I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
p
ute
r
E
ng
in
ee
ring
(
I
J
E
CE
)
Vo
l.
7
,
No
.
4
,
A
u
g
u
s
t
201
7
,
p
p
.
1
9
1
5
~
1
9
2
2
I
SS
N:
2
0
8
8
-
8708
,
DOI
: 1
0
.
1
1
5
9
1
/
i
j
ec
e
.
v7
i
4
.
p
p
1
9
1
5
-
1922
1915
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ia
e
s
jo
u
r
n
a
l.c
o
m/o
n
lin
e/in
d
ex
.
p
h
p
/I
JE
C
E
Uni
m
o
da
l Mu
lti
-
Feature
F
usio
n a
nd
o
ne
-
di
m
e
nsio
na
l H
idden
M
a
rk
o
v
Mo
dels
f
or
Low
-
Reso
lution Face
Recog
nitio
n
O
t
h
m
a
ne
E
l M
eslo
uh
i
1
,
Z
ine
b E
lg
a
rr
a
i
2
,
M
us
t
a
ph
a
K
a
rdo
uchi
3
,
H
a
k
i
m
Alla
li
4
1,
4
L
a
b
.
S
IE
F
P
O
o
f
Ib
n
Zo
h
r
U
n
iv
e
rsit
y
A
g
a
d
ir,
M
o
ro
c
c
o
2
F
S
T
o
f
Ha
ss
a
n
1
st
Un
iv
e
rsity
S
e
tt
a
t,
M
o
r
o
c
c
o
3
D
é
p
a
rte
m
e
n
t
d
’In
f
o
rm
a
ti
q
u
e
o
f
Un
iv
e
rsit
é
d
e
M
o
n
c
to
n
,
Ca
n
a
d
a
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
an
6
,
2
0
1
7
R
ev
i
s
ed
Ma
r
1
0
,
2
0
1
7
A
cc
ep
ted
Ma
r
2
4
,
2
0
1
7
T
h
e
o
b
jec
ti
v
e
o
f
lo
w
-
re
so
lu
ti
o
n
fa
c
e
r
e
c
o
g
n
it
io
n
is
to
i
d
e
n
ti
f
y
f
a
c
e
s
f
ro
m
s
m
a
ll
siz
e
o
r
p
o
o
r
q
u
a
li
ty
i
m
a
g
e
s
w
it
h
v
a
r
y
in
g
p
o
se
,
il
lu
m
in
a
ti
o
n
,
e
x
p
re
ss
io
n
,
e
tc.
In
t
h
is
w
o
rk
,
w
e
p
ro
p
o
se
a
r
o
b
u
st
l
o
w
f
a
c
e
re
c
o
g
n
it
io
n
tec
h
n
i
q
u
e
b
a
se
d
o
n
o
n
e
-
d
im
e
n
sio
n
a
l
Hi
d
d
e
n
M
a
r
k
o
v
M
o
d
e
ls.
F
e
a
tu
re
s
o
f
e
a
c
h
f
a
c
ial
ima
g
e
a
re
e
x
tr
a
c
ted
u
sin
g
th
re
e
ste
p
s:
fi
rstl
y
,
b
o
th
G
a
b
o
r
fi
lt
e
rs
a
n
d
Histo
g
ra
m
o
f
Orie
n
ted
G
ra
d
ien
ts
(HO
G
)
d
e
sc
r
ip
to
r
a
re
c
a
lcu
late
d
.
S
e
c
o
n
d
ly
,
th
e
siz
e
o
f
th
e
se
fe
a
tu
re
s
is
r
e
d
u
c
e
d
u
sin
g
th
e
L
in
e
a
r
Dis
c
ri
m
in
a
n
t
A
n
a
l
y
sis
(L
D
A)
m
e
th
o
d
in
o
r
d
e
r
to
re
m
o
v
e
re
d
u
n
d
a
n
t
in
f
o
rm
a
ti
o
n
.
F
in
a
ll
y
,
t
h
e
re
d
u
c
e
d
f
e
a
tu
re
s
a
re
c
o
m
b
in
e
d
u
si
n
g
Ca
n
o
n
ica
l
C
o
rre
latio
n
A
n
a
ly
si
s
(CCA
)
m
e
th
o
d
.
Un
li
k
e
e
x
isti
n
g
te
c
h
n
iq
u
e
s
u
sin
g
HMM
s,
in
w
h
ich
a
u
th
o
rs
c
o
n
sid
e
r
e
a
c
h
sta
te
to
re
p
re
se
n
t
o
n
e
f
a
c
ial
re
g
i
o
n
(e
y
e
s,
n
o
se
,
m
o
u
th
,
e
tc),
th
e
p
ro
p
o
se
d
s
y
ste
m
e
m
p
lo
y
s
1
D
-
HMM
s
w
it
h
o
u
t
a
n
y
p
rio
r
k
n
o
w
led
g
e
a
b
o
u
t
t
h
e
lo
c
a
li
z
a
ti
o
n
o
f
in
tere
st
re
g
io
n
s
in
th
e
f
a
c
ial
i
m
a
g
e
.
P
e
rf
o
r
m
a
n
c
e
o
f
th
e
p
ro
p
o
se
d
m
e
th
o
d
w
il
l
b
e
m
e
a
su
re
d
u
si
n
g
th
e
A
R
d
a
tab
a
se
.
K
ey
w
o
r
d
:
1D
-
HM
Ms
C
an
o
n
ical
co
r
r
elatio
n
an
al
y
s
is
Face
r
ec
o
g
n
itio
n
I
n
ter
p
o
latio
n
L
o
w
-
r
e
s
o
lu
tio
n
Co
p
y
rig
h
t
©
2
0
1
7
In
stit
u
te o
f
A
d
v
a
n
c
e
d
E
n
g
i
n
e
e
rin
g
a
n
d
S
c
ien
c
e
.
Al
l
rig
h
ts
re
se
rv
e
d
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Oth
m
an
e
E
l M
e
s
lo
u
h
i
,
L
ab
.
SIE
FP
O
o
f
I
b
n
Z
o
h
r
U
n
i
v
er
s
it
y
Ag
ad
ir
,
Mo
r
o
cc
o
.
E
m
ail: o
.
el
m
e
s
lo
u
h
i
@
u
iz.
ac
.
m
a
1.
I
NT
RO
D
UCT
I
O
N
F
ace
r
ec
o
g
n
itio
n
(
F
R
)
is
t
h
e
p
r
o
ce
s
s
of
au
to
m
atica
ll
y
id
en
ti
f
y
in
g
or
v
er
if
y
i
n
g
a
p
er
s
o
n
f
r
o
m
h
i
s
f
ac
ial
i
m
a
g
e.
In
r
ea
l
lif
e
ap
p
li
ca
tio
n
,
s
u
c
h
as,
v
id
eo
s
u
r
v
e
il
lan
ce
ca
m
er
as
(
p
lace
d
in
b
an
k
s
,
s
u
p
er
m
ar
k
et
s
o
r
in
p
u
b
lic
s
tr
ee
ts
)
,
s
u
b
j
ec
ts
ar
e
f
ar
f
r
o
m
ca
m
er
a
s
,
so
f
ace
r
e
g
io
n
s
ten
d
to
be
s
m
all.
Ma
c
h
i
n
e
f
ac
e
id
en
tific
atio
n
f
r
o
m
t
h
i
s
k
i
n
d
of
i
m
a
g
es
b
ec
o
m
e
s
an
i
m
p
o
r
ta
n
t
s
cien
tific
i
s
s
u
e
to
s
o
l
v
e.
T
h
is
is
s
u
e
is
c
alled
l
o
w
-
r
eso
l
u
tio
n
f
ace
r
ec
o
g
n
itio
n
.
L
i
k
e
cla
s
s
ical
f
ac
e
r
ec
o
g
n
iti
o
n
s
y
s
te
m
s
,
lo
w
-
r
e
s
o
lu
t
io
n
f
ac
e
r
ec
o
g
n
it
io
n
p
r
o
ce
d
u
r
e
co
n
s
i
s
ts
o
f
2
m
aj
o
r
task
s
:
f
ea
tu
r
es
e
x
tr
ac
tio
n
an
d
class
if
ier
d
esig
n
in
g
.
B
o
th
h
a
v
e
a
s
i
g
n
i
f
ica
n
t
i
m
p
ac
t
o
n
th
e
r
eliab
ilit
y
o
f
r
ec
o
g
n
itio
n
m
et
h
o
d
.
Ho
w
e
v
er
,
a
lo
w
-
r
eso
l
u
tio
n
f
ac
e
r
ec
o
g
n
itio
n
s
y
s
te
m
n
ee
d
s
to
co
n
s
id
er
t
h
e
p
ar
ticu
lar
p
r
o
b
lem
o
f
d
i
m
en
s
io
n
al
m
i
s
m
a
tch
b
et
w
e
en
p
r
o
b
e
(
w
it
h
lo
w
-
r
e
s
lo
u
tio
n
)
an
d
g
aller
y
(
g
e
n
er
all
y
w
i
th
h
ig
h
-
r
eso
lu
t
io
n
)
i
m
a
g
es
b
ef
o
r
e
ex
tr
ac
ti
n
g
f
ac
e
f
ea
tu
r
e
s
.
T
h
is
is
s
u
e
ca
n
b
e
r
eso
lv
ed
b
y
u
s
i
n
g
t
h
r
ee
m
ai
n
w
a
y
s
:
(
1
)
A
p
p
l
y
in
g
in
ter
p
o
latio
n
o
r
s
u
p
er
-
r
e
s
o
lu
t
io
n
to
th
e
p
r
o
b
e
f
ac
ial
i
m
ag
es,
(
2
)
p
r
o
j
ec
tin
g
b
o
th
p
r
o
b
e
an
d
g
aller
y
f
ac
ia
l
i
m
a
g
es
i
n
to
an
i
n
ter
-
r
eso
lu
t
io
n
s
p
ac
e
o
r
(
3
)
d
o
w
n
s
a
m
p
l
i
n
g
all
g
aller
y
i
m
a
g
e
s
an
d
th
e
n
p
er
f
o
r
m
m
a
tch
i
n
g
in
lo
w
r
eso
l
u
tio
n
.
An
o
v
er
v
ie
w
o
f
th
is
m
eth
o
d
s
i
s
g
i
v
e
n
in
[
1
]
.
Fo
r
th
e
f
ea
tu
r
e
s
ex
tr
ac
tio
n
a
n
d
class
i
f
ica
tio
n
tas
k
s
,
v
ar
io
u
s
ap
p
r
o
ac
h
es
h
av
e
b
ee
n
p
r
o
p
o
s
ed
s
u
c
h
as:
R
QC
r
co
lo
r
f
ea
t
u
r
es
w
i
th
n
ea
r
est
n
ei
g
h
b
o
r
class
i
f
ier
[
2
]
,
L
o
ca
l
Fre
q
u
en
c
y
De
s
cr
ip
to
r
w
i
th
n
ea
r
est
n
ei
g
h
b
o
r
class
i
f
ier
[
3
]
an
d
Su
p
p
o
r
t V
ec
to
r
Data
Descr
ip
tio
n
w
i
th
s
u
p
p
o
r
t v
ec
to
r
m
ac
h
i
n
es (
SVM)
cl
ass
i
f
ier
[
4
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
J
E
C
E
Vo
l.
7
,
No
.
4
,
A
u
g
u
s
t
2
0
1
7
:
1
9
1
5
–
1
9
2
2
1916
I
n
t
h
is
p
ap
er
,
w
e
u
s
e
th
e
cu
b
i
c
in
ter
p
o
latio
n
to
in
cr
ea
s
e
t
h
e
s
p
ec
ial
r
eso
l
u
tio
n
o
f
L
R
i
m
a
g
es
to
d
ea
l
w
it
h
t
h
e
d
i
m
e
n
s
io
n
al
m
is
m
atc
h
p
r
o
b
lem
.
T
h
is
ch
o
ice
i
s
j
u
s
ti
f
ied
b
y
th
e
f
ac
t
t
h
at
it
’
s
ex
tr
em
el
y
f
a
s
t
an
d
it
ca
n
b
e
ap
p
lied
o
n
a
s
in
g
le
i
m
a
g
e
u
n
l
ik
e
s
u
p
er
-
r
eso
l
u
tio
n
tec
h
n
i
q
u
es
w
h
ic
h
u
s
e
g
aller
y
s
et
o
f
L
R
o
r
HR
i
m
a
g
e
s
.
R
eg
ar
d
in
g
t
h
e
f
ac
ial
f
ea
tu
r
e
s
ex
tr
ac
tio
n
,
t
h
e
p
r
o
p
o
s
ed
s
y
s
te
m
e
x
tr
ac
t
f
ac
ia
l
f
ea
t
u
r
es
b
y
f
u
s
in
g
t
w
o
lo
ca
l
d
escr
ip
to
r
s
:
Gab
o
r
w
a
v
elet
s
[
5
]
,
[
6
]
an
d
HOG
d
escr
ip
to
r
s
[
7
]
.
Gab
o
r
w
a
v
elet
s
(
f
il
ter
s
)
ar
e
in
v
ar
ian
t
to
r
o
tatio
n
,
s
ca
le
an
d
tr
a
n
s
la
tio
n
.
I
n
ad
d
itio
n
,
th
e
y
ar
e
r
o
b
u
s
t
to
i
m
ag
e
n
o
i
s
e.
A
l
s
o
,
His
to
g
r
a
m
s
o
f
Or
ie
n
ted
Gr
ad
ien
ts
(
HOG)
ar
e
r
o
b
u
s
t
ag
ain
s
t
ill
u
m
in
a
tio
n
v
ar
iat
io
n
s
[
8
]
.
T
h
eir
c
o
m
p
le
m
en
tar
y
n
a
tu
r
e
m
ak
e
s
th
e
m
g
o
o
d
ca
n
d
id
ate
f
o
r
co
m
b
in
at
io
n
.
C
o
n
ce
r
n
i
n
g
t
h
e
c
lass
if
ica
tio
n
ta
s
k
s
,
w
e
u
s
e
o
n
e
d
i
m
e
n
s
io
n
al
Hid
d
en
Ma
r
k
o
v
Mo
d
el
to
d
ea
l
w
it
h
t
h
e
cla
s
s
i
f
i
ca
tio
n
tas
k
.
O
u
r
ap
p
r
o
ac
h
d
if
f
er
s
f
r
o
m
ex
i
s
ti
n
g
m
eth
o
d
s
u
s
i
n
g
HM
Ms
in
th
a
t
it
d
o
esn
’
t
n
ee
d
a
n
y
p
r
io
r
k
n
o
w
led
g
e
ab
o
u
t
t
h
e
lo
ca
lizatio
n
o
f
in
ter
est
r
e
g
io
n
s
i
n
t
h
e
f
ac
ial
i
m
a
g
es.
T
h
i
s
ad
v
an
ta
g
e
m
a
k
e
s
o
u
r
ap
p
r
o
ac
h
f
u
ll
y
au
to
m
atic
a
n
d
r
o
b
u
s
t e
v
en
t
h
e
p
r
esen
ce
o
f
n
o
f
r
o
n
tal
f
ac
ial
i
m
a
g
es.
T
h
e
f
o
llo
w
in
g
s
ec
tio
n
g
i
v
es
a
d
etailed
d
escr
ip
tio
n
o
f
th
e
p
r
o
p
o
s
ed
s
y
s
te
m
.
E
x
p
er
i
m
e
n
tal
r
esu
lt
s
ar
e
p
r
esen
ted
in
s
ec
t
io
n
3
.
Fin
al
l
y
,
w
e
co
n
cl
u
d
e
th
i
s
p
ap
er
in
s
ec
t
io
n
4
.
2.
CO
M
B
I
NING
G
AB
O
R
AND
H
O
G
F
E
AT
URE
S
F
O
R
H
I
DDEN
M
ARK
O
V
M
O
DE
L
RE
CO
G
NI
T
I
O
N
T
h
is
s
ec
tio
n
d
escr
ib
es
t
h
e
co
m
p
o
n
en
ts
o
f
o
u
r
f
ac
e
r
ec
o
g
n
i
tio
n
s
y
s
te
m
in
d
e
t
a
i
l
:
Gab
o
r
a
n
d
H
O
G
f
e
a
t
u
r
e
s
,
L
D
A
d
i
m
e
n
s
io
n
alit
y
r
ed
u
ctio
n
,
f
ea
t
u
r
e
f
u
s
io
n
u
s
in
g
t
h
e
C
an
o
n
ical
C
o
r
r
elatio
n
An
al
y
s
i
s
(
C
C
A
)
m
et
h
o
d
an
d
Hid
d
en
Ma
r
k
o
v
Mo
d
el
f
o
r
r
ec
o
g
n
itio
n
.
T
h
e
s
tag
es
of
p
r
o
ce
s
s
in
g
ar
e
d
iag
r
a
m
m
ed
in
Fi
g
u
r
e
1.
Fig
u
r
e
1
.
T
h
e
o
v
er
all
o
f
th
e
p
r
o
p
o
s
ed
f
ac
e
r
ec
o
g
n
itio
n
s
y
s
te
m
.
Featu
r
es e
x
tr
ac
tio
n
:
a.
Gab
o
r
Featu
r
es
R
ep
r
esen
tatio
n
:
Gab
o
r
w
av
elet
r
ep
r
esen
tati
o
n
o
f
f
ac
e
i
m
a
g
es
d
er
iv
e
s
d
esira
b
le
f
ea
tu
r
es
g
ain
ed
b
y
s
p
atial
f
r
eq
u
en
c
y
,
s
p
atial
lo
ca
lit
y
,
an
d
o
r
ien
tatio
n
s
elec
ti
v
it
y
.
T
h
ese
d
is
cr
i
m
i
n
ati
v
e
f
ea
t
u
r
es
ex
tr
ac
ted
f
r
o
m
th
e
Gab
o
r
f
ilt
er
ed
i
m
ag
e
s
co
u
ld
b
e
r
o
b
u
s
t
to
illu
m
i
n
atio
n
an
d
f
ac
ial
e
x
p
r
ess
io
n
c
h
a
n
g
e
s
.
A
Gab
o
r
w
a
v
elet
f
ilter
is
a
Ga
u
s
s
ian
k
er
n
el
f
u
n
ct
io
n
m
o
d
u
late
d
b
y
a
s
i
n
u
s
o
id
al
p
lan
e
w
av
e
[
5
]
:
(
)
‖
‖
‖
‖
‖
‖
[
]
(
1
)
W
h
er
e
(
)
is
th
e
p
ix
el
w
it
h
co
o
r
d
in
ate
(
x
,
y
)
in
t
h
e
i
m
a
g
e
p
lan
.
u
an
d
v
d
ef
in
e
o
r
ien
tatio
n
an
d
s
ca
le
o
f
t
h
e
Gab
o
r
k
er
n
els
.
‖
‖
d
en
o
tes th
e
n
o
r
m
o
p
er
ato
r
.
A
lar
g
e
n
u
m
b
er
o
f
lo
ca
l
f
ea
tu
r
es
ca
n
b
e
g
e
n
er
ated
b
y
v
ar
y
in
g
p
ar
a
m
eter
s
a
s
s
o
ciate
d
w
it
h
t
h
e
p
o
s
itio
n
,
s
ca
le,
a
n
d
o
r
ien
tat
io
n
o
f
th
e
f
ilter
s
.
Fo
r
e
x
a
m
p
le,
th
e
m
a
g
n
i
tu
d
e
r
esp
o
n
s
e
o
f
th
e
co
n
v
o
l
u
tio
n
o
f
a
n
i
m
a
g
e
w
it
h
4
0
b
an
k
s
o
f
Gab
o
r
k
er
n
el
s
(
8
o
r
ien
tatio
n
s
a
n
d
5
s
ca
les)
is
4
0
m
a
g
n
it
u
d
e
m
ap
s
in
t
h
e
s
a
m
e
s
ize
a
s
th
e
o
r
ig
i
n
a
l i
m
ag
e,
as i
llu
s
tr
at
ed
in
th
e
F
i
g
u
r
e
2
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
C
E
I
SS
N:
2
0
8
8
-
8708
Un
imo
d
a
l Mu
lti
-
F
ea
tu
r
e
F
u
s
i
o
n
a
n
d
On
e
-
d
ime
n
s
io
n
a
l H
id
d
en
Ma
r
ko
v
Mo
d
els
…. (
Oth
ma
n
e
E
l Meslo
u
h
i
)
1917
Fig
u
r
e
2
.
An
ex
a
m
p
le
o
f
t
h
e
G
ab
o
r
m
ag
n
it
u
d
e
o
u
tp
u
t: (
a)
th
e
in
itial i
m
a
g
e
(
b
)
th
e
m
a
g
n
itu
d
e
o
u
tp
u
t o
f
t
h
e
f
ilter
i
n
g
o
p
er
atio
n
w
it
h
a
b
an
k
o
f
4
0
Gab
o
r
f
ilter
s
.
b.
His
to
g
r
a
m
o
f
Or
ien
ted
Gr
ad
ien
ts
Feat
u
r
es
(
HOG)
:
T
h
e
HOG
d
escr
ip
to
r
is
a
lo
ca
l
s
tatis
tic
o
f
t
h
e
o
r
ien
tatio
n
s
o
f
t
h
e
i
m
a
g
e
g
r
ad
ien
ts
.
I
t
i
s
c
h
ar
ac
ter
ized
b
y
its
in
v
ar
ian
ce
to
r
o
tatio
n
an
d
ill
u
m
in
at
io
n
c
h
a
n
g
e
s
.
T
h
e
HOG
f
ea
t
u
r
e
d
iv
id
es
t
h
e
i
m
a
g
e
i
n
t
o
m
a
n
y
ce
lls
,
in
ea
c
h
o
f
t
h
e
m
a
h
i
s
to
g
r
a
m
co
u
n
t
s
th
e
o
cc
u
r
r
en
ce
s
o
f
p
ix
els
o
r
ien
ta
tio
n
s
g
i
v
e
n
b
y
t
h
eir
g
r
ad
ien
ts
.
T
h
e
f
in
a
l
HOG
d
escr
ip
to
r
is
th
en
b
u
ilt
w
i
th
c
o
m
b
i
n
atio
n
o
f
t
h
es
e
h
is
to
g
r
a
m
s
[
7
].
Fig
u
r
e
3
s
h
o
ws an
ex
a
m
p
le
o
f
f
ac
ial
i
m
a
g
e
w
it
h
e
x
tr
ac
ted
HOG
d
escr
ip
to
r
.
Fig
u
r
e
3
.
An
ex
a
m
p
le
o
f
HOG
d
escr
ip
to
r
f
r
o
m
a
f
ac
ia
l i
m
a
g
e:
(
a)
Facial
i
m
ag
e,
(
b
)
Facial
i
m
a
g
e
w
it
h
HOG
d
escr
ip
to
r
.
c.
Featu
r
e
R
ed
u
c
tio
n
:
T
h
e
d
i
m
e
n
s
io
n
s
o
f
Gab
o
r
an
d
HOG
f
ea
tu
r
e
v
ec
to
r
s
ar
e
to
o
h
ig
h
,
wh
ich
i
s
f
ar
to
o
ex
ten
s
i
v
e
f
o
r
ef
ficien
t
p
r
o
ce
s
s
in
g
an
d
s
to
r
ag
e.
T
o
o
v
er
co
m
e
th
i
s
i
s
s
u
e,
m
an
y
tec
h
n
iq
u
e
s
a
r
e
p
r
o
p
o
s
ed
in
t
h
e
liter
atu
r
e:
P
r
in
cip
al
C
o
m
p
o
n
e
n
t
A
n
al
y
s
i
s
(
P
C
A
)
[
9
]
,
I
n
d
ep
en
d
en
t
C
o
m
p
o
n
e
n
t
A
n
al
y
s
i
s
(
I
C
A
)
[
10
]
,
L
in
ea
r
D
is
cr
i
m
in
a
n
t
A
n
al
y
s
is
(
L
D
A
)
[
1
1
-
13]
.
I
n
th
is
p
ap
er
,
th
e
em
p
lo
y
ed
d
i
m
e
n
s
io
n
alit
y
r
ed
u
c
ti
o
n
tech
n
iq
u
e
is
t
h
e
L
i
n
ea
r
Dis
cr
i
m
i
n
an
t
A
n
a
l
y
s
is
(
L
D
A
)
.
I
t
ai
m
s
at
fin
d
in
g
a
f
ea
t
u
r
e
r
ep
r
esen
tatio
n
b
y
w
h
i
ch
th
e
w
it
h
in
-
cla
s
s
d
is
tan
ce
is
m
i
n
i
m
ized
an
d
th
e
b
et
w
ee
n
-
cla
s
s
d
is
ta
n
ce
is
m
a
x
i
m
ized
[
1
4
]
.
I
n
o
r
d
er
t
o
av
o
id
s
in
g
u
lar
it
y
is
s
u
e
s
,
w
h
e
n
co
m
p
u
t
in
g
t
h
e
i
n
v
er
s
e
o
f
t
h
e
w
it
h
in
-
cla
s
s
s
ca
tter
m
a
tr
ix
,
th
e
L
D
A
r
ed
u
ctio
n
m
et
h
o
d
is
i
m
p
le
m
e
n
ted
i
n
th
e
P
r
in
cip
al
C
o
m
p
o
n
e
n
t
An
al
y
s
i
s
(
P
C
A
)
s
u
b
s
p
ac
e
as s
u
g
g
e
s
ted
in
[
1
5
].
d.
Featu
r
e
f
u
s
io
n
u
s
in
g
ca
n
o
n
ica
l
co
r
r
elatio
n
an
al
y
s
i
s
:
I
n
th
i
s
s
tag
e,
w
e
co
m
b
in
e
t
h
e
t
w
o
r
e
d
u
ce
d
f
ea
tu
r
e
v
ec
to
r
s
to
o
b
tain
a
s
in
g
le
o
n
e
,
w
h
ic
h
is
m
o
r
e
d
is
cr
i
m
in
at
iv
e
th
an
u
s
in
g
o
n
l
y
o
n
e
f
ea
t
u
r
e
m
o
d
alit
y
.
T
h
is
is
ac
h
iev
ed
b
y
u
s
in
g
a
f
ea
t
u
r
e
f
u
s
io
n
tec
h
n
iq
u
e
b
ased
o
n
C
an
o
n
ical
C
o
r
r
elatio
n
A
n
a
l
y
s
is
(
C
C
A
)
[
16
].
C
an
o
n
ical
co
r
r
elatio
n
an
al
y
s
is
h
as b
ee
n
w
id
el
y
u
s
ed
to
an
al
y
ze
a
s
s
o
cia
tio
n
s
b
et
w
ee
n
t
w
o
s
ets o
f
v
ar
i
ab
les.
Giv
e
n
t
w
o
co
l
u
m
n
v
ec
to
r
s
(
)
an
d
(
)
o
f
r
an
d
o
m
v
ar
iab
les
w
it
h
f
i
n
ite
s
ec
o
n
d
m
o
m
e
n
t
s
,
o
n
e
m
a
y
d
ef
in
e
t
h
e
cr
o
s
s
-
co
v
ar
ian
ce
(
)
to
b
e
th
e
m
atr
i
x
w
h
o
s
e
(
)
en
tr
y
is
th
e
co
v
ar
ia
n
ce
(
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
J
E
C
E
Vo
l.
7
,
No
.
4
,
A
u
g
u
s
t
2
0
1
7
:
1
9
1
5
–
1
9
2
2
1918
C
an
o
n
ical
-
co
r
r
elatio
n
an
al
y
s
i
s
atte
m
p
t
to
f
in
d
v
ec
to
r
s
an
d
s
u
c
h
th
at
t
h
e
r
an
d
o
m
v
ar
iab
les
U
an
d
m
ax
i
m
ize
th
e
co
r
r
elatio
n
ρ
=
(
,
)
=
(
)
an
d
ar
e
d
ef
in
ed
b
y
[
8
]
:
(
2
)
(
3
)
W
h
er
e,
(
)
,
(
)
,
an
d
Af
ter
s
o
m
e
s
tep
s
o
f
ca
lc
u
latio
n
,
th
e
s
o
l
u
tio
n
is
t
h
er
ef
o
r
e:
1)
is
th
e
ei
g
en
v
ec
to
r
w
it
h
th
e
m
a
x
i
m
u
m
ei
g
en
v
al
u
e
f
o
r
th
e
m
at
r
ix
.
2)
is
th
e
ei
g
en
v
ec
to
r
w
it
h
th
e
m
a
x
i
m
u
m
ei
g
en
v
al
u
e
f
o
r
th
e
m
at
r
ix
.
e.
Featu
r
e
clas
s
i
fi
ca
t
io
n
:
An
H
MM
,
w
h
ic
h
w
as
d
e
v
elo
p
ed
b
y
B
a
u
m
a
n
d
P
etr
ie
[
1
7
]
,
is
a
p
r
o
b
a
b
ilis
tic
m
o
d
el
i
n
w
h
ic
h
t
h
e
s
y
s
te
m
is
ass
u
m
ed
to
b
e
a
Ma
r
k
o
v
p
r
o
ce
s
s
w
it
h
h
id
d
en
s
tates.
HM
M
is
u
s
ed
f
o
r
r
ep
r
esen
tin
g
p
r
o
b
ab
ilit
y
d
is
tr
i
b
u
tio
n
o
v
er
s
eq
u
en
ce
s
o
f
o
b
s
e
r
v
atio
n
s
.
T
h
e
f
o
l
lo
w
i
n
g
k
e
y
el
e
m
en
ts
c
h
ar
ac
ter
ize
an
HM
M:
a.
th
e
n
u
m
b
er
o
f
s
tates o
f
th
e
m
o
d
el.
b.
is
an
s
tate
tr
an
s
it
io
n
m
atr
ix
t
h
at
g
i
v
e
s
th
e
s
tate
tr
an
s
itio
n
p
r
o
b
ab
ilit
ies b
et
w
ee
n
s
tates.
c.
is
an
em
is
s
io
n
p
r
o
b
ab
ilit
y
m
a
tr
ix
w
h
ile
b
ein
g
in
a
p
ar
ticu
lar
s
tate
.
d.
is
a
m
a
tr
ix
,
ca
lled
in
itial
s
tat
e
p
r
o
b
ab
ilit
y
m
atr
ix
,
an
d
it
g
i
v
es
th
e
p
r
o
b
ab
ilit
y
o
f
b
ein
g
in
a
p
ar
ticu
lar
s
tate
at
t
h
e
s
tar
t
o
f
th
e
p
r
o
ce
s
s
.
As
m
e
n
tio
n
ed
ab
o
v
e,
a
lef
t
to
r
ig
h
t
1
D
Hid
d
en
Ma
r
k
o
v
Mo
d
el
is
u
s
ed
i
n
o
u
r
s
y
s
te
m
to
a
ch
iev
e
th
e
r
ec
o
g
n
itio
n
tas
k
.
T
h
e
F
ig
u
r
e
4
s
h
o
w
s
a
n
ex
a
m
p
le
o
f
a
L
ef
t
-
R
ig
h
t 1
D
-
HM
M
w
ith
N
s
tates.
Fig
u
r
e
4
.
An
ex
a
m
p
le
o
f
a
L
ef
t
-
R
ig
h
t 1
D
-
HM
M
w
it
h
N
s
ta
te
s
.
I
n
o
u
r
s
y
s
te
m
,
th
e
n
u
m
b
er
o
f
s
tates
N
d
ep
en
d
s
to
th
e
d
ataset
s
u
b
j
ec
t
n
u
m
b
er
.
T
h
e
en
tr
y
s
t
ate
1
an
d
th
e
ex
it
s
ta
te
N
ar
e
n
o
n
-
e
m
it
tin
g
.
T
h
e
tr
an
s
itio
n
m
atr
i
x
w
il
l
h
a
v
e
r
o
w
s
an
d
co
lu
m
n
s
.
Fi
n
all
y
w
e
ass
u
m
e
th
a
t
ea
c
h
o
b
s
er
v
atio
n
p
r
o
b
ab
ilit
y
d
i
s
tr
ib
u
tio
n
i
s
r
e
p
r
esen
ted
w
it
h
s
in
g
le
Gau
s
s
ia
n
d
is
tr
ib
u
tio
n
s
.
A
ll
th
ese
p
ar
a
m
eter
s
w
i
ll
b
e
esti
m
ated
d
u
r
in
g
th
e
tr
ai
n
i
n
g
s
te
p
u
s
in
g
an
iter
ati
v
e
p
r
o
ce
d
u
r
e,
k
n
o
w
n
as
B
au
m
-
W
elch
al
g
o
r
ith
m
[
1
8
]
.
A
ls
o
,
e
ac
h
s
u
b
j
ec
t
w
ill
b
e
r
ep
r
esen
te
d
w
i
th
o
n
e
m
o
d
el.
Fin
a
ll
y
,
th
e
r
ec
o
g
n
i
tio
n
s
tep
i
s
ac
h
iev
ed
u
s
in
g
t
h
e
V
iter
b
i
alg
o
r
ith
m
[
1
8
]
to
fin
d
th
e
h
i
g
h
es
t
lik
el
ih
o
o
d
s
co
r
e
b
et
w
ee
n
f
ea
tu
r
es
i
n
t
h
e
te
s
ti
n
g
s
ets a
n
d
m
o
d
els o
b
tain
ed
in
t
h
e
tr
ain
in
g
s
tep
.
3.
E
XP
E
R
I
M
E
NT
A
L
RE
SUL
T
S
T
h
e
ex
p
er
im
e
n
t
w
as
d
esi
g
n
ed
w
it
h
th
e
ai
m
o
f
v
alid
ati
n
g
t
h
e
ef
f
ec
ti
v
e
n
es
s
o
f
th
e
p
r
o
p
o
s
ed
s
y
s
te
m
.
W
e
p
er
f
o
r
m
ed
ex
p
er
i
m
e
n
ts
u
s
in
g
th
e
cr
o
p
p
ed
A
R
f
ac
e
d
atab
ase
[
1
9
]
.
T
h
e
A
R
f
ac
e
d
atab
ase
co
n
tain
s
o
v
e
r
3
,
2
0
0
f
r
o
n
tal
co
lo
r
f
ac
e
i
m
a
g
es
o
f
1
2
6
s
u
b
j
ec
ts
(
2
6
d
if
f
er
en
t
i
m
a
g
es
f
o
r
ea
ch
p
er
s
o
n
)
,
in
cl
u
d
in
g
d
i
f
f
er
en
t
f
ac
ial
ex
p
r
e
s
s
io
n
s
,
w
it
h
v
ar
io
u
s
o
cc
lu
s
io
n
s
a
n
d
u
n
d
e
r
d
if
f
er
en
t
lig
h
ti
n
g
co
n
d
itio
n
s
.
Mo
s
t
o
f
th
e
p
ictu
r
es
w
er
e
r
ec
o
r
d
e
d
in
t
w
o
s
e
s
s
io
n
s
(
s
ep
a
r
ated
b
y
t
w
o
w
ee
k
s
)
.
A
ll
i
m
a
g
es
w
er
e
tak
e
n
b
y
t
h
e
s
a
m
e
ca
m
er
a
u
n
d
er
tig
h
tl
y
co
n
tr
o
lled
co
n
d
itio
n
s
o
f
ill
u
m
i
n
atio
n
an
d
v
ie
w
p
o
i
n
t.
Fo
r
o
u
r
ex
p
er
i
m
e
n
ts
,
li
k
e
i
n
th
e
w
o
r
k
o
f
[
1
9
]
,
in
th
e
fi
r
s
t
ti
m
e,
1
0
0
d
if
f
er
en
t
s
u
b
j
ec
ts
(
5
0
m
ale
s
an
d
5
0
f
e
m
ale
s
)
w
e
r
e
r
an
d
o
m
l
y
s
elec
ted
f
r
o
m
t
h
i
s
d
atab
ase,
th
en
,
all
s
elec
ted
i
m
a
g
es
w
er
e
s
e
g
m
e
n
t
ed
u
s
in
g
a
n
o
v
al
-
s
h
ap
ed
m
as
k
an
d
fin
all
y
all
co
lo
r
i
m
a
g
es
a
r
e
tr
an
s
f
o
r
m
ed
in
to
g
r
a
y
i
m
a
g
es.
Fi
g
u
r
e
5
s
h
o
w
s
s
o
m
e
s
a
m
p
le
i
m
a
g
es
o
f
o
n
e
s
u
b
j
ec
t
ex
tr
ac
ted
f
r
o
m
t
h
e
o
b
tain
ed
f
ac
e
d
atab
ase.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
C
E
I
SS
N:
2
0
8
8
-
8708
Un
imo
d
a
l Mu
lti
-
F
ea
tu
r
e
F
u
s
i
o
n
a
n
d
On
e
-
d
ime
n
s
io
n
a
l H
id
d
en
Ma
r
ko
v
Mo
d
els
…. (
Oth
ma
n
e
E
l Meslo
u
h
i
)
1919
Fig
u
r
e
5
.
I
m
a
g
es o
f
o
n
e
s
u
b
j
ec
t in
th
e
AR
f
ac
e
d
atab
ase.
I
n
t
h
is
ex
p
er
i
m
en
ts
,
w
e
i
n
v
e
s
tig
a
te
t
h
e
f
ac
e
r
ec
o
g
n
i
tio
n
as
a
f
u
n
c
tio
n
o
f
th
e
i
m
ag
e
r
eso
lu
tio
n
.
I
n
s
tead
o
f
tak
in
g
i
m
ag
e
s
f
r
o
m
f
ac
es
w
h
ich
h
a
v
e
d
if
f
er
en
t
d
i
s
tan
ce
s
to
th
e
ca
m
er
a,
w
e
h
av
e
tr
ied
to
s
im
u
late
th
e
ef
f
ec
t
o
f
lo
w
er
in
g
t
h
e
r
es
o
lu
tio
n
.
Fro
m
o
r
i
g
in
a
l
p
ictu
r
e
s
w
it
h
th
e
r
eso
l
u
tio
n
o
f
p
ix
els
,
all
o
f
p
ictu
r
es
h
a
v
e
b
ee
n
d
o
w
n
s
a
m
p
led
2
,
4
,
8
,
1
6
an
d
ev
en
3
2
tim
e
s
.
I
t
co
r
r
esp
o
n
d
s
to
r
e
s
o
lu
tio
n
s
,
,
,
,
an
d
r
esp
ec
tiv
el
y
.
Si
n
c
e
ex
tr
ac
tin
g
f
ea
t
u
r
es
f
r
o
m
f
ac
ial
w
it
h
lo
w
est
r
eso
lu
tio
n
w
o
u
ld
lead
to
s
er
io
u
s
q
u
an
t
izatio
n
er
r
o
r
s
,
w
e
d
ec
id
ed
to
u
p
s
ca
le
all
i
m
a
g
es
to
t
h
e
s
a
m
e
r
eso
l
u
tio
n
.
Af
ter
d
o
w
n
s
ca
lin
g
,
w
e
s
ca
led
th
e
i
m
ag
e
u
p
to
o
r
ig
in
a
l
s
ize
(
i.e
)
f
o
r
f
ac
e
r
ec
o
g
n
it
io
n
u
s
i
n
g
a
cu
b
ic
in
ter
p
o
latio
n
[
20
]
.
T
h
e
r
esu
lts
ca
n
b
e
s
ee
n
i
n
Fi
g
u
r
e
6.
Fig
u
r
e
6
.
Sa
m
p
les
f
o
r
d
o
w
n
s
a
m
p
led
f
ac
e
i
m
ag
e:
(
a)
th
e
o
r
i
g
in
al
f
ac
e
i
m
a
g
e
.
(
b
)
d
o
w
n
s
a
m
p
lin
g
t
h
e
i
m
a
g
e
in
a
b
y
a
f
ac
to
r
o
f
1
/2
to
th
e
s
ize
. (
c)
b
y
f
ac
to
r
o
f
1
/4
to
th
e
s
ize
.
(
d
)
b
y
f
ac
to
r
o
f
1
/
8
to
th
e
s
ize
.
(
e)
b
y
f
ac
to
r
o
f
1
/1
6
to
th
e
s
ize
.
(
f
)
b
y
f
ac
to
r
of
1
/3
2
to
th
e
s
ize
.
T
h
e
ex
p
er
i
m
en
tal
r
es
u
lt
s
p
r
ese
n
ted
i
n
t
h
is
s
ec
tio
n
ar
e
d
iv
id
ed
in
to
t
h
r
ee
p
ar
ts
.
T
h
e
ai
m
o
f
th
e
fi
r
s
t
s
et
of
ex
p
er
i
m
e
n
ts
i
s
to
s
ee
w
h
et
h
er
th
e
s
y
s
te
m
co
n
v
er
g
e
s
q
u
i
ck
l
y
o
r
n
o
t.
R
ec
o
g
n
itio
n
r
ates
g
i
v
en
at
d
if
f
er
e
n
t
ep
o
ch
s
ar
e
p
lo
tt
ed
in
F
g
u
r
e
7
s
h
o
w
in
g
t
h
at
th
e
s
y
s
te
m
co
n
v
er
g
es a
t t
h
e
en
d
o
f
t
h
e
s
ec
o
n
d
ep
o
ch
.
I
n
s
ec
o
n
d
e
x
p
er
i
m
e
n
t,
w
e
m
a
k
e
a
co
m
p
ar
is
o
n
b
et
w
ee
n
t
h
e
co
m
b
in
ed
Gab
o
r
+H
OG
d
esc
r
ip
to
r
w
ith
is
o
lated
d
escr
ip
to
r
s
Gab
o
r
,
H
OG
an
d
L
B
P
(
L
o
ca
l
B
in
ar
y
P
atter
n
s
)
f
o
r
d
if
f
er
e
n
t
r
eso
l
u
tio
n
.
W
e
n
o
tice
th
a
t
th
e
f
ea
t
u
r
es
s
ize
o
f
all
t
h
o
s
e
d
es
cr
ip
to
r
s
is
r
ed
u
ce
d
u
s
i
n
g
L
D
A
m
et
h
o
d
.
T
h
e
co
m
p
ar
is
o
n
r
esu
lt
s
ar
e
s
h
o
w
n
i
n
T
ab
le
1
.
Fro
m
th
i
s
tab
le,
w
e
n
o
tice
th
at
th
e
r
es
u
lts
o
b
tain
ed
b
y
t
h
e
co
m
b
i
n
ed
Gab
o
r
+H
OG
f
ea
tu
r
e
s
o
u
tp
er
f
o
r
m
t
h
o
s
e
o
b
tain
e
d
u
s
in
g
s
ep
ar
ated
Gab
o
r
,
HOG
a
n
d
L
B
P
d
escr
ip
to
r
s
in
AR
d
atab
ase
f
o
r
all
f
ac
ia
l
i
m
a
g
e
r
eso
lu
tio
n
.
Fin
all
y
,
a
s
s
h
o
w
n
in
T
ab
le
2,
T
ab
le
3,
T
ab
le
4
an
d
T
a
b
le
5
,
w
e
ca
lc
u
lated
r
ec
o
g
n
itio
n
r
ate
ac
co
r
d
in
g
to
th
e
s
ize
o
f
tr
ain
in
g
i
m
a
g
e
s
et
an
d
th
e
r
eso
lu
tio
n
o
b
tai
n
ed
b
y
t
h
e
1
D
-
HM
M
clas
s
i
fier
an
d
o
th
er
th
r
ee
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
J
E
C
E
Vo
l.
7
,
No
.
4
,
A
u
g
u
s
t
2
0
1
7
:
1
9
1
5
–
1
9
2
2
1920
class
i
fi
er
s
:
S
u
p
p
o
r
t
Vec
to
r
Ma
ch
i
n
e
(
SVM)
w
ith
R
ad
ia
l
B
asic
Fu
n
ctio
n
(
R
DF)
Ker
n
el,
A
d
aB
o
o
s
t
an
d
Dec
is
io
n
T
r
ee
C
lass
i
fier
s
.
Fr
o
m
t
h
ese
tab
les,
w
e
ca
n
s
ee
th
at
th
e
ac
cu
r
ac
ie
s
o
b
tain
ed
b
y
t
h
e
1
D
-
HM
M
class
i
fi
er
,
i
n
g
e
n
e
r
al,
o
u
tp
er
f
o
r
m
th
e
SVM,
A
d
aB
o
o
s
t a
n
d
Dec
is
io
n
T
r
ee
C
lass
i
fi
er
s
.
Fig
u
r
e
7
.
R
ec
o
g
n
i
tio
n
R
ate
v
s
.
T
h
e
n
u
m
b
er
o
f
tr
ai
n
in
g
ep
o
ch
s
.
T
ab
le
1
.
B
est f
ac
e
r
ec
o
g
n
it
io
n
r
ates u
s
i
n
g
d
if
f
er
en
t d
escr
ip
to
r
s
w
ith
d
i
f
f
er
en
t i
m
a
g
e
r
eso
lu
ti
o
n
D
e
scri
p
t
o
r
s
Ex
p
e
r
i
me
n
t
a
l
D
a
t
a
se
t
s
L
B
P
1
2
.
0
3
%
5
0
%
7
7
.
2
6
%
9
2
.
2
3
%
9
1
.
3
5
%
9
5
.
4
3
%
HOG
2
2
.
4
5
%
5
5
.
1
4
%
8
8
.
5
%
9
8
.
8
9
%
9
9
.
8
1
%
9
9
.
8
1
%
G
a
b
o
r
1
3
.
3
6
%
5
3
.
2
3
%
5
9
.
3
7
%
9
4
.
9
5
%
9
9
.
4
4
%
1
0
0
%
G
a
b
o
r
+
H
O
G
2
7
.
6
4
%
6
5
.
4
9
%
8
9
.
0
5
%
9
9
.
6
3
%
1
0
0
%
1
0
0
%
T
ab
le
2
.
Face
r
ec
o
g
n
itio
n
r
ates o
f
f
o
u
r
class
i
f
ier
s
w
i
th
d
i
f
f
er
en
t I
m
a
g
e
r
es
o
lu
tio
n
u
s
i
n
g
5
tr
ain
i
n
g
i
m
a
g
es
C
l
a
ssi
f
i
e
r
s
Ex
p
e
r
i
me
n
t
a
l
D
a
t
a
se
t
s
S
V
M
0
.
0
7
%
1
1
%
4
5
.
4
8
%
7
7
.
4
5
%
8
4
.
8
3
%
9
2
.
0
3
%
A
d
a
B
o
o
st
0
.
0
2
4
%
3
.
3
%
3
.
8
%
3
.
5
%
2
.
9
%
3
.
5
%
D
e
c
i
si
o
n
T
r
e
e
0
.
0
5
%
7
.
2
%
1
1
%
1
2
.
9
2
%
1
5
.
9
%
1
0
.
2
%
1D
-
H
M
M
1
2
.
0
5
%
1
9
.
5
3
%
4
5
.
8
7
%
8
5
.
0
3
%
9
0
.
2
8
%
9
3
.
4
9
%
T
ab
le
3
.
Face
r
ec
o
g
n
itio
n
r
ates o
f
f
o
u
r
class
i
f
ier
s
w
i
th
d
i
f
f
er
en
t I
m
a
g
e
r
eso
lu
tio
n
u
s
i
n
g
10
tr
ain
i
n
g
i
m
a
g
es.
C
l
a
ssi
f
i
e
r
s
Ex
p
e
r
i
me
n
t
a
l
D
a
t
a
se
t
s
S
V
M
0
.
1
3
%
2
0
.
8
%
5
2
.
4
2
%
9
0
.
3
%
9
7
.
1
9
%
9
8
.
7
2
%
A
d
a
B
o
o
st
0
.
0
3
%
3
.
4
%
6
.
1
%
5
.
2
2
%
9
.
1
8
%
4
.
6
%
D
e
c
i
si
o
n
T
r
e
e
0
.
0
7
%
1
0
.
5
8
%
9
.
7
%
1
6
.
4
5
%
1
1
.
2
%
1
6
.
8
3
%
1D
-
H
M
M
1
7
.
8
6
%
3
2
%
6
2
.
3
7
%
9
5
.
0
3
%
9
8
.
0
1
%
9
8
.
8
5
%
T
ab
le
4
.
Face
r
ec
o
g
n
itio
n
r
ates o
f
f
o
u
r
class
i
f
ier
s
w
i
th
d
i
f
f
er
en
t I
m
a
g
e
r
eso
lu
tio
n
u
s
i
n
g
1
5
tr
ain
i
n
g
i
m
a
g
es
C
l
a
ssi
f
i
e
r
s
Ex
p
e
r
i
me
n
t
a
l
D
a
t
a
se
t
s
S
V
M
1
8
.
9
%
3
7
.
1
%
7
3
.
8
4
%
9
9
.
4
4
%
1
0
0
%
1
0
0
%
A
d
a
B
o
o
st
5
.
1
9
%
5
.
7
%
3
.
3
3
%
7
.
2
%
6
.
6
7
%
5
.
9
%
D
e
c
i
si
o
n
T
r
e
e
5
.
5
%
1
1
.
1
3
%
1
7
.
6
%
1
4
.
6
5
%
2
0
%
2
3
%
1D
-
H
M
M
2
0
.
5
9
%
4
6
.
7
5
%
8
0
.
7
1
%
9
9
.
6
3
%
1
0
0
%
1
0
0
%
T
ab
le
5
.
Face
r
ec
o
g
n
itio
n
r
ates o
f
f
o
u
r
class
i
f
ier
s
w
i
th
d
i
f
f
er
en
t I
m
a
g
e
r
eso
lu
tio
n
u
s
i
n
g
20
tr
ain
i
n
g
i
m
a
g
es
C
l
a
ssi
f
i
e
r
s
Ex
p
e
r
i
me
n
t
a
l
D
a
t
a
se
t
s
S
V
M
1
0
%
3
1
.
2
9
%
7
5
.
8
5
%
1
0
0
%
1
0
0
%
1
0
0
%
A
d
a
B
o
o
st
2%
5
.
4
%
6
.
4
6
%
5
.
4
%
8
.
1
%
5
.
1
%
D
e
c
i
si
o
n
T
r
e
e
6
.
1
2
%
9
.
1
8
%
1
8
.
0
2
%
1
9
.
3
8
%
1
8
.
3
6
%
1
7
.
6
%
1D
-
H
M
M
1
6
%
4
8
.
6
4
%
9
0
.
4
8
%
1
0
0
%
1
0
0
%
1
0
0
%
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
C
E
I
SS
N:
2
0
8
8
-
8708
Un
imo
d
a
l Mu
lti
-
F
ea
tu
r
e
F
u
s
i
o
n
a
n
d
On
e
-
d
ime
n
s
io
n
a
l H
id
d
en
Ma
r
ko
v
Mo
d
els
…. (
Oth
ma
n
e
E
l Meslo
u
h
i
)
1921
4.
CO
NCLU
SI
O
N
A
lo
w
f
ac
e
r
ec
o
g
n
itio
n
s
y
s
te
m
b
ased
o
n
1
D
-
HM
M
s
w
as
p
r
esen
ted
in
t
h
i
s
p
ap
er
.
First,
th
e
s
y
s
te
m
ex
tr
ac
t
f
ac
ial
f
ea
tu
r
es
u
s
i
n
g
G
ab
o
r
an
d
HOG
d
e
s
cr
ip
to
r
s
.
T
h
en
it
co
m
b
i
n
es
t
h
e
m
u
s
i
n
g
C
C
A
f
u
s
io
n
tech
n
iq
u
e
to
o
b
tain
o
n
e
f
ea
t
u
r
e
v
ec
to
r
af
ter
d
im
e
n
s
io
n
alit
y
r
ed
u
ctio
n
s
t
ep
u
s
in
g
L
D
A
m
et
h
o
d
.
T
h
e
s
tan
d
ar
d
d
atab
ase
AR
i
s
u
s
ed
to
e
v
al
u
ate
th
e
p
r
o
p
o
s
ed
s
y
s
te
m
.
T
h
e
o
b
tain
ed
r
ec
o
g
n
itio
n
r
ate
s
b
y
t
h
e
co
m
b
in
ed
f
ea
t
u
r
es
an
d
1
D
-
HM
Ms
class
i
fier
o
u
tp
er
f
o
r
m
th
e
s
e
o
b
tain
ed
b
y
is
o
lat
ed
d
escr
ip
to
r
s
an
d
s
o
m
e
clas
s
ical
clas
s
i
fi
er
s
f
o
r
all
f
ac
ial
i
m
a
g
e
r
eso
lu
tio
n
s
.
Ou
r
f
u
t
u
r
e
r
esear
c
h
w
il
l
b
e
f
o
cu
s
ed
o
n
u
s
i
n
g
t
h
e
p
r
o
p
o
s
ed
s
y
s
te
m
i
n
r
ea
l
ti
m
e
co
n
te
x
t
i
n
o
r
d
er
to
in
te
g
r
ate
it in
v
id
eo
s
u
r
v
eilla
n
ce
ca
m
er
as sec
u
r
it
y
ap
p
licatio
n
.
RE
F
E
R
E
NC
E
S
[1
]
Z.
W
a
n
g
,
e
t
a
l.
,
“
L
o
w
-
Re
so
lu
ti
o
n
fa
c
e
Re
c
o
g
n
it
io
n
:
A
R
e
v
ie
w
,
”
T
h
e
Vi
su
a
l
Co
mp
u
ter
,
v
o
l.
3
0
,
n
o
.
4
,
p
p
.
3
5
9
–
3
8
6
,
2
0
1
4
.
[2
]
J.
Y.
Ch
o
i,
e
t
a
l.
,
“
Co
l
o
r
f
a
c
e
Re
c
o
g
n
it
io
n
f
o
r
De
g
ra
d
e
d
f
a
c
e
I
m
a
g
e
s,”
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
S
y
st
e
ms
,
M
a
n
,
a
n
d
Cy
b
e
rn
e
ti
c
s,
Pa
rt B
(
Cy
b
e
rn
e
ti
c
s)
,
v
o
l.
3
9
,
n
o
.
5
,
p
p
.
1
2
1
7
–
1
2
3
0
,
Oc
t
2
0
0
9
.
[3
]
T
.
A
h
o
n
e
n
,
e
t
a
l.
,
“
Rec
o
g
n
it
io
n
o
f
Bl
u
rr
e
d
fa
c
e
s u
sin
g
lo
c
a
l
p
h
a
se
Q
u
a
n
t
iza
ti
o
n
,
”
P
a
tt
e
rn
Re
c
o
g
n
it
i
o
n
,
2
0
0
8
.
IC
P
R
2
0
0
8
.
1
9
th
I
n
tern
a
ti
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
,
p
p
.
1
–
4
,
2
0
0
8
.
[4
]
S.
-
W
.
L
e
e
e
t
a
l.
,
“
L
o
w
Re
so
l
u
ti
o
n
f
a
c
e
R
e
c
o
g
n
it
io
n
b
a
se
d
o
n
S
u
p
p
o
rt
V
e
c
to
r
Da
ta
D
e
sc
r
ip
ti
o
n
,
”
Pa
tt
e
r
n
Rec
o
g
n
it
io
n
,
v
o
l.
3
9
,
n
o
.
9
,
p
p
.
1
8
0
9
–
1
8
1
2
,
2
0
0
6
.
[5
]
C.
L
iu
,
e
t
a
l.
,
“
G
a
b
o
r
F
e
a
tu
re
b
a
se
d
Clas
sifi
c
a
ti
o
n
u
sin
g
th
e
En
h
a
n
c
e
d
F
sh
e
r
L
in
e
a
r
Dis
c
ri
m
in
a
n
t
M
o
d
e
l
f
o
r
f
a
c
e
R
e
c
o
g
n
it
io
n
,
”
T
ra
n
s.
Img
.
Pro
c
.
,
v
o
l.
1
1
,
n
o
.
4
,
p
p
.
4
6
7
–
4
7
6
,
A
p
r.
2
0
0
2
.
[6
]
H.
Ku
su
m
a
,
e
t
a
l.
“
G
a
b
o
r
-
b
a
se
d
f
a
c
e
Re
c
o
g
n
it
io
n
w
it
h
Ill
u
m
in
a
ti
o
n
V
a
riatio
n
u
sin
g
S
u
b
s
p
a
c
e
-
li
n
e
a
r
Disc
ri
m
in
a
n
t
A
n
a
l
y
si
s”
,
T
EL
KOM
NIKA
(
T
e
lec
o
mm
u
n
ic
a
ti
o
n
C
o
mp
u
t
in
g
El
e
c
tr
o
n
ics
a
n
d
Co
n
tro
l)
,
v
o
l.
10
,
n
o
.
1
,
p
p
1
1
9
-
1
2
8
,
2
0
1
2
.
[7
]
C.
S
h
u
,
e
t
a
l.
,
“
Histo
g
ra
m
o
f
t
h
e
Orie
n
ted
G
ra
d
ien
t
f
o
r
fa
c
e
R
e
c
o
g
n
it
io
n
,
”
T
sin
g
h
u
a
S
c
ie
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
,
v
o
l.
1
6
,
n
o
.
2
,
p
p
.
2
1
6
–
2
2
4
,
A
p
ril
2
0
1
1
.
[8
]
M
.
Ha
g
h
ig
h
a
t,
e
t
a
l.
,
“
F
u
ll
y
A
u
to
m
a
ti
c
fa
c
e
No
r
m
a
li
z
a
ti
o
n
a
n
d
si
n
g
le
S
a
m
p
le
fa
c
e
R
e
c
o
g
n
it
io
n
i
n
Un
c
o
n
stra
in
e
d
E
n
v
iro
n
m
e
n
ts,
”
Exp
e
rt S
y
st
e
ms
wit
h
A
p
p
li
c
a
ti
o
n
s
,
v
o
l
.
4
7
,
n
o
.
C,
p
p
.
2
3
–
3
4
,
A
p
r.
2
0
1
6
.
[9
]
K.
Kim
,
“
Fa
c
e
Rec
o
g
n
it
i
o
n
u
sin
g
Prin
c
i
p
le
C
o
mp
o
n
e
n
t
A
n
a
lys
is,
”
In
tern
a
ti
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
ter
V
isi
o
n
a
n
d
P
a
tt
e
rn
Re
c
o
g
n
it
i
o
n
,
1
9
9
6
,
p
p
.
5
8
6
–
5
9
1
.
[1
0
]
M
.
S
.
Ba
rtl
e
tt
,
e
t
a
l.
,
“
F
a
c
e
Re
c
o
g
n
it
io
n
b
y
In
d
e
p
e
n
d
e
n
t
C
o
m
p
o
n
e
n
t
A
n
a
ly
sis,”
Ne
u
ra
l
Ne
two
rk
s,
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
,
v
o
l.
1
3
,
n
o
.
6
,
p
p
.
1
4
5
0
–
1
4
6
4
,
2
0
0
2
.
[1
1
]
D.
L
.
S
we
ts,
e
t
a
l.
,
“
Us
in
g
D
isc
rim
in
a
n
t
Ei
g
e
n
f
e
a
tu
re
s
f
o
r
I
m
a
g
e
R
e
tri
e
v
a
l,
”
IEE
E
T
ra
n
s.
Pa
t
ter
n
An
a
l.
M
a
c
h
.
In
tell.
,
v
o
l.
1
8
,
n
o
.
8
,
p
p
.
8
3
1
–
8
3
6
,
A
u
g
.
1
9
9
6
.
[1
2
]
P
.
N.
Be
lh
u
m
e
u
r,
e
t
a
l.
,
“
Ei
g
e
n
f
a
c
e
s
v
s.
F
i
sh
e
rf
a
c
e
s:
Re
c
o
g
n
it
io
n
u
sin
g
c
l
a
ss
sp
e
c
ifi
c
li
n
e
a
r
P
r
o
jec
ti
o
n
,
”
IEE
E
T
ra
n
s.
Pa
tt
e
rn
An
a
l.
M
a
c
h
.
I
n
tell.
,
v
o
l.
1
9
,
n
o
.
7
,
p
p
.
7
1
1
–
7
2
0
,
Ju
l.
1
9
9
7
.
[1
3
]
I.
W
ij
a
y
a
,
e
t
a
l.
“
F
a
c
e
Re
c
o
g
n
it
io
n
Us
in
g
Ho
l
isti
c
F
e
a
tu
re
s
a
n
d
L
DA
S
im
p
li
f
ica
ti
o
n
”
,
T
EL
KOM
NIKA
(
T
e
lec
o
mm
u
n
ica
ti
o
n
Co
mp
u
ti
n
g
El
e
c
tro
n
ics
a
n
d
C
o
n
tro
l)
,
v
o
l.
1
0
,
n
o
4
,
p
.
7
7
1
-
7
8
2
,
2
0
1
2
.
[1
4
]
T
.
S
a
v
ič
,
e
t
a
l.
,
“
P
e
rso
n
a
l
Re
c
o
g
n
it
io
n
b
a
se
d
o
n
a
n
Im
a
g
e
o
f
th
e
P
a
lm
a
r
S
u
rf
a
c
e
o
f
th
e
h
a
n
d
”
,
Pa
tt
e
r
n
Rec
o
g
n
it
io
n
,
p
p
.
3
1
5
2
–
3
1
6
3
,
2
0
0
7
.
[1
5
]
P
.
N.
Be
lh
u
m
e
u
r,
e
t
a
l.
,
“
Ei
g
e
n
f
a
c
e
s
v
s.
F
i
sh
e
rf
a
c
e
s:
Re
c
o
g
n
it
io
n
u
sin
g
c
las
s
sp
e
c
ifi
c
li
n
e
a
r
P
r
o
jec
ti
o
n
,
”
IEE
E
T
ra
n
s.
Pa
tt
e
rn
An
a
l.
M
a
c
h
.
I
n
tell.
,
v
o
l.
1
9
,
n
o
.
7
,
p
p
.
7
1
1
–
7
2
0
,
Ju
l.
1
9
9
7
.
[1
6
]
Q.
-
S
.
S
u
n
,
e
t
a
l.
,
“
A
n
e
w
M
e
th
o
d
o
f
F
e
a
tu
re
F
u
si
o
n
a
n
d
it
s
A
p
p
li
c
a
ti
o
n
in
Im
a
g
e
R
e
c
o
g
n
it
io
n
,
”
Pa
tt
e
rn
Rec
o
g
n
it
io
n
.
,
v
o
l
.
3
8
,
n
o
.
1
2
,
p
p
.
2
4
3
7
–
2
4
4
8
,
2
0
0
5
.
[1
7
]
L
.
E.
Ba
u
m
,
e
t
a
l.
,
“
S
tatisti
c
a
l
In
f
e
re
n
c
e
f
o
r
P
r
o
b
a
b
il
isti
c
F
u
n
c
ti
o
n
s
o
f
F
in
i
te
S
tate
M
a
rk
o
v
C
h
a
in
s,”
An
n
a
ls
o
f
M
a
th
e
ma
ti
c
a
l
S
t
a
ti
stics
,
v
o
l
.
3
7
,
p
p
.
1
5
5
4
–
1
5
6
3
,
1
9
6
6
.
[1
8
]
L
.
Ra
b
in
e
r,
“
A
tu
to
rial
o
n
Hid
d
e
n
M
a
rk
o
v
M
o
d
e
ls
a
n
d
S
e
lec
ted
A
p
p
li
c
a
ti
o
n
s
in
sp
e
e
c
h
R
e
c
o
g
n
it
io
n
,
”
Pro
c
e
e
d
in
g
s
o
f
t
h
e
IEE
E
,
v
o
l.
7
7
,
n
o
.
2
,
p
p
.
2
5
7
–
2
8
6
,
1
9
8
9
.
[1
9
]
A
.
M
.
M
a
rti
n
e
z
,
e
t
a
l.
,
“
P
c
a
v
e
rsu
s
ld
a
,
”
IEE
E
T
ra
n
s.
Pa
tt
e
rn
An
a
l
.
M
a
c
h
.
In
tell
.
,
v
o
l.
2
3
,
n
o
.
2
,
p
p
.
2
2
8
–
2
3
3
,
2
0
0
1
.
[2
0
]
R.
Ke
y
s,
“
Cu
b
ic
Co
n
v
o
lu
ti
o
n
In
terp
o
latio
n
f
o
r
Dig
it
a
l
I
m
a
g
e
P
r
o
c
e
ss
in
g
,
”
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
Aco
u
stics
,
S
p
e
e
c
h
,
a
n
d
S
ig
n
a
l
Pro
c
e
ss
in
g
,
v
o
l.
2
9
,
n
o
.
6
,
p
p
.
1
1
5
3
–
1
1
6
0
,
De
c
1
9
8
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
J
E
C
E
Vo
l.
7
,
No
.
4
,
A
u
g
u
s
t
2
0
1
7
:
1
9
1
5
–
1
9
2
2
1922
B
I
O
G
RAP
H
I
E
S
O
F
AUTH
O
RS
O
th
m
a
n
e
El
m
e
sl
o
u
h
i
is
a
n
A
s
so
c
iate
P
r
o
f
e
ss
o
r
in
t
h
e
M
a
th
e
m
a
ti
c
s
a
n
d
Co
m
p
u
ter
S
c
ien
c
e
De
p
a
rtme
n
t
o
f
th
e
P
o
ly
d
isc
ip
li
n
a
r
y
F
a
c
u
lt
y
Ou
a
rz
a
z
a
te
o
f
Ib
n
Z
o
h
r
U
n
iv
e
rsity
,
M
o
ro
c
c
o
.
H
e
h
a
s
a
P
h
D f
ro
m
th
e
F
S
T
o
f
Ha
ss
a
n
1
st
Un
iv
e
rsit
y
,
M
o
ro
c
c
o
,
in
c
o
m
p
u
te
r
e
n
g
in
e
e
rin
g
.
His res
e
a
r
c
h
in
tere
sts
in
c
lu
d
e
:
Big
Da
ta,
Co
m
p
u
ter
V
is
io
n
,
P
a
tt
e
rn
Re
c
o
g
n
i
ti
o
n
a
n
d
M
e
d
ica
l
Im
a
g
in
g
.
E
-
m
a
il
:
o
.
e
lme
slo
u
h
i@u
iz.ac
.
m
a
Z
i
n
e
b
Elg
a
r
r
a
i
re
c
e
iv
e
d
h
e
r
e
n
g
in
e
e
r
d
e
g
re
e
in
Co
m
p
u
ter
sc
ien
c
e
in
2
0
0
9
a
t
t
h
e
Na
ti
o
n
a
l
S
c
h
o
o
l
o
f
c
o
m
p
u
ter
sc
ien
c
e
a
n
d
s
y
ste
m
s
a
n
a
ly
sis
(ENS
IA
S
)
S
c
h
o
o
l,
Ra
b
a
t,
M
o
ro
c
c
o
.
I
n
2
0
1
2
sh
e
jo
in
e
d
t
h
e
LA
V
ET
E
Lab
o
ra
to
ry
o
f
F
S
T
o
f
Ha
ss
a
n
1
st
Un
iv
e
rsity
,
S
e
tt
a
t,
M
o
ro
c
c
o
.
He
r
a
c
tu
a
l
m
a
in
re
se
a
rc
h
in
tere
sts c
o
n
c
e
rn
F
a
c
e
Re
c
o
g
n
it
i
o
n
a
n
d
Co
m
p
u
ter
V
isio
n
.
E
-
m
a
il
:
e
lg
a
rra
i@g
m
a
il
.
c
o
m
M
u
sta
p
h
a
K
a
r
d
o
u
c
h
i
is
a
f
u
ll
p
ro
f
e
ss
o
r
a
t
th
e
Co
m
p
u
ter
S
c
ien
c
e
De
p
a
rtem
e
n
t,
M
o
n
c
to
n
U
n
iv
e
rsity
,
Ca
n
a
d
a
.
His
re
se
a
rc
h
in
tere
sts
a
re
p
rim
a
ril
y
in
th
e
f
ield
o
f
Co
m
p
u
ter
V
isio
n
w
it
h
a
f
o
c
u
s
o
n
Im
a
g
e
Re
c
o
g
n
it
io
n
a
n
d
Vid
e
o
A
n
a
l
y
sis.
E
-
m
a
il
:
m
u
sta
p
h
a
.
k
a
rd
o
u
c
h
i@
u
m
o
n
c
to
n
.
c
a
H
a
k
i
m
Alla
li
w
a
s
b
o
rn
in
M
o
r
o
c
c
o
o
n
1
9
6
6
.
He
re
c
e
iv
e
d
th
e
P
h
.
D
d
e
g
re
e
f
ro
m
Clau
d
e
Be
rn
a
rd
L
y
o
n
I
Un
iv
e
rsit
y
(F
ra
n
c
e
)
in
1
9
9
3
a
n
d
th
e
“
Do
c
teu
r
d
’Et
a
t”
d
e
g
re
e
f
ro
m
Ha
ss
a
n
II
-
M
o
h
a
m
e
d
ia
Un
iv
e
rs
it
y
,
Ca
sa
b
lan
c
a
(M
o
ro
c
c
o
)
in
1
9
9
7
.
He
is
c
u
rre
n
tl
y
P
ro
f
e
ss
o
r
a
t
F
a
c
u
lt
y
o
f
S
c
ien
c
e
s
a
n
d
Tec
h
n
o
lo
g
ies
o
f
Ha
ss
a
n
1
st
Un
iv
e
rsit
y
o
f
S
e
tt
a
t
(
M
o
ro
c
c
o
)
a
n
d
d
irec
to
r
o
f
LA
V
ET
E
L
a
b
o
ra
to
ry
.
His
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
tec
h
n
o
lo
g
y
e
n
h
a
n
c
e
d
lea
rn
in
g
,
m
o
d
e
li
n
g
,
im
a
g
e
p
ro
c
e
ss
in
g
,
c
o
m
p
u
ter n
e
tw
o
rk
in
g
a
n
d
G
IS
.
E
-
m
a
il
:
h
a
k
i
m
a
ll
a
li
@h
o
tma
il
.
f
r
Evaluation Warning : The document was created with Spire.PDF for Python.