg
ree
2000
0
.11
6
3
5
74
9
5
3
.91
2
4
5
deg
ree
1
0
,00
0
0
.08
9
6
3
80
4
.93
9
7
e+0
4
3
9
0
deg
ree
2000
0
.07
2
7
82
8
9
1
.79
4
1
3
5
deg
ree
1
0
,00
0
0
.10
7
4
6
76
2
.39
2
3
e+0
4
5
Total f
eatu
re
v
ecto
r
1
0
,00
0
0
.13
9
5
5
68
3
.56
5
7
e+0
4
Table
3
.
Cl
assi
ficat
ion
acc
ura
cy
, b
est
c
os
t a
nd ti
m
e d
ur
at
io
n
of GA
-
FLAN
N
(
f
or
onli
ne
i
m
ages)
Sl.
No
.
Selecte
d
Fe
atu
re
Vector
Maxi
m
u
m
Gen
eration
Bes
t Co
st
v
alu
e
Clas
sif
icatio
n
Accurac
y
(
%)
Ti
m
e
Duratio
n
(in Seco
n
d
s)
1
0
deg
ree
1
0
,00
0
0
.10
7
2
2
94
3
3
1
.13
2
4
5
deg
ree
2
0
,00
0
0
.18
7
0
5
90
6
9
7
.82
3
9
0
deg
ree
5000
0
.00
8
7
100
1
4
3
.9
4
1
3
5
deg
ree
1
0
,00
0
0
.02
6
9
94
2
2
8
.96
5
Total f
eatu
re
v
ecto
r
1
0
,00
0
0
.09
8
1
5
94
1
.20
7
e+0
3
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
Real
-
ti
me o
nline fi
nger
pr
int i
mage
cl
as
sif
ic
ation u
sin
g adap
ti
ve
hybri
d
t
echn
i
qu
e
s
(
An
napurna
Mi
sh
r
a
)
4379
Table
4
sho
ws
the
resu
lt
s
of
PSO
-
FL
A
NN
cl
assifi
cat
ion
.
It
is
pr
oduci
ng
an
m
axi
m
u
m
accuracy
of
100%
for
45
degree
feat
ure
vecto
r
with
a
bestcost
value
of
0.0
17
in
73.47
sec
onds
and
98%
ac
cur
acy
consi
der
i
ng
th
e
total
feat
ur
e
vecto
r
with
a
bestc
os
t
valu
e
of
0.022
i
n
146.6
5
sec
onds.
As
per
the
resu
lt
s
,
the
al
gorithm
i
s
producin
g
be
st
resu
lt
s
as
co
m
par
ed
to
BB
O
-
F
LA
N
N
an
d
G
A
-
FLAN
N
,
and
t
he
sta
bi
li
zi
ng
par
am
et
er
is
s
el
ect
ed
and
optim
iz
ed
automa
ti
cal
ly
,
wh
ic
h
m
akes
it
eff
ect
ive
fo
r
real
tim
e
on
li
ne
database
.
Durin
g
the
fea
ture
e
xtracti
on
sta
ge
a
152
di
m
ension
al
feat
ur
e
vecto
r
is
extracte
d
by
col
le
ct
ively
con
sideri
ng
four
a
ngle
or
ie
ntati
on
s
(0,
45,
90
a
nd
135
de
gr
ees
)
res
pecti
vely
.
Her
e
eac
h
an
gle
ori
enta
ti
on
vect
or
pro
vid
es
38
no.
of
featu
res
a
nd
fin
al
ly
it
form
s
15
2
f
e
at
ur
es
f
or
t
he
w
ho
le
fin
gerp
rint.
T
he
a
ngul
ar
feat
ur
e
vect
or
is
te
ste
d
in
the
ne
twork
for
(0,
45,
90
a
nd
135
de
gr
ee
s)
re
sp
ect
ively
an
d
finall
y
the
w
ho
le
featu
re
ve
ct
or
is
te
ste
d
f
or
acc
uracy
.
From
the
ou
t
pu
t
gr
a
ph
a
s
show
n
i
n
Fig
ur
e
7,
it
is
cl
ea
r
that
by
pro
vid
in
g
the
t
otal
f
eat
ur
e
vecto
r
as
input
to
the
networ
k,
it
is
pr
ovidi
ng
best
cl
assifi
cat
ion
accu
rac
y
of
98%
in
P
SO
-
FL
ANN
al
gorithm
wh
e
re
as
it
is
94%
in
GA
-
F
LANN
al
gorit
hm
,
wh
ic
h
sho
ws
that
PSO
-
F
LANN
is
ref
le
ct
ing
bette
r
res
ults
as
com
par
ed
t
o G
A
-
F
LA
N
N
. T
he
b
est
c
os
t
gr
a
ph
s
a
nd the c
onf
u
sio
n
m
at
rix are s
how
n
in
th
e Fig
ur
e
7.
Table
4
.
Cl
assi
ficat
ion
acc
ura
cy
, b
est
c
os
t a
nd ti
m
e d
ur
at
io
n o
f
PS
O
-
FL
ANN
(
f
or
onli
ne
i
m
ages)
Sl.
No
.
Selecte
d
Fe
atu
re
Vector
Maxi
m
u
m
Gen
eration
Bes
t Co
st
v
alu
e
Clas
sif
icatio
n
Accurac
y
(
%)
Ti
m
e
Duratio
n
(in Seco
n
d
s)
1
0
deg
ree
500
0
.01
5
98
8
5
.64
2
4
5
deg
ree
500
0
.01
7
100
7
3
.47
3
9
0
deg
ree
500
0
.03
6
92
7
1
.9
4
1
3
5
deg
ree
500
0
.02
3
96
7
2
.82
5
Total f
eatu
re
v
ecto
r
500
0
.02
2
98
1
4
6
.65
(a)
(b)
(c)
(d)
Figure
7.
The
Confus
i
on m
atr
ix a
nd
best cost
g
ra
ph
for
tot
al
f
eat
ure
vector
,
(a
)
c
onf
us
io
n
m
at
rix
of PS
O
-
FLANN
for t
ot
al
f
eat
ure
vector com
bin
in
g
a
ll
an
gle
featu
re
v
ect
ors,
(
b) b
e
st cost
gr
a
ph
of PS
O
-
F
LA
N
N
for
total
featu
r
e v
ect
or c
om
bin
in
g
al
l an
gl
e
f
eat
ur
e
vecto
rs,
(c)
c
onf
us
io
n m
at
rix
of GA
-
FLANN
for t
ot
al
featur
e
v
e
ct
or
com
bin
ing
al
l ang
le
featu
re
ve
ct
or
s,
(
d)
bes
t cost
gr
a
ph of
GA
-
FL
A
NN f
or total
f
eat
ur
e
vecto
r
c
om
bin
ing al
l an
gle f
ea
ture vect
ors
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.
9
, N
o.
5
,
Oct
ober
20
19
:
4
3
7
2
-
4
3
8
1
4380
(e)
(f)
Figure
7.
The
Confus
i
on m
atr
ix
a
nd
best cost
g
ra
ph
for
tot
al
f
eat
ure
vector
,
(e
)
c
onf
us
io
n
m
at
rix
of BB
O
-
FLANN
for t
ot
al
f
eat
ure
vector com
bin
in
g
a
ll
an
gle
featu
re
v
ect
ors,
(
f) b
e
st cost
gr
a
ph
of BB
O
-
F
LA
N
N
for
total
f
eat
ure
ve
ct
or
c
om
b
ining
al
l ang
le
f
eat
ure
vectors
5.
CONCL
US
I
O
N
In
this
pa
per
,
t
he
cl
assifi
cat
io
n
of
real
-
ti
m
e
fin
gerpr
i
nts
into
f
our
br
oad
c
la
sses
by
ada
pt
ive
hybr
i
d
cl
assifi
ers
is
s
uccess
fu
ll
y
car
ried
ou
t.
D
ur
i
ng
feat
ur
e
extra
ct
ion
the
Ga
bo
r
filt
er
bank
pl
ay
s
a
m
ajo
r
r
ol
e
for
extracti
ng
the
vital
featur
es
of
sig
nifica
nce
fo
r
the
resp
ec
ti
ve
fing
e
rprint
s.
In
the
feature
extracti
on
st
age
a
152
dim
ensional
featur
e
vector
is
e
xtracted
for
the
whole
f
ing
e
rprint,
by
colle
ct
ively
con
side
rin
g
f
our
ang
le
s
of
ori
entat
ions
su
c
h
as
0,
45
,
90
a
nd
13
5
degrees,
w
her
e
each
a
ngle
or
ie
ntati
on
vecto
r
pr
ov
i
des
38
no.
of
featur
e
s.
T
he
angular
f
eat
ur
e
vecto
rs
f
or
0,
45,
90
a
nd
135
degrees
as
wel
l
as
the
total
featur
e
vect
or
a
r
e
us
ed
in
the
ada
ptive
hybr
i
d
cl
assifi
ers
f
or
te
sti
ng
cl
assifi
cat
ion
a
ccur
acy
.
Her
e
,
un
li
ke
FLAN
N
,
w
her
e
the
t
uning
of
sta
bil
iz
ing
fact
or
(µ)
is
done
rand
om
l
y
and
m
anu
al
ly
by
hi
t
and
tria
l,
to
i
m
pr
ov
e
it
s
cl
assifi
cat
ion
acc
ur
acy
,
wh
ic
h
is
no
t
po
s
sible
in
re
al
tim
e
on
li
ne
pr
oce
ss,
the
adap
ti
ve
hy
b
rid
cl
assifi
ers
li
ke
BB
O
-
FL
ANN,
GA
-
FL
A
NN
a
nd
PS
O
-
FLANN
a
re
desig
ned
an
d
te
ste
d
f
or
cl
assifi
c
at
i
on
acc
ur
acy
.
As
pe
r
the
resu
lt
,
the
PS
O
-
FL
A
NN
te
ch
nique
is
pro
du
ci
ng
th
e
m
axi
m
u
m
ac
cur
acy
of
10
0%
f
or
45
de
gr
e
e
featu
re
vecto
r
wit
h
a
bestcost
value
of
0.0
17
a
nd
98%
acc
ur
acy
c
on
si
der
i
ng
the
total
featu
re
ve
ct
or
with
a
bes
tc
os
t
val
ue
of
0.022,
wh
ic
h
is
the
be
st
cl
assifi
cat
ion
acc
ur
acy
as
c
om
par
ed
t
o
B
BO
-
FL
A
NN
a
nd
G
A
-
FL
ANN.
Her
e
the
sta
bili
zi
ng
par
am
et
er
(µ)
is
ada
pted
a
nd
opti
m
iz
ed
autom
at
ic
ally,
w
hich
m
akes
it
eff
ect
ive
f
or
real
ti
m
e
on
li
ne
fin
gerpr
i
nt clas
sific
at
ion
.
ACKN
OWLE
DGE
MENTS
The
first
a
uthor
w
ould
li
ke
to
t
hank
the
te
ch
nical
sup
port
of
De
pa
r
t
m
ent
of
I
nform
at
ion
an
d
Com
m
un
ic
at
io
n
Tec
hnol
og
y,
Fakir
Mo
han
Un
i
ver
sit
y,Vya
saViha
r,
Bal
as
or
e
.
REFERE
NCE
S
[1]
W
.
B
i
a
n
,
D
.
X
u
,
Q
.
L
i
,
Y
.
C
h
e
n
g
,
B
.
J
i
e
a
n
d
X
.
D
i
n
g
,
"
A
S
u
r
v
e
y
o
f
t
h
e
M
e
t
h
o
d
s
o
n
F
i
n
g
e
r
p
r
i
n
t
O
r
i
e
n
t
a
t
i
o
n
F
i
e
l
d
E
s
t
i
m
a
t
i
o
n
,
"
i
n
I
E
E
E
A
c
c
e
s
s
,
v
o
l
.
7
,
p
p
.
3
2
6
4
4
-
3
2
6
6
3
,
2
0
1
9
.
[2]
D
.
E
.
H
a
m
d
i
,
I
.
E
l
o
u
e
d
i
,
A
.
F
a
t
h
a
l
l
a
h
,
M
.
K
.
N
g
u
y
u
e
n
a
n
d
A
.
H
a
m
o
u
d
a
,
"
C
o
m
b
i
n
i
n
g
F
i
n
g
e
r
p
r
i
n
t
s
a
n
d
t
h
e
i
r
R
a
d
o
n
T
r
a
n
s
f
o
r
m
a
s
I
n
p
u
t
t
o
D
e
e
p
L
e
a
r
n
i
n
g
f
o
r
a
F
i
n
g
e
r
p
r
i
n
t
C
l
a
s
s
i
f
i
c
a
t
i
o
n
T
a
s
k
,
"
2
0
1
8
1
5
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
C
o
n
t
r
o
l
,
A
u
t
o
m
a
t
i
o
n
,
R
o
b
o
t
i
c
s
a
n
d
V
i
s
i
o
n
(
I
C
A
R
C
V
)
,
S
i
n
g
a
p
o
r
e
,
p
p
.
1
4
4
8
-
1
4
5
3
,
2
0
1
8
.
[3]
P
a
r
k
,
C
.
H
.
a
n
d
P
a
r
k
,
H
.
,
“
F
i
n
g
e
r
p
r
i
n
t
c
l
a
s
s
i
f
i
c
a
t
i
o
n
u
s
i
n
g
f
a
s
t
F
o
u
r
i
e
r
t
r
a
n
s
f
o
r
m
a
n
d
n
o
n
l
i
n
e
a
r
d
i
s
c
r
i
m
i
n
a
n
t
a
n
a
l
y
s
i
s
,
”
P
a
t
t
e
r
n
R
e
c
o
g
n
i
t
i
o
n
,
p
p
.
4
9
5
-
5
0
3
,
2
0
0
5
.
[4]
L
i
,
J
u
n
,
W
e
i
-
Y
u
n
Y
a
u
,
a
n
d
H
a
n
W
a
n
g
,
"
C
o
m
b
i
n
i
n
g
s
i
n
g
u
l
a
r
p
o
i
n
t
s
a
n
d
o
r
i
e
n
t
a
t
i
o
n
i
m
a
g
e
i
n
f
o
r
m
a
t
i
o
n
f
o
r
f
i
n
g
e
r
p
r
i
n
t
c
l
a
s
s
i
f
i
c
a
t
i
o
n
,
"
P
a
t
t
e
r
n
R
e
c
o
g
n
i
t
i
o
n
4
1
,
3
5
3
-
3
6
6
,
2
0
0
8
.
[5]
R
a
t
h
a
,
N
a
l
i
n
i
,
a
n
d
R
u
u
d
B
o
l
l
e
e
d
s
,
“
A
u
t
o
m
a
t
i
c
f
i
n
g
e
r
p
r
i
n
t
r
e
c
o
g
n
i
t
i
o
n
s
y
s
t
e
m
s
,”
S
p
r
i
n
g
e
r
S
c
i
e
n
c
e
&
B
u
s
i
n
e
s
s
M
e
d
i
a
,
2
0
0
3
.
[6]
T
a
n
,
X
.
,
B
h
a
n
u
,
B
.
a
n
d
L
i
n
,
Y
.
,
“
F
i
n
g
e
r
p
r
i
n
t
c
l
a
s
s
i
f
i
c
a
t
i
o
n
b
a
s
e
d
o
n
l
e
a
r
n
e
d
f
e
a
t
u
r
e
s
,
”
I
E
E
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
S
y
s
t
e
m
s
,
M
a
n
,
a
n
d
C
y
b
e
r
n
e
t
i
c
s
,
P
a
r
t
C
(
A
p
p
l
i
c
a
t
i
o
n
s
a
n
d
R
e
v
i
e
w
s
)
,
p
p
.
2
8
7
-
3
0
0
,
2
0
0
5
.
[7]
D
e
h
u
r
i
,
S
a
t
c
h
i
d
a
n
a
n
d
a
,
R
a
h
u
l
R
o
y
,
S
u
n
g
-
B
a
e
C
h
o
,
a
n
d
A
s
h
i
s
h
G
h
o
s
h
,
"
A
n
i
m
p
r
o
v
e
d
s
w
a
r
m
o
p
t
i
m
i
z
e
d
f
u
n
c
t
i
o
n
a
l
l
i
n
k
a
r
t
i
f
i
c
i
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
(
I
S
O
-
F
L
A
N
N
)
f
o
r
c
l
a
s
s
i
f
i
c
a
t
i
o
n
,
"
J
o
u
r
n
a
l
o
f
S
y
s
t
e
m
s
a
n
d
S
o
f
t
w
a
r
e
8
5
,
n
o
.
6
,
1
3
3
3
-
1
3
4
5
,
2
0
1
2
.
[8]
N
a
i
k
,
B
i
g
h
n
a
r
a
j
,
J
a
n
m
e
n
j
o
y
N
a
y
a
k
,
a
n
d
H
.
S
.
B
e
h
e
r
a
,
"
A
h
o
n
e
y
b
e
e
m
a
t
i
n
g
o
p
t
i
m
i
z
a
t
i
o
n
b
a
s
e
d
g
r
a
d
i
e
n
t
d
e
s
c
e
n
t
l
e
a
r
n
i
n
g
–
F
L
A
N
N
(
H
B
M
O
-
G
D
L
-
F
L
A
N
N
)
f
o
r
C
l
a
s
s
i
f
i
c
a
t
i
o
n
,
"
I
n
E
m
e
r
g
i
n
g
I
C
T
f
o
r
B
r
i
d
g
i
n
g
t
h
e
F
u
t
u
r
e
-
P
r
o
c
e
e
d
i
n
g
s
o
f
t
h
e
4
9
t
h
A
n
n
u
a
l
C
o
n
v
e
n
t
i
o
n
o
f
t
h
e
C
o
m
p
u
t
e
r
S
o
c
i
e
t
y
o
f
I
n
d
i
a
C
S
I
V
o
l
.
2
,
p
p
.
2
1
1
-
220
,
2
0
1
5
.
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
Real
-
ti
me o
nline fi
nger
pr
int i
mage
cl
as
sif
ic
ation u
sin
g adap
ti
ve
hybri
d
t
echn
i
qu
e
s
(
An
napurna
Mi
sh
r
a
)
4381
[9]
D
a
s
h
,
T
i
r
t
h
a
r
a
j
,
S
a
n
j
i
b
K
u
m
a
r
N
a
y
a
k
,
a
n
d
H
.
S
.
B
e
h
e
r
a
,
"
H
y
b
r
i
d
g
r
a
v
i
t
a
t
i
o
n
a
l
s
e
a
r
c
h
a
n
d
p
a
r
t
i
c
l
e
s
w
a
r
m
b
a
s
e
d
f
u
z
z
y
M
L
P
f
o
r
m
e
d
i
c
a
l
d
a
t
a
c
l
a
s
s
i
f
i
c
a
t
i
o
n
,
"
I
n
C
o
m
p
u
t
a
t
i
o
n
a
l
I
n
t
e
l
l
i
g
e
n
c
e
i
n
D
a
t
a
M
i
n
i
n
g
-
V
o
l
u
m
e
1
,
p
p
.
35
-
43
,
S
p
r
i
n
g
e
r
,
N
e
w
D
e
l
h
i
,
2
0
1
5
.
[10]
J
a
i
n
,
A
n
i
l
K
.
,
S
a
l
i
l
P
r
a
b
h
a
k
a
r
,
a
n
d
L
i
n
H
o
n
g
,
"
A
m
u
l
t
i
c
h
a
n
n
e
l
a
p
p
r
o
a
c
h
t
o
f
i
n
g
e
r
p
r
i
n
t
c
l
a
s
s
i
f
i
c
a
t
i
o
n
,
"
I
E
E
E
t
r
a
n
s
a
c
t
i
o
n
s
o
n
p
a
t
t
e
r
n
a
n
a
l
y
s
i
s
a
n
d
m
a
c
h
i
n
e
i
n
t
e
l
l
i
g
e
n
c
e
2
1
,
n
o
.
4
,
3
4
8
-
3
5
9
,
1
9
9
9
.
[11]
S
i
m
o
n
,
D
.
,
“
B
i
o
g
e
o
g
r
a
p
h
y
-
b
a
s
e
d
o
p
t
i
m
i
z
a
t
i
o
n
,”
I
E
E
E
t
r
a
n
s
a
c
t
i
o
n
s
o
n
e
v
o
l
u
t
i
o
n
a
r
y
c
o
m
p
u
t
a
t
i
o
n
,
p
p
.
7
0
2
-
7
1
3
,
2
0
0
8
.
[12]
D
e
h
u
r
i
,
S
.
,
R
o
y
,
R
.
,
C
h
o
,
S
.
B
.
a
n
d
G
h
o
s
h
,
A
.
,
“
A
n
i
m
p
r
o
v
e
d
s
w
a
r
m
o
p
t
i
m
i
z
e
d
f
u
n
c
t
i
o
n
a
l
l
i
n
k
a
r
t
i
f
i
c
i
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
(
I
S
O
-
F
L
A
N
N
)
f
o
r
c
l
a
s
s
i
f
i
c
a
t
i
o
n
,”
J
o
u
r
n
a
l
o
f
S
y
s
t
e
m
s
a
n
d
S
o
f
t
w
a
r
e
,
p
p
.
1
3
3
3
-
1
3
4
5
,
2
0
1
2
.
[13]
J
u
a
n
g
,
C
.
F
.
,
“
A
h
y
b
r
i
d
o
f
g
e
n
e
t
i
c
a
l
g
o
r
i
t
h
m
a
n
d
p
a
r
t
i
c
l
e
s
w
a
r
m
o
p
t
i
m
i
z
a
t
i
o
n
f
o
r
r
e
c
u
r
r
e
n
t
n
e
t
w
o
r
k
d
e
s
i
g
n
,”
I
E
E
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
S
y
s
t
e
m
s
,
M
a
n
,
a
n
d
C
y
b
e
r
n
e
t
i
c
s
,
P
a
r
t
B
(
C
y
b
e
r
n
e
t
i
c
s
)
,
p
p
.
9
9
7
-
1
0
0
6
,
2
0
0
4
.
[14]
C
h
o
,
M
.
Y
.
a
n
d
H
o
a
n
g
,
T
.
T
.
,
“
F
e
a
t
u
r
e
s
e
l
e
c
t
i
o
n
a
n
d
p
a
r
a
m
e
t
e
r
s
o
p
t
i
m
i
z
a
t
i
o
n
o
f
s
v
m
u
s
i
n
g
p
a
r
t
i
c
l
e
s
w
a
r
m
o
p
t
i
m
i
z
a
t
i
o
n
f
o
r
f
a
u
l
t
c
l
a
s
s
i
f
i
c
a
t
i
o
n
i
n
p
o
w
e
r
d
i
s
t
r
i
b
u
t
i
o
n
s
y
s
t
e
m
s
,”
C
o
m
p
u
t
a
t
i
o
n
a
l
i
n
t
e
l
l
i
g
e
n
c
e
a
n
d
n
e
u
r
o
s
c
i
e
n
c
e
,
2
0
1
7
.
[15]
P
a
t
r
a
,
J
.
C
.
,
P
a
l
,
R
.
N
.
,
C
h
a
t
t
e
r
j
i
,
B
.
N
.
a
n
d
P
a
n
d
a
,
G
.
,
“
I
d
e
n
t
i
f
i
c
a
t
i
o
n
o
f
n
o
n
l
i
n
e
a
r
d
y
n
a
m
i
c
s
y
s
t
e
m
s
u
s
i
n
g
f
u
n
c
t
i
o
n
a
l
l
i
n
k
a
r
t
i
f
i
c
i
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
,
”
I
E
E
E
t
r
a
n
s
a
c
t
i
o
n
s
o
n
s
y
s
t
e
m
s
,
m
a
n
,
a
n
d
c
y
b
e
r
n
e
t
i
c
s
,
p
a
r
t
b
(
c
y
b
e
r
n
e
t
i
c
s
)
,
p
p
.
2
5
4
-
2
6
2
,
1
9
9
9
.
[16]
K
u
m
a
r
,
P
.
S
.
a
n
d
V
a
l
a
r
m
a
t
h
y
,
S
.
,
“
M
a
r
c
h
.
D
e
v
e
l
o
p
m
e
n
t
o
f
a
n
o
v
e
l
a
l
g
o
r
i
t
h
m
f
o
r
S
V
M
B
D
T
f
i
n
g
e
r
p
r
i
n
t
c
l
a
s
s
i
f
i
e
r
b
a
s
e
d
o
n
c
l
u
s
t
e
r
i
n
g
a
p
p
r
o
a
c
h
,”
In
I
E
E
E
-
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
O
n
A
d
v
a
n
c
e
s
I
n
E
n
g
i
n
e
e
r
i
n
g
,
S
c
i
e
n
c
e
A
n
d
M
a
n
a
g
e
m
e
n
t
(
I
C
A
E
S
M
-
2
0
1
2
)
,
p
p
.
2
5
6
-
2
6
1
,
2
0
1
2
.
[17]
J
a
v
e
d
,
S
.
a
n
d
U
s
m
a
n
,
A
.
,
“
C
o
m
p
u
t
e
r
i
z
e
d
s
y
s
t
e
m
f
o
r
f
i
n
g
e
r
p
r
i
n
t
c
l
a
s
s
i
f
i
c
a
t
i
o
n
u
s
i
n
g
s
i
n
g
u
l
a
r
p
o
i
n
t
s
,”
In
2
0
1
1
I
E
E
E
1
4
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
M
u
l
t
i
t
o
p
i
c
C
o
n
f
e
r
e
n
c
e
,
p
p
.
9
6
-
1
0
1
,
D
e
c
e
m
b
e
r
2
0
1
1
.
[18]
B
h
a
t
t
a
c
h
a
r
y
a
,
A
.
a
n
d
C
h
a
t
t
o
p
a
d
h
y
a
y
,
P
.
K
.
,
“
H
y
b
r
i
d
d
i
f
f
e
r
e
n
t
i
a
l
e
v
o
l
u
t
i
o
n
w
i
t
h
b
i
o
g
e
o
g
r
a
p
h
y
-
b
a
s
e
d
o
p
t
i
m
i
z
a
t
i
o
n
f
o
r
s
o
l
u
t
i
o
n
o
f
e
c
o
n
o
m
i
c
l
o
a
d
d
i
s
p
a
t
c
h
,”
I
E
E
E
t
r
a
n
s
a
c
t
i
o
n
s
o
n
p
o
w
e
r
s
y
s
t
e
m
s
,
p
p
.
1
9
5
5
-
1
9
6
4
,
2
0
1
0
.
[19]
D
e
h
u
r
i
,
S
.
,
M
i
s
h
r
a
,
B
.
B
.
a
n
d
C
h
o
,
S
.
B
.
,
“
G
e
n
e
t
i
c
f
e
a
t
u
r
e
s
e
l
e
c
t
i
o
n
f
o
r
o
p
t
i
m
a
l
f
u
n
c
t
i
o
n
a
l
l
i
n
k
a
r
t
i
f
i
c
i
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
i
n
c
l
a
s
s
i
f
i
c
a
t
i
o
n
,
”
In
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
I
n
t
e
l
l
i
g
e
n
t
D
a
t
a
E
n
g
i
n
e
e
r
i
n
g
a
n
d
A
u
t
o
m
a
t
e
d
L
e
a
r
n
i
n
g
,
p
p
.
1
5
6
-
163
.
S
p
r
i
n
g
e
r
,
B
e
r
l
i
n
,
H
e
i
d
e
l
b
e
r
g
,
N
o
v
e
m
b
e
r
,
2
0
0
8
.
[20]
S
i
r
i
w
a
r
d
e
n
e
,
N
.
R
.
a
n
d
P
e
r
e
r
a
,
B
.
J
.
C
.
,
“
S
e
l
e
c
t
i
o
n
o
f
g
e
n
e
t
i
c
a
l
g
o
r
i
t
h
m
o
p
e
r
a
t
o
r
s
f
o
r
u
r
b
a
n
d
r
a
i
n
a
g
e
m
o
d
e
l
p
a
r
a
m
e
t
e
r
o
p
t
i
m
i
s
a
t
i
o
n
,
”
M
a
t
h
e
m
a
t
i
c
a
l
a
n
d
C
o
m
p
u
t
e
r
M
o
d
e
l
l
i
n
g
,
44
(5
-
6
)
,
p
p
.
4
1
5
-
4
2
9
,
2
0
0
6
.
[21]
J
o
h
n
s
o
n
,
J
.
M
.
a
n
d
R
a
h
m
a
t
-
S
a
m
i
i
,
Y
.
,
1
9
9
4
,
“
G
e
n
e
t
i
c
a
l
g
o
r
i
t
h
m
o
p
t
i
m
i
z
a
t
i
o
n
a
n
d
i
t
s
a
p
p
l
i
c
a
t
i
o
n
t
o
a
n
t
e
n
n
a
d
e
s
i
g
n
,
”
In
P
r
o
c
e
e
d
i
n
g
s
o
f
I
E
E
E
A
n
t
e
n
n
a
s
a
n
d
P
r
o
p
a
g
a
t
i
o
n
S
o
c
i
e
t
y
I
n
t
e
r
n
a
t
i
o
n
a
l
S
y
m
p
o
s
i
u
m
a
n
d
U
R
S
I
N
a
t
i
o
n
a
l
R
a
d
i
o
S
c
i
e
n
c
e
M
e
e
t
i
n
g
,
V
o
l
.
1
,
p
p
.
3
2
6
-
3
2
9
,
I
E
E
E
.
,
J
u
n
e
1
9
9
4
.
[22]
S
i
m
o
n
,
D
.
,
“
B
i
o
g
e
o
g
r
a
p
h
y
-
b
a
s
e
d
o
p
t
i
m
i
z
a
t
i
o
n
,
”
I
E
E
E
t
r
a
n
s
a
c
t
i
o
n
s
o
n
e
v
o
l
u
t
i
o
n
a
r
y
c
o
m
p
u
t
a
t
i
o
n
,
p
p
.
7
0
2
-
7
1
3
,
2
0
0
8
.
[23]
G
u
o
,
Y
.
H
.
a
n
d
H
u
a
n
g
,
C
.
L
.
,
“
F
u
n
c
t
i
o
n
a
l
l
i
n
k
a
r
t
i
f
i
c
i
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
f
i
l
t
e
r
f
o
r
G
a
u
s
s
i
a
n
n
o
i
s
e
,”
In
A
p
p
l
i
e
d
M
e
c
h
a
n
i
c
s
a
n
d
M
a
t
e
r
i
a
l
s
,
V
o
l
.
3
4
7
,
p
p
.
2
5
8
0
-
2
5
8
5
,
2
0
1
3
.
[24]
J
a
i
n
,
A
.
K
.
,
P
r
a
b
h
a
k
a
r
,
S
.
a
n
d
H
o
n
g
,
L
.
,
“
A
m
u
l
t
i
c
h
a
n
n
e
l
a
p
p
r
o
a
c
h
t
o
f
i
n
g
e
r
p
r
i
n
t
c
l
a
s
s
i
f
i
c
a
t
i
o
n
,
”
I
E
E
E
t
r
a
n
s
a
c
t
i
o
n
s
o
n
p
a
t
t
e
r
n
a
n
a
l
y
s
i
s
a
n
d
m
a
c
h
i
n
e
i
n
t
e
l
l
i
g
e
n
c
e
,
p
p
.
3
4
8
-
3
5
9
,
1
9
9
9
.
Evaluation Warning : The document was created with Spire.PDF for Python.