T
E
L
K
O
MN
I
K
A
T
el
eco
m
m
u
n
i
ca
t
i
o
n
,
C
o
m
p
u
t
i
n
g
,
E
l
ect
ro
n
i
cs
a
n
d
C
o
n
t
ro
l
Vo
l
.
18
, N
o
.
5
,
O
ct
o
b
er
2020
, p
p
.
2265~
2275
I
S
S
N
:
1693
-
6930,
a
c
c
r
e
di
t
e
d F
i
r
s
t
G
r
a
de
by K
e
m
e
nr
i
s
t
e
kdi
kt
i
,
D
e
c
r
e
e
N
o:
21/
E
/
K
P
T
/
2018
D
O
I
:
10.
12928/
T
E
L
K
O
M
N
I
K
A
.
v18i
5.
13725
2265
Jou
r
n
al
h
om
e
page
:
ht
t
p:
/
/
j
our
nal
.
uad
.
ac
.
i
d/
i
nde
x
.
php/
T
E
L
K
O
M
N
I
K
A
H
a
nd g
es
t
u
re r
eco
g
ni
t
i
o
n
us
i
ng
di
s
c
ret
e w
a
v
el
et
t
ra
ns
f
o
rm
a
nd
hi
d
de
n
Ma
rko
v
m
o
del
s
E
r
i
z
k
a B
an
u
w
at
i
C
an
d
r
as
ar
i
,
L
e
d
y
a
N
o
v
a
m
iz
a
n
t
i,
S
u
c
i A
u
lia
Te
l
kom
U
ni
ve
r
s
it
y
,
I
nd
one
sia
A
rt
i
cl
e I
n
f
o
AB
S
T
RACT
A
r
tic
le
h
is
to
r
y
:
R
ecei
v
ed
J
ul
25,
2019
R
e
vi
s
e
d M
a
r
30,
2020
A
ccep
t
ed
M
a
y 1,
2020
G
e
stur
e
r
e
c
o
gn
it
io
n ba
se
d on c
om
pu
te
r
-
v
is
io
n is a
n im
por
ta
n
t pa
r
t of
hum
a
n
-
c
om
pu
te
r
in
te
r
a
c
t
io
n
.
B
ut i
t la
c
k
s in se
ve
r
a
l p
o
int
s,
tha
t wa
s im
a
ge
br
i
gh
tne
ss,
r
e
c
og
ni
ti
on t
im
e
,
a
nd a
c
c
ur
a
c
y.
B
e
c
a
use
of
tha
t g
oa
l of
t
hi
s
r
e
se
a
r
c
h wa
s to
c
r
e
a
te
a
ha
nd ge
st
ur
e
r
e
c
og
ni
ti
on s
yst
e
m
tha
t ha
d g
oo
d
pe
r
f
or
m
a
nc
e
s
u
si
ng
d
isc
r
e
te
wa
ve
le
t
tr
a
n
sf
or
m
a
nd h
id
de
n
Mark
o
v
m
od
e
ls
.
The
f
ir
st pr
oc
e
s
s wa
s pr
e
-
pr
oc
e
s
si
ng,
w
hic
h d
one
b
y r
e
siz
in
g the
im
a
ge
t
o
128
x1
28
pi
xe
l
s a
n
d t
he
n
se
gm
e
nte
d t
he
s
ki
n c
o
lor
.
Th
e
se
c
o
nd pr
oc
e
ss wa
s
f
e
a
tur
e
e
x
tr
a
c
ti
on u
si
ng
the
d
isc
r
e
te
wa
ve
le
t tr
a
nsf
o
r
m
.
The
r
e
su
lt
wa
s
the
f
e
a
t
ur
e
va
lue
in
the
f
o
r
m
of
a
f
e
a
tur
e
ve
c
t
or
f
r
om
the
im
a
ge
.
T
he
la
s
t
pr
oc
e
ss wa
s ge
s
tur
e
c
la
ss
if
ic
a
ti
on u
si
ng
h
id
de
n
M
a
r
kov
m
ode
ls
t
o
c
a
l
c
u
l
a
t
e
the
hi
ghe
st pr
oba
bi
li
ty of
f
e
a
tur
e
m
a
tr
ix w
hic
h ha
d ob
ta
ine
d f
r
om
the
f
e
a
tur
e
e
xtr
a
c
ti
on pr
oc
e
ss.
T
he
r
e
s
ul
t of
t
he
s
ys
te
m
ha
d 7
2
% of
a
c
c
u
r
a
c
y
u
s
i
n
g
150 tr
a
in
in
g a
nd 1
00 te
st da
ta
im
a
ge
s tha
t c
on
si
st f
i
ve
g
e
st
ur
e
s.
The
ne
wne
ss
thi
ng f
ou
nd i
n th
is e
x
pe
r
im
e
nt we
r
e
the
e
f
f
e
c
t
of
a
c
qui
si
ti
on a
n
d
p
re
-
pr
oc
e
s
si
ng.
The
a
c
c
ur
a
c
y ha
d be
e
n e
sc
a
la
te
d b
y 14
% c
om
pa
r
e
d to
S
e
ba
s
tie
n’
s da
ta
se
t
a
t 58
%.
T
he
i
nc
r
e
m
e
n
t e
f
f
e
c
t pr
op
p
e
d b
y br
ig
ht
ne
s
s a
nd
c
on
tr
a
s
t va
l
ue
.
Ke
y
wo
r
d
s
:
D
i
s
cr
et
e
w
av
el
et
t
r
an
s
f
o
r
m
H
a
nd ge
s
t
ur
e
R
e
c
ogni
t
i
on
T
his
is
a
n
o
pe
n
ac
c
e
s
s
ar
tic
le
u
nde
r
the
CC
B
Y
-
SA
lic
e
n
se
.
C
or
r
e
s
pon
di
n
g A
u
t
h
or
:
E
r
i
zk
a B
an
u
w
at
i
C
an
d
r
as
ar
i
,
T
e
l
kom
U
ni
ve
r
s
i
t
y,
T
e
l
e
kom
uni
ka
s
i
S
t
.,
T
er
u
s
an
B
u
ah
B
at
u
S
t.
,
S
uka
pur
a
,
D
a
ye
uhkol
ot
,
Ba
n
dung,
Wes
t
J
av
a,
40257,
I
ndone
s
i
a
.
E
ma
il:
er
i
zk
ab
an
u
w
at
i
c@
g
m
ai
l
.
co
m
1.
I
NT
RO
DUC
T
I
O
N
T
he
t
e
c
hnol
ogy c
a
n be
us
e
d f
or
hum
a
n
-
phys
i
c
a
l
r
e
c
ogni
t
i
on [
1
-
10]
,
one
of
w
hi
c
h i
s
t
he
r
e
c
ogni
t
i
on
of
t
he
ha
nd
t
ha
t
c
a
n be
a
ppl
i
e
d
a
s
a
c
om
m
uni
c
a
t
i
on
t
ool
[
11
-
13]
.
T
he
pr
e
vi
ous
r
e
s
e
a
r
c
h us
e
d D
W
T
a
nd
s
uppor
t
v
ect
o
r
m
ach
i
n
e
(
S
V
M
)
c
la
s
s
if
ic
a
tio
n
[
14]
.
I
n
th
is
s
t
udy,
t
he
y obt
a
i
ne
d 94%
of
a
c
c
ur
a
c
y w
he
r
e
t
he
y di
d c
r
os
s
va
l
i
da
t
i
on f
i
ve
t
i
m
e
s
w
i
t
h 50 da
t
a
s
a
m
pl
e
s
f
r
om
s
e
ve
n a
c
t
i
ons
.
T
he
n w
he
n
231 s
a
m
pl
e
s
w
e
r
e
us
e
d f
or
t
r
a
i
ni
ng
da
t
a
a
nd t
he
r
e
m
a
i
ni
ng
119
w
e
r
e
us
e
d
f
or
t
e
s
t
da
t
a
,
t
he
y
obt
a
i
ne
d a
n
a
c
c
ur
a
c
y of
93
.
27%
.
T
es
t
s
al
s
o
car
r
i
ed
out
w
i
t
h
256×
256
pi
xe
l
i
m
a
ge
s
w
i
t
h
l
e
ve
l
5
de
c
om
pos
i
t
i
on w
hi
c
h p
r
oduc
e
d a
n
a
c
c
ur
a
c
y of
93
.
14%
.
D
W
T
c
a
n
pr
ovi
de
s
t
he
i
n
f
or
m
a
t
i
on
of
t
i
m
e
a
nd
f
r
e
que
nc
y s
i
m
ul
t
a
ne
ous
l
y a
nd w
a
ve
l
e
t
s
c
a
n be
a
r
r
a
nge
d a
nd
a
da
pt
e
d a
s
ne
e
de
d [
15]
.
H
M
M
ha
s
t
he
a
dva
nt
a
ge
a
t
be
i
ng
a
bl
e
t
o ove
r
c
om
e
t
he
pr
obl
e
m
of
e
va
l
ua
t
i
on,
i
nf
e
r
e
nc
e
,
a
nd
l
e
a
r
ni
ng [
16
]
.
H
M
M
of
t
e
n
us
e
d i
n
va
r
i
ous
a
ppl
i
c
a
t
i
ons
,
a
n e
f
f
e
c
t
i
ve
l
e
a
r
n
i
ng a
l
go
r
i
t
hm
,
a
nd
c
a
n
ha
ndl
e
va
r
i
a
t
i
ons
i
n r
e
c
or
d s
t
r
uc
t
ur
e
[
2]
.
R
e
f
e
r
r
i
ng t
o t
he
r
e
s
e
a
r
c
h [
17]
,
a
s
t
a
t
i
c
ha
nd ge
s
t
ur
e
r
e
c
ogni
t
i
on us
i
ng
t
h
e H
M
M
h
as
an
av
er
ag
e accu
r
acy
r
at
e o
f
9
3
.
3
8
%
.
T
he
pur
pos
e
of
t
hi
s
r
e
s
e
a
r
c
h
w
as
t
o be
a
bl
e
t
o de
s
i
gn a
ha
nd
ge
s
t
ur
e
r
e
c
ogni
t
i
on
s
ys
t
e
m
ba
s
e
d on
d
i
g
i
t
al
i
m
ag
es
u
s
i
n
g
D
WT
as
f
eat
u
r
e
ex
t
r
act
i
o
n
an
d
H
M
M
as
a cl
as
s
i
f
i
cat
i
o
n
al
g
o
r
i
th
m.
T
h
e
n
,
te
s
t th
e
r
e
s
u
lts
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
1693
-
6930
T
E
L
KOM
NI
KA
T
el
eco
m
m
u
n
C
om
put
E
l
C
ont
r
o
l
,
Vo
l
.
18
, N
o
.
5
,
O
c
t
obe
r
2020:
2265
-
2275
2266
a
nd a
na
l
yz
e
t
he
s
ys
t
e
m
pe
r
f
or
m
a
nc
e
s
.
T
he
pr
ob
l
e
m
s
c
ont
a
i
ne
d i
n t
hi
s
r
e
s
e
a
r
c
h i
nc
l
udi
ng how
t
o m
a
ke
t
he
s
ys
t
e
m
de
s
i
gn a
nd s
i
m
ul
a
t
i
on
of
ha
nd ge
s
t
ur
e
r
e
c
ogni
t
i
on f
o
r
da
t
a
s
e
t
us
i
ng
t
he
D
W
T
m
e
t
hod
a
n
d H
M
M
c
l
a
s
s
i
f
i
c
a
t
i
on,
how
doe
s
t
he
e
f
f
e
c
t
i
n
c
ha
ngi
ng t
h
e
va
l
ue
of
i
nput
pa
r
a
m
e
t
e
r
s
on s
ys
t
e
m
pe
r
f
or
m
a
n
c
e
s
us
i
ng
D
WT
f
eat
u
r
e
ex
t
r
act
i
o
n
m
et
h
o
d
s
an
d
H
M
M
cl
as
s
i
f
i
cat
i
o
n
,
h
o
w
i
s
t
h
e p
er
f
o
r
m
an
ces
,
an
d
h
o
w
t
h
e
accu
r
acy
an
d
t
i
m
i
ng of
c
om
put
i
ng s
ys
t
e
m
c
om
pa
r
e
d t
o
. T
h
e d
at
a
w
a
s se
l
f
co
l
l
ect
ed
us
i
ng ow
n da
t
a
s
e
t
c
ons
i
s
t
of
a
c
ol
l
e
c
t
i
on
of
ha
nd ge
s
t
ur
e
i
m
a
ge
s
us
i
ng s
m
a
r
t
phone
w
i
t
h 13
M
P
r
e
s
ol
ut
i
on.
2.
R
ES
E
A
R
C
H M
E
T
HOD
2.
1.
D
i
s
cret
e
w
a
v
el
et
t
ra
n
s
f
o
rm
D
W
T
m
e
t
h
od i
s
ut
i
l
i
z
e
d
t
o
c
o
nd
uc
t
h
a
n
d f
e
a
t
ur
e
e
xt
r
a
c
t
i
on.
T
he
m
e
t
h
od
i
s
u
s
e
d t
o c
r
e
a
t
e
ch
ar
ac
t
e
r
i
s
t
i
c
m
at
r
i
x
o
f
a
n
i
m
ag
e t
o
r
e
p
r
e
s
e
n
t
v
al
u
e
o
f
t
h
e m
at
r
i
x
o
f
t
h
e r
el
at
ed
i
m
a
g
e.
E
x
p
l
a
n
at
i
o
n
o
f
t
h
e f
e
at
u
r
e e
x
t
r
ac
t
i
o
n
pr
o
c
e
s
s
u
s
i
n
g D
W
T
w
i
t
h
e
xa
m
p
l
e
i
n
T
a
bl
e
1
. F
i
r
s
t
,
c
al
cu
l
a
t
e
t
h
e
a
ve
r
a
ge
v
a
l
ue
of
pi
xl
e
r
ow
of
a
h
a
n
d g
e
s
t
ur
e
c
o
nt
ur
e
i
m
a
ge
a
s
i
n
T
a
b
l
e
2
a
nd
t
he
r
e
s
ul
t
i
s
i
n T
a
b
l
e
3.
S
e
c
o
nd
,
c
al
cu
l
at
e
t
h
e av
er
ag
e v
a
l
u
e o
f
e
a
ch
p
i
x
e
l
c
o
l
u
m
n
s
e
t
i
n T
a
bl
e
3 by
i
np
u
t
i
ng
t
he
p
r
e
vi
ous
ca
l
cu
l
a
t
i
o
n
r
e
s
ul
t
i
l
l
us
t
r
a
t
e
d
i
n T
a
b
l
e
4
.
T
he
n
,
i
t
w
ou
l
d pr
od
uc
e
a
n ou
t
p
u
t
o
f
i
m
ag
e co
n
t
u
r
e v
a
l
u
e
s
ub
-
ba
nd
L
L
,
L
H
, H
L
, H
H
il
lu
s
t
r
a
te
d
in
T
a
b
le
5
.
F
i
n
al
p
r
o
ce
s
s
,
p
r
o
ce
s
s
o
f
r
ep
et
i
t
i
v
e
ex
tr
a
c
t
i
o
n
o
f
D
WT
ch
ar
ac
t
e
r
i
s
t
i
c
w
i
l
l
f
i
n
i
s
h i
f
e
ve
r
y
c
o
nt
ur
e
i
m
a
g
e
da
t
a
i
s
s
u
cc
es
f
u
l
l
y
e
x
t
r
a
ct
ed
[1
8
-
21
]
.
T
ab
l
e 1
.
6x6 m
a
t
r
i
x s
a
m
pl
e
M
a
tr
ix
S
a
mp
le
135
120
90
98
132
122
140
126
95
94
121
114
144
129
88
90
119
111
129
121
85
78
109
109
116
106
73
72
106
99
98
80
50
53
88
79
Ta
bl
e
2.
I
llu
s
tr
a
tio
n
o
f
cal
cu
l
at
i
o
n
pr
oc
e
s
s
of
a
ve
r
a
ge
pi
xe
l
pa
i
r
ba
s
e
d on t
he
r
ow
P
ix
e
l P
a
ir
135
+
120
2
90
+
98
2
132
+
122
2
135
−
120
2
90
−
98
2
132
−
122
2
140
+
126
2
95
+
94
2
121
+
114
2
140
−
126
2
95
−
94
2
121
−
114
2
144
+
129
2
88
+
90
2
119
+
111
2
144
−
129
2
88
−
90
2
119
−
111
2
129
+
121
2
85
+
78
2
109
+
109
2
129
−
121
2
85
−
78
2
109
−
109
2
116
+
106
2
73
+
72
2
106
+
99
2
116
−
106
2
73
−
72
2
106
−
99
2
98
+
80
2
50
+
53
2
88
+
79
2
98
−
80
2
50
−
53
2
88
−
79
2
T
ab
l
e
3.
I
llu
s
tr
a
tio
n
of
pi
xe
l
pa
i
r
ba
s
e
d on t
he
r
ow
cal
cu
l
at
i
o
n
r
e
su
l
t
P
ix
e
l P
a
ir
127
.
5
94
127
7
.
5
-
4
5
133
94
.
5
117
.
5
7
0
.
5
3
.
5
136
.
5
89
115
7
.
5
-
1
4
125
81
.
5
109
4
3
.
5
0
111
72
.
5
102
.
5
5
0
.
5
3
.
5
89
51
.
5
83
.
5
9
-
1
.
5
4
.
5
T
ab
l
e
4.
I
l
l
us
t
r
a
t
i
on
of
t
he
pr
oc
e
s
s
of
cal
cu
l
at
i
n
g
t
h
e av
er
ag
e p
i
x
el
p
ai
r
P
ix
e
l P
a
ir
127
.
5
+
133
2
94
+
94
.
5
2
127
+
117
.
5
2
7
.
5
+
7
2
−
4
+
0
.
5
2
5
+
3
.
5
2
…
.
.
…
.
.
…
.
.
…
.
.
…
.
.
…
.
.
…
.
.
…
.
.
…
.
.
…
.
.
…
.
.
…
.
.
127
.
5
−
133
2
94
−
94
.
5
2
127
−
117
.
5
2
7
.
5
−
7
2
−
4
−
0
.
5
2
5
−
3
.
5
2
…
.
.
…
.
.
…
.
.
…
.
.
…
.
.
…
.
.
…
.
.
…
.
.
…
.
.
…
.
.
…
.
.
…
.
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
el
eco
m
m
u
n
C
om
put
E
l
C
ont
r
o
l
H
and ge
s
t
ur
e
r
e
c
ogni
t
i
on us
i
ng di
s
c
r
e
t
e
w
av
e
l
e
t
t
r
ans
f
or
m
and.
.
.
(
E
r
i
z
k
a B
anuw
at
i
C
andr
as
ar
i
)
2267
T
ab
l
e
5.
I
llu
s
tr
a
tio
n
o
f
th
e
r
e
s
u
lt
f
r
o
m
cal
cu
l
a
t
i
ng a
ve
r
a
ge
pi
xe
l
pa
i
r
ba
s
e
d on t
he
c
ol
um
n
P
ix
e
l P
a
ir
LL
130
.
25
94
.
25
122
.
25
7
.
25
-
1
.
75
4
.
25
HL
130
.
75
85
.
25
112
5
.
75
1
.
25
2
100
62
93
7
-
0
.
5
4
LH
-
2
.
75
-
0
.
25
4
.
75
0
.
25
-
2
.
25
0
.
75
HH
5
.
75
3
.
75
3
1
.
75
-
2
.
25
2
11
10
.
5
9
.
5
-
2
1
-
0
.
5
2.
2.
H
i
d
d
en
Ma
rk
o
v
m
o
d
el
s
E
ach
h
i
dde
n
M
a
r
kov
m
ode
l
s
a
r
e
d
e
f
in
e
d
b
y
s
ta
te
,
p
r
o
b
a
b
ility
s
ta
te
,
p
r
o
b
a
b
ility
o
f
t
r
a
n
s
itio
n
,
pr
oba
bi
l
i
t
y of
e
m
i
s
s
i
on a
nd t
he
e
a
r
l
y
pr
oba
bi
l
i
t
y.
T
o de
s
c
r
i
be
t
he
e
nt
i
r
e
H
M
M
,
t
he
f
ol
l
ow
i
ng
f
i
ve
e
l
e
m
e
nt
s
s
houl
d be
e
l
a
bor
a
t
e
d:
a.
N
i
s
a s
t
at
e o
f
a m
o
d
el
,
d
ef
i
n
ed
as
f
o
l
l
o
w
s
:
=
{
1
,
…
,
}
(
1)
b.
M
i
s
s
ym
bol
r
e
pr
e
s
e
nt
i
ng obs
e
r
va
t
i
on pe
r
s
t
a
t
e
=
{
1
,
…
,
}
.
T
he
obs
e
r
va
t
i
on ha
s
c
ont
i
nuous
va
l
ue
a
s
th
e
M
v
a
lu
e
is
in
f
in
ity
.
c
.
P
r
o
b
a
b
ility
d
is
tr
ib
u
tio
n
o
f
tr
a
n
s
itio
n
s
ta
te
=
,
s
t
a
nds
f
or
s
t
a
t
e
pr
oba
bi
l
i
t
y a
t
t
+
1 s
ym
bol
i
z
e
d a
s
,
g
iv
e
n
w
h
e
n
s
ta
te
in
time
t v
a
lu
e
d
.
=
{
+
1
=
|
=
}
,
w
h
er
e 1
≤
,
≤
(
2)
T
h
e
d
i
sp
l
a
y
s
t
h
e
c
u
r
r
e
n
t
st
a
t
e
.
T
r
a
n
s
i
t
i
on
pr
o
ba
bi
l
i
t
y s
ho
ul
d,
m
e
e
t
t
h
e
n
or
m
a
l
l
i
m
i
t
.
≥
0
,
1
≤
,
≤
a
nd
∑
=
1
,
1
≤
≤
.
=
1
d.
T
he
O
bs
e
r
va
t
i
on
of
s
ym
bol
p
r
oba
bi
l
i
t
y
di
s
t
r
i
but
i
on i
n
e
a
c
h s
t
a
t
e
,
=
(
)
w
h
er
e
(
)
se
r
v
e
s a
s
pr
oba
bi
l
i
t
y of
s
ym
bol
o
ccu
r
r
ed
i
n
s
t
at
e
S
j
.
(
)
=
{
=
|
=
}
,
1
≤
≤
, 1
≤
≤
(
3)
s
how
s
s
ym
bol
i
n
obs
e
r
va
t
i
on
w
i
t
h
al
p
h
ab
et
an
d
s
er
v
e as
cu
r
r
en
t
v
ect
o
r
p
a
r
am
et
er
.
F
ol
l
ow
i
ng
s
to
c
h
a
s
tic
limit mu
s
t b
e
me
t
(
)
≥
0
,
1
≤
≤
, 1
≤
≤
a
nd
∑
(
)
=
1
,
1
≤
≤
=
1
.
e
.
H
M
M
is
th
e
f
ir
s
t d
is
tr
ib
u
tio
n
o
f
s
ta
te
=
{
}
,
s
t
a
nds
f
or
m
o
de
l
pr
oba
bi
l
i
t
y
i
n
s
ta
te
in
time
=
0
=
{
1
=
}
a
nd
1
≤
≤
(
4)
I
n or
de
r
t
o c
a
r
r
y out
f
u
r
t
he
r
a
na
l
ys
i
s
,
f
i
r
s
t
l
y t
w
o
ba
s
i
c
i
s
s
ue
s
of
H
M
M
s
houl
d be
s
ol
ve
d a
s
f
ol
l
ow
:
a.
E
va
l
ua
t
i
on a
nd
f
or
w
a
r
d a
nd
ba
c
kw
a
r
d i
s
s
ue
s
C
al
cu
l
at
e
t
he
va
l
ue
by i
ns
e
r
t
i
ng s
c
a
l
i
ng
f
unc
t
i
on
.
−
S
c
a
l
i
ng f
unc
t
i
on
=
1
∑
(
)
=
1
(
5)
−
F
o
rw
a
rd
−
I
n
itia
liz
a
tio
n
:
(
1
)
=
1
(
1
)
(
1
)
=
(
1
)
(
6)
−
R
ecu
r
s
i
o
n
:
1
≤
≤
−
1
,
1
≤
≤
.
+
1
(
)
=
(
+
1
)
∑
(
)
.
=
1
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
1693
-
6930
T
E
L
KOM
NI
KA
T
el
eco
m
m
u
n
C
om
put
E
l
C
ont
r
o
l
,
Vo
l
.
18
, N
o
.
5
,
O
c
t
obe
r
2020:
2265
-
2275
2268
+
1
(
)
=
[
∏
+
1
=
1
]
+
1
(
)
,
(
7)
−
T
e
r
min
a
tio
n
:
log
[
(
|
)
]
=
−
∑
log
=
1
(
8)
−
B
ack
w
ar
d
−
I
n
itia
liz
a
tio
n
:
̂
(
)
=
(
)
(
)
=
1
(
9)
−
R
ecu
r
si
o
n
:
=
−
1
,
−
2
,
…
,
1
≤
≤
.
(
)
=
∑
(
+
1
)
+
1
(
)
,
=
1
(
10)
b.
Le
a
r
ni
ng i
s
s
ue
F
ol
l
ow
i
ngs
a
r
e
t
he
s
t
e
p t
o
c
om
put
e
f
or
s
ol
vi
ng
l
e
a
r
ni
ng i
s
s
ue
:
(
)
=
(
)
(
)
∑
(
)
(
)
=
1
(
11)
(
)
=
(
)
(
+
1
)
(
+
1
)
∑
∑
(
)
(
+
1
)
(
+
1
)
=
1
=
1
(
12)
N
e
x
t s
te
p
is
r
e
-
e
s
tima
tin
g
p
ar
am
et
er
A
,
B
,
an
d
:
=
∑
(
)
−
1
=
1
∑
(
)
−
1
=
1
,
1
≤
≤
,
1
≤
≤
(
13)
(
)
=
∑
(
)
−
1
=
1
=
∑
(
)
−
1
=
1
(
14)
=
(
)
,
1
≤
≤
(
15)
T
he
pr
oc
e
s
s
a
bove
s
ho
ul
d be
c
a
r
r
i
e
d
out
unt
i
l
a
de
c
e
nt
va
l
ue
i
s
obt
a
i
ne
d
[
22
-
27]
.
2.
3.
I
m
age
p
re
-
p
ro
ces
s
i
n
g
I
n t
hi
s
r
e
s
e
a
r
c
h,
a
s
ys
t
e
m
ha
s
be
e
n de
s
i
gn
e
d t
o r
e
c
ogni
z
i
ng ha
nd ge
s
t
ur
e
s
t
hr
ough i
m
a
ge
s
.
I
n ge
ne
r
a
l
,
t
he
de
s
i
gn i
l
l
us
t
r
a
t
e
d i
n F
i
gur
e
1.
T
he
i
nput
s
w
e
r
e
t
r
a
i
ni
ng i
m
a
ge
s
f
r
om
a
R
G
B
-
l
ay
er
ed
d
at
as
et
.
T
he
i
nput
s
w
e
r
e
te
s
tin
g
ima
g
e
s
f
r
o
m a
R
G
B
-
l
ay
er
ed
d
at
as
et
u
s
i
ng
D
WT
a
s
f
eat
u
r
e
ex
t
r
act
i
o
n
m
et
h
o
d
.
T
he
f
i
na
l
pr
oc
e
s
s
w
a
s
t
o t
r
a
i
n t
he
pa
r
a
m
e
t
e
r
s
of
f
or
w
a
r
d
a
nd ba
c
kw
a
r
d t
r
a
i
ni
ng
i
m
a
ge
s
i
n e
a
c
h c
l
a
s
s
us
i
ng H
M
M
a
nd
t
he
i
nput
s
w
er
e f
eat
u
r
e v
ect
o
r
f
r
o
m
t
r
ai
n
i
n
g
i
m
ag
es
as
s
een
i
n
F
i
gur
e
1
(
a
)
. I
n
F
i
gur
e
1
(
b
)
t
he
i
nput
s
ha
d be
e
n
te
s
tin
g
f
r
o
m
a d
at
as
et
t
h
at
h
ad
a
R
G
B
l
ay
er
t
he
n ge
ne
r
a
t
e
d
a
c
ont
ou
r
i
m
a
ge
b
y
i
m
a
ge
r
e
s
i
z
i
ng a
nd s
k
i
n c
ol
or
s
e
gm
e
nt
a
t
i
on.
T
h
e
l
as
t
h
ad
b
een
p
r
o
c
e
s
s
i
ng w
i
t
h
D
WT
m
et
h
o
d
a
nd
H
M
M
m
e
t
hod,
t
h
e p
r
o
ces
s
t
h
at
h
ap
p
en
ed
w
a
s
c
a
l
c
ul
a
t
i
ng t
he
f
or
w
a
r
d
pa
r
a
m
e
t
e
r
s
a
nd de
t
e
r
m
i
ne
d t
he
c
l
a
s
s
f
r
om
t
he
hi
ghe
s
t
pr
oba
bi
l
i
t
y.
T
h
e i
m
ag
e p
r
e
-
pr
oc
e
s
di
ng ba
s
e
d on F
i
gur
e
1 w
a
s
t
o r
e
s
i
z
e
t
he
i
m
a
ge
t
o 128×
128 pi
xe
l
s
t
he
n s
e
c
ond
st
e
p
w
a
s
c
ha
nge
t
he
i
m
a
ge
f
r
om
R
G
B
t
o
Y
c
bC
r
,
B
l
ue
-
l
a
ye
r
e
d.
I
n t
hi
s
p
r
oc
e
s
s
,
t
he
i
nput
w
a
s
R
G
B
-
l
a
ye
r
e
d ha
nd
g
es
t
u
r
e i
m
ag
e
.
T
he
t
hi
r
d
s
t
e
p
w
as
s
e
gm
e
nt
t
he
s
ki
n
by
s
e
ttin
g
u
p
t
he
p
i
xe
l
va
l
ue
t
hr
e
s
hol
d,
t
he
f
i
na
l
r
e
s
ul
t
f
r
om
t
h
i
s p
r
o
c
e
ss w
a
s a
s
eg
m
en
t
ed
i
m
ag
e
.
T
he
f
our
t
h
s
t
e
p c
a
l
l
e
d de
noi
s
i
ng,
w
h
er
e t
h
i
s
p
r
o
c
e
ss
ha
d be
e
n r
e
m
ovi
ng
t
he
noi
s
e
i
n
t
he
s
ig
n
a
l w
h
ile
ma
in
ta
in
s
ig
n
a
l
c
h
a
r
a
c
te
r
is
tic
s
.
T
h
e
f
if
th
s
te
p
w
as
t
o
f
il
le
d
u
p
t
he
noi
s
e
t
h
a
t
c
a
nnot
be
r
e
m
ove
d f
r
om
t
he
pr
e
vi
ous
pr
oc
e
s
s
.
T
he
s
i
xt
h s
t
e
p
w
as
a
di
l
a
t
i
on pr
oc
e
s
s
to
t
hi
c
ke
n t
he
e
dge
of
th
e
s
e
g
me
n
te
d
ima
g
e
f
r
o
m
th
e
l
a
st
p
r
o
ces
s
s
o
t
h
at
t
h
e
r
eq
u
i
r
ed
p
i
x
el
s
can
b
e
d
et
ect
ed
.
T
he
s
e
ve
nt
h s
t
e
p
w
as
t
he
e
r
os
i
on pr
oc
e
s
s
w
hi
c
h w
oul
d e
r
ode
d
t
he
e
d
ge
of
t
he
s
e
gm
e
nt
e
d i
m
a
ge
f
r
om
t
he
l
a
st
p
r
o
c
e
ss so
t
h
a
t
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
el
eco
m
m
u
n
C
om
put
E
l
C
ont
r
o
l
H
and ge
s
t
ur
e
r
e
c
ogni
t
i
on us
i
ng di
s
c
r
e
t
e
w
av
e
l
e
t
t
r
ans
f
or
m
and.
.
.
(
E
r
i
z
k
a B
anuw
at
i
C
andr
as
ar
i
)
2269
unne
c
e
s
s
a
r
y pi
xe
l
s
c
a
n be
r
e
m
ove
d
.
T
h
e o
ut
put
w
a
s
Y
Cb
Cr
-
l
ay
er
ed
ha
nd ge
s
t
ur
e
c
ont
our
s
.
T
h
e
m
ai
n
p
r
o
ces
s
i
n pr
e
-
pr
oc
e
s
s
i
ng w
a
s
a
pr
oc
e
s
s
of
s
e
pa
r
a
t
i
ng t
he
b
a
c
kgr
ound a
nd obj
e
c
t
s
,
w
hi
c
h i
n
t
hi
s
r
e
s
e
a
r
c
h
w
a
s
t
he
r
i
ght
ha
nd
a
s
s
e
e
n i
n F
i
gur
e
2
.
(
a)
(
b)
(
c)
(
d)
(
e)
F
i
gur
e
2.
H
an
d
ge
s
t
ur
e
;
(
a
)
le
t
te
r
A
,
(
b
)
le
tte
r
B
,
(
c
)
le
tte
r
C
,
(
d
)
p
o
in
t g
e
s
tu
r
e
,
a
nd
(
e
)
num
be
r
5
2.
4.
F
ea
t
u
re
ex
t
ra
ct
i
o
n
a
n
d
c
la
s
s
if
ic
a
t
io
n
I
n
t
h
i
s
r
e
s
e
ar
ch
,
w
e
u
s
ed
t
h
e
D
WT
t
o
f
i
n
d
t
h
e
h
an
d
f
e
at
u
r
e
s
a
n
d
t
o
cr
ea
t
e
a f
e
at
u
r
e
m
at
r
i
x
f
r
o
m
an
i
m
ag
e t
o
d
e
n
o
t
e
t
h
e m
at
r
i
x
v
al
u
e
o
f
t
h
e i
m
a
g
e
i
t
s
el
f
.
T
h
e r
e
s
u
l
t
w
a
s
a c
o
n
t
o
u
r
i
m
ag
e v
al
u
e w
i
t
h
i
n
L
L
,
L
H
,
H
L
,
H
H
s
u
bba
nd
a
s
a
n
ex
am
p
l
e s
ee
n
i
n
T
ab
l
e
5
.
T
he
c
la
s
s
if
i
c
a
ti
o
n
p
r
o
c
e
s
s
w
i
th
H
M
M
a
s
il
l
u
s
tr
a
t
e
d
i
n F
i
gu
r
e
3
,
i
np
ut
w
a
s
a
c
om
bi
ne
d v
e
c
t
o
r
f
r
o
m
t
r
a
i
ni
n
g i
m
a
ge
’
s
c
ha
r
a
c
t
e
r
i
s
t
i
c
v
e
c
t
or
r
e
s
ul
t
i
ng f
r
om
t
h
e
f
e
a
t
ur
e
e
x
t
r
a
c
t
i
o
n
pr
o
c
e
s
s
us
i
n
g
D
W
T
.
I
n
a
d
di
t
i
o
n,
H
M
M
r
e
qui
r
e
d
A
,
B
,
π
,
s
t
a
t
e
,
a
n
d
c
l
us
t
e
r
va
l
u
e
s
.
I
t
w
a
s
n
e
c
e
s
s
ar
y
t
o
d
et
er
m
i
n
e t
h
e r
eq
u
i
r
e
d
s
t
at
e v
al
u
e
an
d
ca
l
c
u
l
at
ed
t
h
e c
l
u
s
t
e
r
v
al
u
e
a
s
t
h
e
o
bs
e
r
va
t
i
on
va
l
u
e
b
y
s
e
e
ki
ng t
h
e
k
-
m
e
an
s
v
al
u
e.
T
h
e
n
e
xt
pr
o
c
e
s
s
w
a
s
c
a
l
c
u
l
a
t
i
ng t
h
e
f
or
w
a
r
d va
r
i
a
bl
e
,
na
m
e
l
y
t
he
pr
oc
e
s
s
of
i
ni
t
i
a
l
i
z
a
t
i
o
n [
1
0,
2
8]
,
r
e
c
ur
s
i
o
n a
nd
t
e
r
m
i
na
t
i
on [
2]
.
B
e
f
o
r
e
t
h
e
p
r
o
ces
s
,
t
h
e
r
e
w
as
a
n
a
d
d
ed
p
r
o
ces
s
o
f
ca
l
c
u
l
at
e
t
h
e
s
c
al
i
n
g
f
u
n
ct
i
o
n
.
N
e
x
t
w
a
s
a b
ac
k
w
ar
d
a
l
gor
i
t
hm
c
a
l
c
ul
a
t
i
o
n.
T
h
e
pr
o
c
e
s
s
c
on
s
i
s
t
e
d of
t
w
o
s
t
a
g
e
s
,
t
he
i
n
i
t
i
a
l
i
z
a
t
i
on a
n
d
r
e
c
ur
s
i
on s
t
a
g
e
s
.
C
a
l
c
u
l
a
t
e
d
t
h
e v
ar
i
a
b
l
e
ξ
(
,
)
a
nd
(
)
ba
s
e
d
on t
h
e
v
a
r
i
a
b
l
e
s
d
e
f
i
ne
d i
n
t
h
e
pr
e
v
i
o
u
s
f
or
w
a
r
d
a
n
d b
a
c
k
w
a
r
d
pr
o
c
e
d
ur
e
s
.
A
f
t
e
r
t
h
e
f
o
u
r
v
a
r
i
ab
l
es
w
as
o
b
r
t
a
i
n
ed
,
r
ees
t
i
m
a
t
ed
t
h
e
p
ar
am
et
er
s
A
, B
,
a
nd
π
.
T
he
f
i
n
a
l
s
t
e
p w
a
s
t
o t
a
ke
t
h
e
hi
gh
e
s
t
pr
ob
a
b
i
l
i
t
y
v
a
l
u
e
of
t
he
t
e
s
t
i
ng
i
m
ag
e
t
o
b
e u
s
e
d
as
t
h
e
f
i
n
al
v
a
l
u
e
o
f
t
h
e h
a
n
d
g
e
s
t
u
r
e cl
a
s
s
i
f
i
c
at
i
o
n
.
(
a)
(
b)
F
i
gur
e
1
. S
y
s
t
e
m
de
s
i
gn
f
l
o
w
ch
ar
t
;
(a
) t
ra
i
n
i
n
g
,
(
b
)
te
s
tin
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
1693
-
6930
T
E
L
KOM
NI
KA
T
el
eco
m
m
u
n
C
om
put
E
l
C
ont
r
o
l
,
Vo
l
.
18
, N
o
.
5
,
O
c
t
obe
r
2020:
2265
-
2275
2270
(
a)
(
b)
F
i
gur
e
3.
F
l
ow
c
ha
r
t
of
c
la
s
s
if
ic
a
tio
n
p
r
o
c
e
ss;
(
a
)
tr
a
in
in
g
,
(
b
)
te
s
tin
g
3.
RE
S
UL
T
S
AND ANAL
YS
I
S
S
y
s
te
m te
s
tin
g
w
a
s
p
e
r
f
o
r
me
d
o
u
t
f
r
o
m
s
el
f
co
l
l
e
ct
ed
d
a
ta
s
e
t w
ith
a
n
ima
g
e
th
a
t th
r
o
u
g
h
a
r
e
s
iz
in
g
pr
oc
e
s
s
m
e
a
s
ur
e
d a
t
128×
128 pi
xe
l
s
.
T
he
pur
pos
e
of
e
xa
m
i
ni
ng t
hi
s
s
ys
t
e
m
w
a
s
t
o c
om
pa
r
e
t
he
a
c
c
ur
a
c
y,
s
y
s
t
em
p
er
f
o
r
m
an
ces
,
an
d
t
h
e b
es
t
-
pe
r
f
or
m
e
d pa
r
a
m
e
t
e
r
s
f
or
ha
nd
ge
s
t
ur
e
r
e
c
ogni
t
i
on s
ys
t
e
m
s
.
I
n
t
hi
s
r
e
s
e
a
r
ch
,
t
he
t
ot
a
l
i
m
a
ge
da
t
a
us
e
d w
a
s
250 i
m
a
ge
s
f
r
om
d
a
t
a
s
e
t
.
T
he
ha
nd ge
s
t
ur
e
i
m
a
ge
c
ons
i
s
t
s
of
5 w
o
r
d c
l
a
s
s
e
s
w
hi
c
h
e
a
c
h c
ons
i
s
t
e
d of
50 i
m
a
ge
s
.
3.
1.
T
es
t
i
n
g
t
h
e
s
y
s
t
em
p
a
ra
m
e
t
ers
T
h
e p
ar
am
et
er
t
es
t
i
n
g
g
o
al
w
as
to
o
b
ta
in
th
e
r
e
s
u
l
ts
o
f
p
a
r
a
me
te
r
s
w
ith
th
e
b
es
t
p
er
f
o
r
m
an
ce,
m
o
r
e
sp
e
si
f
i
c
,
t
h
e accu
r
acy
an
d
t
i
m
i
n
g
o
f
t
h
e s
y
s
t
em
.
−
L
ay
er
-
t
ype
p
ar
am
et
er
s
i
m
p
act
D
one
by
us
i
ng
one
t
ype
o
f
l
ay
er
f
o
r
t
es
t
i
n
g
an
d
t
h
en
D
WT
f
eat
u
r
e
ex
t
r
act
i
o
n
w
as
p
er
f
o
r
m
ed
a
n
d
cl
as
s
i
f
i
ed
u
s
i
n
g
H
M
M
as
s
h
o
w
n
i
n
T
ab
l
e 6
.
I
t
ca
n
b
e s
een
t
h
at
t
h
e
b
es
t
p
ar
am
et
er
w
as
i
n
t
h
e Y
C
b
C
r
l
ay
er
.
I
n
T
ab
l
e 6
i
t
ap
p
ear
s
t
h
at
t
h
e b
l
u
e l
ay
er
h
av
e t
h
e h
i
g
h
es
t
accu
r
acy
.
T
h
i
s
w
as
due
t
o t
he
hi
gh
f
r
e
que
nc
y o
f
pi
xe
l
s
,
fro
m
0
t
o 45
f
o
r
hi
gh
-
in
te
n
s
ity
va
l
ue
s
a
t
pi
xe
l
s
0
t
o
231 c
om
pa
r
e
d t
o
ot
he
r
t
ype
s
of
l
a
ye
r
s
a
s
i
n F
i
gu
r
e
4
.
−
S
ub
-
ba
nd
-
t
y
p
e p
ar
am
et
er
s
i
m
p
act
D
one
by
us
i
ng l
a
ye
r
s
t
ha
t
ha
d
t
he
be
s
t
pe
r
f
o
r
m
a
n
c
e
i
n
t
he
p
r
e
vi
ous
t
e
s
t
,
t
he
bl
ue
l
ay
er
,
an
d
D
WT
pa
r
a
m
e
t
e
r
s
,
t
ha
t
w
a
s
t
he
f
our
t
ype
s
of
s
ub
-
ba
nd c
o
ns
i
s
t
i
ng i
n
l
ow
-
l
o
w
(
LL)
,
l
ow
-
hi
gh
(L
H
),
hi
gh
-
l
o
w
(H
L
),
hi
gh
-
hi
gh
(
H
H
)
.
T
he
pe
r
f
or
m
a
nc
e
r
e
s
ul
t
s
i
n s
ub
-
ch
ap
t
er
w
er
e d
es
cr
i
b
ed
i
n
T
ab
l
e 7
an
d
can
b
e s
een
t
h
at
t
h
e b
es
t
p
ar
am
et
er
w
as
i
n
t
h
e L
L
s
u
b
-
ba
nd t
ype
.
T
he
L
L
s
ub
-
ba
nd ha
d a
s
m
oot
h
e
st
t
h
a
n
ot
he
r
su
b
-
ba
nd t
ype
s
a
s
s
how
n i
n F
i
gur
e
5
.
−
D
eco
m
p
o
s
i
t
i
o
n
l
ev
el
p
ar
am
et
er
s
i
m
p
act
T
he
p
r
e
vi
ous
t
e
s
t
w
a
s
c
onduc
t
e
d by
a
na
l
yz
i
ng
t
h
e
va
l
ue
o
f
D
W
T
de
c
om
pos
i
t
i
on
l
ev
el
p
ar
am
et
er
s
o
f
l
ev
el
1,
2
,
3
,
a
nd 4
i
n
t
he
da
t
a
s
e
t
.
T
h
e
t
es
t
s
w
er
e car
r
i
ed
o
u
t
w
i
t
h
t
h
e b
es
t
p
ar
am
e
t
e
r
s
i
n
t
he
t
w
o
pr
e
vi
ous
t
es
t
i
n
g
p
ar
am
et
er
s
,
t
h
e
b
l
u
e l
ay
er
a
nd
LL
su
b
-
b
an
d
.
T
h
e p
er
f
o
r
m
an
ce r
es
u
l
t
s
we
r
e d
es
cr
i
b
ed
i
n
T
ab
l
e 8
.
G
r
ap
h
s
o
f
ch
ar
act
er
i
s
t
i
cs
t
h
at
w
er
e
i
n
f
l
ue
nc
e
d by l
e
ve
l
de
c
om
pos
i
t
i
on
w
er
e
s
how
n i
n F
i
gu
r
e
6.
T
he
ch
an
g
es
o
f
l
ev
el
de
c
om
pos
i
t
i
on
r
es
u
l
t
i
n
g
i
n
t
h
e acq
u
i
r
ed
ch
a
r
act
er
i
s
t
i
cs
ha
d no
m
an
y
ch
ar
act
er
i
s
t
i
c.
T
h
e
s
m
al
l
er
t
he
de
c
om
pos
i
t
i
on l
e
ve
l
,
t
he
f
a
s
t
e
r
t
h
e
c
o
mp
u
ta
ti
o
n
a
l time
w
oul
d
be
.
H
ow
e
ve
r
,
t
hi
s
w
as
not
t
he
c
a
s
e
w
i
t
h
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
el
eco
m
m
u
n
C
om
put
E
l
C
ont
r
o
l
H
and ge
s
t
ur
e
r
e
c
ogni
t
i
on us
i
ng di
s
c
r
e
t
e
w
av
e
l
e
t
t
r
ans
f
or
m
and.
.
.
(
E
r
i
z
k
a B
anuw
at
i
C
andr
as
ar
i
)
2271
accu
r
acy
,
t
h
er
e
w
er
e
s
ev
er
al
v
al
u
es
f
o
r
l
ev
el
d
eco
m
p
o
s
i
t
i
o
n
p
ar
am
et
er
s
t
h
at
h
ad
cl
ear
ch
ar
act
er
i
s
t
i
c v
al
u
es
t
o
ab
l
e t
o
m
ad
e d
i
f
f
er
en
t
b
et
w
een
cl
as
s
es
.
T
ab
l
e 6
.
L
ay
er
-
t
y
p
e p
ar
am
et
er
s
p
er
f
o
r
m
an
ce
s
L
ay
er
T
o
ta
l te
s
tin
g
d
a
ta
T
o
t
al
co
r
r
ect
d
at
a
A
ccu
r
acy
(
%
)
C
o
mp
u
ta
ti
on
time
(
s
)
R
ed
100
50
50
46
G
r
een
100
20
20
41
B
lu
e
100
68
68
43
G
r
ay
s
cal
e
100
59
59
49
B
i
n
a
ry
100
32
32
56
Y
Cb
Cr
(C
r)
100
53
53
44
HS
V (
V)
100
20
20
40
T
ab
l
e
7
.
S
ub
-
ba
nd
-
t
y
p
e p
ar
am
et
er
s
p
er
f
o
r
m
an
ces
Su
b
-
ba
nd
T
o
ta
l te
s
tin
g
d
a
ta
T
o
t
al
co
r
r
ect
d
a
ta
A
ccu
r
acy
(
%
)
C
o
mp
.
time
(
s
)
LL
100
68
68
43
LH
100
20
20
57
HL
100
20
20
54
HH
100
20
20
55
F
i
gur
e
4.
H
i
s
t
ogr
a
m
o
f
bl
ue
l
a
ye
r
i
m
a
ge
s
F
i
gur
e
5.
I
l
l
us
t
r
a
t
i
on of
i
m
a
ge
s
i
n s
ub
-
ba
nd
T
ab
l
e 8
.
P
er
f
o
r
m
an
ces
o
f
d
eco
m
p
o
s
i
t
i
o
n
l
ev
el
p
ar
a
m
et
er
s
L
ev
el
T
o
ta
l te
s
tin
g
d
a
ta
T
o
t
al
co
r
r
ect
d
at
a
A
ccu
r
acy
(
%
)
C
o
mp
u
ta
tio
n
time
(
s
)
1
100
68
68
43
2
100
20
20
57
3
100
20
20
54
4
100
20
20
55
F
i
gur
e
6.
F
e
a
t
ur
e
v
al
u
es
of
va
r
i
ous
l
e
ve
l
s
of
de
c
o
m
pos
i
t
i
on
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
1693
-
6930
T
E
L
KOM
NI
KA
T
el
eco
m
m
u
n
C
om
put
E
l
C
ont
r
o
l
,
Vo
l
.
18
, N
o
.
5
,
O
c
t
obe
r
2020:
2265
-
2275
2272
−
M
ot
he
r
w
av
el
et
p
ar
am
et
er
s
i
m
p
act
T
h
e t
es
t
s
w
er
e
car
r
i
ed
o
u
t
w
i
t
h
f
o
u
r
t
y
p
es
o
f
m
o
t
h
er
w
av
el
et
p
ar
am
et
er
s
:
H
aar
,
d
b
3
,
d
b
5
,
an
d
db7.
T
e
s
t
e
d i
t
w
i
t
h
t
he
be
s
t
pa
r
a
m
e
t
e
r
s
i
n t
he
pr
e
vi
ous
pa
r
a
m
e
t
e
r
s
:
t
he
B
l
ue
l
a
ye
r
,
L
L
s
ub
-
ba
nd,
a
nd
l
e
ve
l
1 de
c
om
pos
i
t
i
on.
T
he
pe
r
f
or
m
a
nc
e
r
e
s
ul
t
s
w
e
r
e
l
i
s
t
e
d i
n T
a
bl
e
9.
T
he
be
s
t
t
e
s
t
r
e
s
ul
t
s
obt
a
i
ne
d f
r
om
H
a
a
r
m
o
t
h
er
w
av
el
et
.
I
n F
i
gur
e
7
,
t
he
gr
a
ph
s
how
s
t
ha
t
t
he
di
f
f
e
r
e
nt
t
ype
s
of
m
ot
he
r
w
a
ve
l
e
t
s
c
a
us
e
di
f
f
e
r
e
nt
f
o
r
m
s
o
f
ch
ar
act
er
i
s
t
i
c i
n
t
h
e
s
am
e cl
as
s
.
S
o,
t
h
e u
s
ed
o
f
cer
t
ai
n
m
o
t
h
er
w
av
el
et
s
i
n
a s
y
s
t
em
can
pr
ovi
de
d
a
uni
que
ne
s
s
f
or
e
a
c
h c
l
a
s
s
s
o t
ha
t
t
he
y c
a
n be
d
i
s
t
i
ngui
s
he
d
b
et
w
een
each
cl
as
s
T
ab
l
e 9
.
M
o
t
h
er
w
av
el
et
p
ar
am
et
er
s
p
er
f
o
r
m
an
ces
M
o
t
h
er
Wav
el
et
T
o
ta
l te
s
tin
g
d
a
ta
T
o
t
al
co
r
r
ect
d
at
a
A
ccu
r
acy
(
%
)
C
om
p
u
ta
tio
n
time
(
s
)
H
aar
100
68
68
43
db3
100
20
20
43
db5
100
34
34
43
db7
100
50
50
44
F
i
gur
e
7.
F
e
a
t
ur
e
va
l
ue
s
of
va
r
i
ous
m
o
t
h
er
w
av
el
et
−
A
m
ount
of
cl
u
s
t
er
p
ar
am
et
er
s
i
m
p
act
D
o
n
e t
o
t
es
t
t
h
e
cl
u
s
t
er
p
ar
am
et
er
s
u
s
ed
i
n
H
M
M
cl
as
s
i
f
i
cat
i
o
n
.
C
l
u
s
t
er
s
t
h
at
w
er
e
b
ei
n
g
an
al
y
zed
e
r
e
50,
100,
200
,
400
,
800,
a
nd 1000
.
T
e
s
t
e
d i
t
w
i
t
h t
he
be
s
t
pa
r
a
m
e
t
e
r
s
f
r
om
pr
e
vi
ous
t
e
s
t
s
.
I
n
T
a
bl
e
10,
t
he
be
s
t
num
be
r
obt
a
i
ne
d i
n 800.
I
n
F
i
gur
e
8
,
i
t
ca
n
b
e s
een
i
n
t
h
e g
r
ap
h
t
h
at
t
h
e ch
ar
act
er
i
s
t
i
cs
o
f
t
h
e cl
u
s
t
er
s
o
f
5
0
cau
s
ed
t
h
e ch
ar
act
er
i
s
t
i
cs
at
t
h
e
s
am
e t
y
p
e o
b
t
ai
n
ed
t
h
e s
m
al
l
accu
r
acy
co
m
p
ar
ed
t
o
8
0
0
cl
u
s
t
er
s
.
T
ab
l
e 1
0
.
A
m
o
u
n
t
o
f
cl
u
s
t
er
p
a
r
am
et
er
s
p
er
f
o
r
m
an
ces
C
lu
s
te
r
to
ta
l
T
o
ta
l
te
s
tin
g
d
a
ta
T
o
t
al
co
r
r
ect
d
at
a
A
ccu
r
acy
(
%
)
C
o
mp
u
ta
tio
n
time
(
s
)
50
100
58
58
52
100
100
36
36
51
200
100
45
45
51
400
100
55
55
52
800
100
68
68
43
1000
100
25
25
53
−
N
u
m
b
er
o
f
s
t
at
e i
m
p
act
T
h
e n
ex
t
s
t
ep
w
as
t
o
t
es
t
t
h
e s
t
at
e
p
ar
am
et
er
s
u
s
ed
i
n
t
h
e
H
M
M
cl
as
s
i
f
i
cat
i
o
n
t
o
s
y
s
t
em
accu
r
acy
an
d
c
om
put
a
t
i
on t
i
m
e
.
T
he
s
t
a
t
e
t
ha
t
w
e
r
e
us
e
d:
4
,
5
,
25,
50,
100,
a
nd 150
.
T
he
be
s
t
pe
r
f
o
r
m
a
nc
e
r
e
s
ul
t
s
w
a
s
5
s
ta
te
s
a
n
d
th
e
r
e
s
t w
e
r
e
lis
te
d
in
T
a
b
le
1
1
.
T
h
e
b
e
s
t p
a
r
a
me
te
r
s
w
ith
th
e
n
u
mb
e
r
o
f
s
imila
r
s
ta
te
s
w
as
5
st
a
t
e
s.
T
hi
s
ha
ppe
ne
d be
c
a
us
e
t
he
c
on
c
e
pt
of
H
M
M
t
ha
t
ba
s
i
c
a
l
l
y br
oke
dow
n t
he
da
t
a
a
s
m
a
ny a
s
t
he
de
s
i
r
e
d s
t
a
t
e
.
S
o,
i
f
th
e
v
a
lu
e
o
f
th
e
s
ta
te
u
s
e
d
is
n
o
t
r
ig
h
t
,
it
w
ill
ma
k
e
it d
if
f
ic
u
lt to
id
e
n
ti
fy
t
he
te
s
t d
a
ta
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
el
eco
m
m
u
n
C
om
put
E
l
C
ont
r
o
l
H
and ge
s
t
ur
e
r
e
c
ogni
t
i
on us
i
ng di
s
c
r
e
t
e
w
av
e
l
e
t
t
r
ans
f
or
m
and.
.
.
(
E
r
i
z
k
a B
anuw
at
i
C
andr
as
ar
i
)
2273
F
i
gur
e
8.
F
e
a
t
ur
e
va
l
ue
i
n c
l
us
t
e
r
50
T
ab
l
e
11.
N
u
m
b
er
o
f
s
t
at
e i
m
p
act
p
er
f
o
r
m
an
ces
S
ta
te
to
ta
l
T
o
ta
l te
s
tin
g
d
a
ta
T
o
t
al
co
r
r
ect
d
at
a
A
ccu
r
acy
(
%
)
C
o
mp
u
ta
tio
n
time
(
s
)
4
100
30
30
42,4
5
100
72
72
53
25
100
49
49
53
50
100
32
32
58
100
100
51
51
63
150
100
42
42
71
3.
2.
T
es
t
i
n
g
t
h
e
d
at
a
b
at
c
h
T
h
e d
at
a t
h
at
t
es
t
ed
w
er
e s
h
o
w
n
i
n
T
ab
l
e 1
2
.
T
he
c
onc
l
us
i
on w
a
s
t
he
r
e
c
ogni
t
i
on s
ys
t
e
m
w
i
t
h
t
h
e D
WT
an
d
t
h
e H
M
M
can
i
d
en
t
i
f
y
w
el
l
i
f
t
h
e
t
r
a
i
n
i
n
g
an
d
t
es
t
i
m
ag
es
p
r
es
en
t
ed
w
as
be
t
w
e
e
n 60%
a
nd 40%
o
f
al
l
d
at
a
i
n
each
cl
as
s
.
T
ab
l
e
12.
D
at
a b
at
ch
t
es
t
i
n
g
p
er
f
o
r
m
an
ces
D
a
ta
(
T
r
a
in
in
g
-
T
es
t
)
A
ccu
r
acy
(
%
)
C
o
mp
u
ta
tio
n
time
(
s
)
40
–
10
20
22
30
–
20
72
53
25
–
25
20
53
3.
3.
C
la
s
s
if
ic
a
t
io
n
te
s
ti
n
g
C
la
s
s
if
ic
a
tio
n
te
s
tin
g
w
a
s
done
by c
om
pa
r
i
ng t
he
a
c
c
ur
a
c
y a
nd c
om
put
a
t
i
on t
i
m
e
of
t
w
o c
l
a
s
s
i
f
i
c
a
t
i
on
m
e
t
hods
,
K
-
N
ear
es
t
N
ei
g
h
b
o
r
(
K
-
N
N
)
an
d
H
M
M
.
T
h
e cl
as
s
i
f
i
cat
i
o
n
d
at
a w
as
t
ak
en
f
r
o
m
t
h
e t
r
ai
n
i
n
g
d
at
a t
o
t
he
t
r
a
i
ni
ng
da
t
a
a
nd
t
r
a
i
ni
ng
da
t
a
t
o
t
he
t
e
s
t
da
t
a
a
s
s
how
n i
n
T
a
bl
e
13
.
Ba
s
e
d on
T
a
bl
e
12,
H
M
M
ha
d a
l
ow
e
r
a
c
c
ur
a
c
y w
he
n t
e
s
t
e
d a
t
r
a
i
ni
ng t
o t
r
a
i
ni
ng da
t
a
,
w
he
n c
om
pa
r
e
d t
o t
r
a
i
ni
ng t
o t
e
s
t
da
t
a
.
T
hi
s
ha
ppe
ne
d due
t
o
t
he
pe
r
c
e
nt
a
ge
of
da
t
a
w
he
n t
he
t
r
a
i
ni
ng t
e
s
t
e
d w
i
t
h t
r
a
i
ni
ng da
t
a
i
s
50
-
50%
.
W
he
r
e
a
s
,
w
he
n t
e
s
t
e
d t
r
a
i
ni
ng t
o
th
e t
es
t
d
at
a h
ad
a p
r
es
en
t
at
i
o
n
o
f
6
0
-
40%
.
B
as
ed
o
n
T
ab
l
e 1
3
,
i
t
can
b
e s
een
t
h
at
t
h
e
t
es
t
w
as
al
s
o
d
o
n
e w
i
t
h
o
t
h
er
d
at
as
et
s
w
i
t
h
accu
r
acy
an
d
co
m
p
u
t
at
i
o
n
al
t
i
m
e b
et
t
er
t
h
an
a
p
er
f
o
r
m
an
ce w
i
t
h
M
ar
cel
s
t
a
t
i
c
ha
nd
pos
t
ur
e
d
at
ab
a
se
[
2
9
,
30]
.
T
h
e d
at
as
et
h
ad
a
l
o
w
er
p
er
f
o
r
m
an
ce co
m
p
ar
ed
t
o
t
h
e A
u
t
h
o
r
d
at
as
et
t
h
at
h
as
a r
es
i
ze i
m
ag
e
m
e
a
s
ur
e
d a
t
76×
66 pi
xe
l
s
w
h
i
ch
i
n
t
h
i
s
r
es
ear
ch
w
a
s
t
he
r
i
ght
ha
nd
as
s
een
i
n
F
i
gur
e
9
.
T
ab
l
e
1
3
.
C
la
s
s
if
ic
a
tio
n
t
e
s
t
i
ng pe
r
f
or
m
a
nc
e
on
c
l
a
s
s
i
f
i
c
a
t
i
on m
e
t
hods
D
at
as
et
C
la
s
s
if
ic
a
tio
n
A
ccu
r
acy
(
%
)
C
om
put
a
tio
n
time
(
s
)
T
r
a
in
in
g
T
es
t
T
r
a
in
in
g
T
es
t
D
at
as
et
S
e
b
a
s
tie
n
K
-
NN
100%
100%
3
.
78
2
.
7
HM
M
38%
58%
2
.
6
2
.
01
Wr
i
t
er
’
s
D
at
as
et
K
-
NN
100%
100%
58
40
HM
M
55%
72%
68
53
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
1693
-
6930
T
E
L
KOM
NI
KA
T
el
eco
m
m
u
n
C
om
put
E
l
C
ont
r
o
l
,
Vo
l
.
18
, N
o
.
5
,
O
c
t
obe
r
2020:
2265
-
2275
2274
(
a)
(
b)
(
c)
(
d)
(
e)
P
i
ct
u
r
e
S
e
ba
t
i
e
n’
s
da
t
a
s
e
t
of
ha
nd ge
s
t
ur
e
;
(a
) L
e
t
t
e
r A
,
(b
) L
e
t
t
e
r B
,
(c
) L
e
t
t
e
r C
,
(d
) P
o
i
n
t
G
e
s
t
u
re
,
(e
)
N
u
m
b
e
r 5
4.
CO
NCL
US
I
O
N
T
hi
s
pa
pe
r
pr
opos
e
d a
ha
nd ge
s
t
ur
e
r
e
c
ogni
t
i
on s
y
s
t
e
m
t
ha
t
ha
s
5 t
ype
s
of
ge
s
t
ur
e
s
:
le
tte
r
A
,
le
tte
r
B
,
l
e
t
t
e
r
C
,
poi
nt
,
a
nd nu
m
be
r
5
(
f
i
ve
)
.
T
h
e b
es
t
p
ar
am
et
er
s
ar
e
bl
ue
l
ay
er
,
l
ow
-
l
ow
s
ub
-
ba
nd
,
l
ev
el
1
de
c
om
pos
i
t
i
on,
H
a
a
r
m
ot
he
r
w
a
ve
l
e
t
on D
W
T
pa
r
a
m
e
t
e
r
s
,
800 c
l
us
t
e
r
num
be
r
s
an
d
5
s
t
at
e s
t
at
es
o
n
t
h
e H
M
M
p
ar
am
et
er
.
T
h
e
accu
r
acy
an
d
co
m
p
u
t
at
i
o
n
t
i
m
e
out
c
om
e
f
r
om
th
e
s
y
s
te
m
w
er
e
72%
a
nd
53
s
e
c
onds
r
es
p
ect
i
v
el
y
.
T
he
be
s
t
a
m
ount
of
da
t
a
t
es
t
ed
i
s
on 30 t
r
a
i
ni
ng i
m
a
ge
s
a
nd 20 t
e
s
t
i
m
a
ge
s
.
L
a
ye
r
s
t
ha
t
ha
d hi
gh
accu
r
acy
w
oul
d
ha
ve
a
good
c
ont
r
a
s
t
a
nd b
r
i
ght
ne
s
s
r
a
tio
.
T
he
da
t
a
s
e
t
i
m
a
ge
ha
d a
hi
gh
c
ont
r
a
s
t
a
nd b
r
i
ght
ne
s
s
on t
he
B
l
ue
l
a
ye
r
due
t
o
t
he
hi
gh
f
r
e
que
nc
y o
f
pi
x
e
l
s
w
hi
c
h a
r
e
0
t
o
45
f
or
hi
gh
-
i
nt
e
ns
i
t
y va
l
ue
s
a
t
pi
xe
l
s
0 t
o
231 c
om
pa
r
e
d t
o o
t
he
r
l
a
ye
r
.
D
WT
h
ad
t
h
r
ee
t
es
t
p
ar
am
et
er
s
t
h
ey
ar
e
su
b
-
ba
n
d t
ype
,
de
c
om
pos
i
t
i
on
l
ev
el
,
an
d
m
o
t
h
er
w
av
el
et
.
T
he
s
ub
-
b
an
d
p
ar
am
et
er
w
as
p
r
o
ces
s
ed
t
o
o
b
t
ai
n
a s
m
o
o
t
h
i
m
ag
e ch
ar
act
er
i
s
t
i
c i
n
t
h
e
L
L
s
u
b
-
ba
nd.
D
e
c
om
pos
i
t
i
on
l
ev
el
p
ar
am
et
er
s
w
as
t
he
p
r
oc
e
s
s
of
c
onve
r
t
i
ng
i
m
a
ge
s
i
nt
o
a
s
i
m
pl
e
f
or
m
t
o obt
a
i
n
uni
que
ch
ar
act
er
i
s
t
i
cs
o
f
a g
o
o
d
i
m
ag
e.
N
ex
t
,
d
i
ffe
re
n
t
t
y
p
e o
f
m
o
t
h
er
w
av
el
et
cau
s
ed
t
h
e
uni
qu
e
ne
s
s
in
t
he
ch
ar
act
er
i
s
t
i
c
.
T
h
e H
M
M
cl
as
s
i
f
i
cat
i
o
n
h
ad
t
w
o
t
es
t
p
ar
am
et
er
s
,
t
h
e cl
u
s
t
er
s
an
d
t
h
e
s
t
at
e
s
.
E
f
f
ect
o
f
cl
u
s
t
er
p
ar
am
et
er
s
w
as
t
a
ki
ng
t
he
f
eat
u
r
es
t
o
b
e u
s
ed
.
T
he
c
l
us
t
e
r
va
l
ue
s
m
us
t
be
a
ppr
opr
i
a
t
e
d
d
e
te
r
min
a
te
b
ecau
s
e
th
e
c
h
a
r
a
c
te
r
is
tic
v
a
lu
e
s
o
f
e
a
c
h
c
la
s
s
w
ill b
e
s
imi
la
r
.
I
n
te
s
tin
g
th
e
c
la
s
s
if
ic
a
tio
n
f
o
r
tr
a
in
in
g
to
t
r
a
i
n
in
g
h
a
d
a
l
ow
e
r
a
c
c
ur
a
c
y of
55
%
c
om
pa
r
e
d w
i
t
h
t
e
s
t
i
ng t
r
a
i
ni
ng t
o
da
t
a
w
hi
c
h i
s
72%
.
T
hi
s
ha
ppe
ne
d due
t
o
t
he
pe
r
c
e
nt
a
ge
of
da
t
a
w
he
n
t
he
t
r
a
i
ni
ng
t
e
s
t
e
d w
i
t
h ot
he
r
t
r
a
i
ni
ng
da
t
a
w
as
50
-
50%
.
W
he
r
e
a
s
,
w
he
n t
e
s
t
i
ng
t
r
a
i
ni
ng da
t
a
t
o
t
he
t
e
s
t
da
t
a
ha
d a
p
r
e
s
e
nt
a
t
i
on of
6
0%
-
40%
.
C
ogi
t
a
t
e
d a
nd c
onc
l
ude
d t
o
c
r
e
a
t
e
our
ow
n da
t
a
s
e
t
b
ecau
s
e w
h
en
S
eb
as
t
i
en
d
at
as
et
w
as
u
s
ed
,
i
t
j
u
s
t
h
ad
t
h
e
accu
r
acy
at
5
8
%
w
i
t
h
t
h
e i
m
ag
e s
i
ze o
f
7
6
×
6
6
p
i
x
el
.
F
ur
t
he
r
,
i
t
ha
d
2
nd
l
e
ve
l
de
c
om
pos
i
t
i
on a
nd db5
m
ot
he
r
w
a
ve
l
e
t
.
T
he
m
e
nt
i
one
d c
a
us
e
d by
f
e
a
t
ur
e
e
xt
r
a
c
t
i
on
pr
oc
e
s
s
w
i
t
h 2
nd
l
ev
el
D
WT
an
d
cl
as
s
i
f
i
cat
i
o
n
w
i
t
h
H
M
M
en
co
u
n
t
er
ed
t
h
r
ee t
i
m
es
t
h
e
co
m
p
r
es
s
i
o
n
p
r
o
ces
s
.
T
h
e co
n
s
eq
u
en
ces
w
er
e t
h
e g
es
t
u
r
es
t
ak
en
f
r
o
m
t
h
e i
m
ag
es
w
as
s
o
s
m
al
l
s
o
i
t
s
h
a
r
d
er
t
o
cl
as
s
i
f
i
ed
i
t
.
H
en
ce,
t
o
m
ad
e t
h
e accu
r
acy
h
i
g
h
er
w
e p
r
oduc
e
d our
ow
n
da
t
a
s
e
t
w
i
t
h a
good br
i
ght
e
s
s
a
nd c
ont
r
a
s
t
va
l
ue
.
T
he
r
e
a
f
t
e
r
,
t
he
r
e
s
ol
ut
i
on w
a
s
boos
t
e
d t
o
128×
128
pi
xe
l
s
r
e
s
ul
t
i
ng t
he
a
c
c
ur
a
c
y j
um
pe
d
up by
14%
t
o 72%
.
R
EF
ER
EN
CE
[1
]
Ast
a
ne
h A
.
A
.
,
G
he
isa
r
i S
.
,
“
R
e
vie
w a
nd C
om
pa
r
is
on o
f
R
out
in
g
M
e
tr
ic
s i
n C
og
ni
ti
ve
R
a
di
o Ne
tw
or
k
s
,
”
Eme
r
g Sc
i
J
.
,
vol.
2
, n
o
.
4
,
pp
.
1
91
-
20
1
,
20
18
.
[2
]
B
uon
o A
.
,
Ramd
h
an
A
.
,
R
uv
in
na
,
“
I
ntr
od
uc
ti
on
of
I
nd
o
ne
s
ia
n
W
or
ds
w
it
h
H
id
de
n
M
a
r
ko
v
M
o
de
l
(
HM
M
)
U
si
n
g
the
Bau
m
-
W
e
lc
h
Al
gor
it
hm
(
I
n B
a
ha
sa
:
P
e
n
ge
na
la
n Ka
ta
B
e
r
ba
ha
sa
I
n
do
n
e
s
ia
de
n
ga
n
Hi
dde
n M
a
r
ko
v M
o
de
l
(
H
M
M
)
M
e
ng
gu
na
ka
n Al
gor
itm
e
B
a
um
-
W
el
c
h
)
,”
J
ur
na
l
Il
m
u
K
om
put
e
r
,
v
ol.
6
,
n
o
.
2
,
pp.
32
-
40
,
20
08
.
[3
]
Da
n
ie
l J
.
,
Mart
i
n
J
.
H.
,
“
S
pe
e
c
h a
nd L
a
n
gua
ge
P
r
oc
e
s
si
ng a
n
in
tr
o
duc
ti
on
to
na
t
ur
a
l
la
n
gua
ge
pr
oc
e
s
sin
g,
c
om
p
uta
ti
ona
l li
ng
ui
st
ic
s,
a
n
d s
pe
e
c
h r
e
c
o
gn
it
io
n
,
”
PE
AR
SO
N
,
S
e
c
on
d Ed
it
io
n
,
20
18.
[4
]
F
ink G
.
A.
,
“
M
a
r
ko
v M
ode
ls f
or
P
a
tte
r
n
R
e
c
og
ni
ti
on
,
”
S
pri
ng
e
r L
o
nd
on,
S
e
c
o
n
E
di
ti
on,
20
14
.
[5
]
G
opi E
.
S.
,
“
Dig
ita
l S
pe
e
c
h P
r
oc
e
s
si
ng U
si
ng M
a
tla
b
,”
2
014
.
[
Onl
ine
]
.
A
va
i
la
b
le
: ht
tp
:/
/l
in
k.
spr
in
ge
r
.
c
om
/1
0.
10
07
/9
78
-
81
-
322
-
16
77
-
3
[6
]
Ja
m
a
l
ud
in
A
.
,
Hu
da
A
.
F
.
,
S
ah
y
an
d
ari
R
.
,
“
I
ntr
od
uc
t
io
n
of
De
a
d
No
on
L
a
w
Us
in
g
the
Hi
dde
n M
a
r
k
ov
M
o
de
l (
I
n B
a
h
a
sa
:
P
e
nge
na
la
n L
a
f
a
l Hu
kum
N
un M
a
ti M
e
n
gg
una
ka
n
Hi
dde
n
M
a
r
ko
v M
ode
l
,”
L
O
G
IK
@
,
vo
l.
6
, n
o
.
1
, p
p
.
1
-
10
,
20
16
.
[7
]
Ke
s
ki
n C
.
,
Er
ka
n A
.
,
Aka
r
un L
.
,
“
R
e
a
l tim
e
ha
nd tr
a
c
ki
ng a
n
d 3D
ge
s
tur
e
r
e
c
o
gn
it
io
n f
or
i
nte
r
a
c
ti
ve
in
te
r
f
a
c
e
s usi
n
g
H
MM
,”
Pr
oc
e
e
d
in
gs
o
f
I
n
te
r
na
ti
on
al
C
o
nfe
re
nc
e
on
A
rt
ifi
c
i
al
N
e
u
ra
l
N
e
tw
ork
s
,
20
03.
[8
]
M
a
r
dh
iy
ya
A
.
,
Hida
ya
t B
.
,
Aul
ia
S
.
,
“
Ha
nd
wr
i
ti
ng
de
te
c
ti
on
u
si
ng
a
da
pt
ive
se
gm
e
n
ta
t
io
n
a
nd
h
id
de
n
m
a
r
k
ov
m
ode
l
m
e
th
od
s (
I
n B
a
ha
sa
: De
te
k
si tu
l
i
sa
n ta
nga
n m
e
ng
gu
na
ka
n m
e
to
de
se
gm
e
nta
si a
da
pt
if
da
n hi
dde
n m
a
r
ko
v m
o
de
l,
”
C
onf
e
re
nc
e
:
C
I
T
EE 20
15
,
a
t Y
og
ya
ka
r
ya
,
vo
l.
7,
20
15.
Evaluation Warning : The document was created with Spire.PDF for Python.