I
nte
rna
t
io
na
l J
o
urna
l o
f
Adv
a
nces in Applie
d Science
s
(
I
J
AAS)
Vo
l.
9
,
No
.
4
,
Dec
em
b
e
r
2
0
2
0
,
p
p
.
326
~
332
I
SS
N:
2
2
5
2
-
8
8
1
4
,
DOI
: 1
0
.
1
1
5
9
1
/ijaas.v
9
.
i4
.
p
p
3
2
6
-
332
326
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
a
s
.
ia
esco
r
e.
co
m
Powerful
proces
si
ng
t
o
t
hr
ee
-
dim
en
sio
na
l f
a
cia
l recog
nition
using
t
riple
inf
o
r
ma
tion
M
o
ha
m
m
a
d K
a
ri
m
i Mo
rida
ni
1
,
Aha
d K
a
rim
i Mo
rida
ni
2
,
M
a
hin
G
ho
lip
o
ur
3
1,
3
D
e
p
a
rtme
n
t
o
f
B
io
m
e
d
ica
l
E
n
g
i
n
e
e
rin
g
,
F
a
c
u
lt
y
o
f
He
a
lt
h
,
Te
h
ra
n
M
e
d
ica
l
S
c
ien
c
e
s,
Isla
m
ic Az
a
d
Un
iv
e
rsit
y
,
Ira
n
2
Tec
h
n
o
l
o
g
ies
d
e
L'
in
fo
rm
a
ti
o
n
C
o
ll
è
g
e
CDI
,
Ca
n
a
d
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ap
r
2
,
2
0
20
R
ev
is
ed
J
u
n
11
,
2
0
20
Acc
ep
ted
J
u
l
9
,
2
0
20
F
a
c
e
De
tec
ti
o
n
p
lay
s
a
c
ru
c
ial
ro
l
e
in
id
e
n
t
ify
i
n
g
i
n
d
iv
i
d
u
a
ls
a
n
d
c
rimin
a
ls
in
S
e
c
u
rit
y
,
s
u
rv
e
il
la
n
c
e
,
a
n
d
f
o
o
t
wo
rk
c
o
n
tro
l
sy
ste
m
s.
F
a
c
e
Re
c
o
g
n
it
io
n
i
n
th
e
h
u
m
a
n
is
su
p
e
r
b
,
a
n
d
p
ict
u
re
s
c
a
n
b
e
e
a
sily
id
e
n
ti
fied
e
v
e
n
a
fter
y
e
a
rs
o
f
se
p
a
ra
ti
o
n
.
T
h
e
se
a
b
il
it
ies
a
lso
a
p
p
l
y
to
c
h
a
n
g
e
s
i
n
a
fa
c
ial
e
x
p
re
ss
io
n
su
c
h
a
s
a
g
e
,
g
las
se
s,
b
e
a
rd
,
o
r
li
tt
le
c
h
a
n
g
e
in
th
e
fa
c
e
.
Th
is
m
e
th
o
d
i
s
b
a
se
d
o
n
1
5
0
t
h
re
e
-
d
ime
n
sio
n
a
l
ima
g
e
s
u
s
in
g
th
e
B
o
sp
h
o
r
u
s
d
a
tab
a
se
o
f
a
h
ig
h
ra
n
g
e
las
e
r
sc
a
n
n
e
r
in
a
Bo
g
a
z
içi
Un
iv
e
rsit
y
in
Tu
r
k
e
y
.
Th
is
p
a
p
e
r
p
re
se
n
ts
p
o
we
rfu
l
p
ro
c
e
ss
in
g
f
o
r
fa
c
e
re
c
o
g
n
it
i
o
n
b
a
se
d
o
n
a
c
o
m
b
i
n
a
ti
o
n
o
f
th
e
sa
li
e
n
t
in
f
o
rm
a
ti
o
n
a
n
d
fe
a
t
u
re
s
o
f
th
e
fa
c
e
,
su
c
h
a
s
e
y
e
s
a
n
d
n
o
se
,
fo
r
th
e
d
e
tec
ti
o
n
o
f
th
re
e
-
d
ime
n
s
io
n
a
l
fi
g
u
re
s
i
d
e
n
ti
f
ied
t
h
ro
u
g
h
a
n
a
ly
sis
o
f
su
rfa
c
e
c
u
rv
a
tu
re
.
T
h
e
Tri
n
it
y
o
f
th
e
n
o
se
a
n
d
two
e
y
e
s
we
re
se
lec
ted
fo
r
a
p
p
ly
in
g
p
ri
n
c
ip
a
l
c
o
m
p
o
n
e
n
t
a
n
a
ly
sis
a
lg
o
rit
h
m
a
n
d
su
p
p
o
r
t
v
e
c
t
o
r
m
a
c
h
in
e
to
re
v
e
a
li
n
g
a
n
d
c
las
sifica
ti
o
n
t
h
e
d
iffere
n
c
e
b
e
twe
e
n
fa
c
e
a
n
d
n
o
n
-
fa
c
e
.
Th
e
re
su
lt
s
with
d
iffere
n
t
fa
c
ial
e
x
p
re
ss
io
n
s
a
n
d
e
x
trac
ted
fro
m
d
iffere
n
t
a
n
g
les
h
a
v
e
i
n
d
ica
ted
th
e
e
fficie
n
c
y
o
f
o
u
r
p
o
we
rf
u
l
p
r
o
c
e
ss
in
g
.
K
ey
w
o
r
d
s
:
3D
im
ag
e
Facial
in
f
o
r
m
atio
n
I
n
f
o
r
m
atio
n
p
r
o
ce
s
s
in
g
,
Me
an
an
d
g
au
s
s
ian
cu
r
v
atu
r
e
SVM
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Mo
h
am
m
ad
Kar
i
m
i M
o
r
id
a
n
i
,
Dep
ar
tm
en
t o
f
B
io
m
ed
ical
E
n
g
in
ee
r
in
g
,
Facu
lty
o
f
Hea
lth
,
T
eh
r
an
Me
d
ical
Scien
ce
s
,
I
s
lam
ic
Aza
d
Un
iv
er
s
ity
,
T
eh
r
an
Pro
v
in
ce
,
T
e
h
r
an
,
Dan
esh
R
d
,
I
r
an
.
E
m
ail:
k
ar
im
i.m
@
iau
tm
u
.
ac
.
ir
1.
I
NT
RO
D
UCT
I
O
N
Face
d
etec
tio
n
h
as
g
r
ea
t
im
p
o
r
tan
ce
f
o
r
m
a
n
y
as
g
r
o
u
p
s
.
T
h
e
cr
u
cial
ap
p
licatio
n
s
o
f
Face
Dete
ctio
n
in
ar
ea
s
s
u
ch
as
h
u
m
an
-
c
o
m
p
u
ter
in
ter
f
ac
e
,
h
u
m
an
au
th
e
n
ticatio
n
,
cr
im
in
als
f
ac
e
d
etec
tio
n
,
f
ac
e
m
o
n
ito
r
in
g
s
y
s
tem
s
is
u
n
d
en
iab
le.
I
n
al
l
m
eth
o
d
s
p
r
o
p
o
s
ed
to
d
eter
m
in
e
th
e
p
o
s
itio
n
o
f
s
p
ec
if
i
c
p
o
in
ts
in
t
h
r
ee
-
d
im
en
s
io
n
al
im
ag
es,
th
e
n
o
s
e
is
r
eg
ar
d
ed
as
a
ty
p
ical
p
o
in
t,
an
d
th
e
r
est
o
f
th
e
p
o
i
n
ts
ca
n
b
e
lo
ca
te
d
.
T
o
d
eter
m
in
e
th
e
lo
ca
tio
n
o
f
t
h
e
n
o
s
e,
s
ev
er
al
m
eth
o
d
s
we
r
e
p
r
o
p
o
s
ed
.
I
n
s
o
m
e
m
eth
o
d
s
,
t
h
e
n
ea
r
est
p
o
s
itio
n
o
f
th
e
ca
m
er
a
is
r
eg
ar
d
ed
as
th
e
n
o
s
e,
an
d
th
en
th
e
o
th
e
r
p
o
in
ts
ar
e
ex
tr
ac
ted
b
a
s
ed
o
n
th
e
n
o
s
e
p
o
s
itio
n
.
T
h
is
ass
u
m
p
tio
n
is
tr
u
e
i
n
th
e
ca
s
e
o
f
n
o
r
o
tatio
n
a
r
o
u
n
d
th
e
x
a
n
d
y
-
ax
is
;
in
f
ac
t,
th
e
im
ag
e
m
u
s
t
b
e
o
n
th
e
f
r
o
n
t sid
e.
Xu
et
al.
[
1
]
i
n
tr
o
d
u
ce
d
a
f
ea
t
u
r
e
ex
tr
ac
tio
n
h
ier
ar
c
h
ical
p
la
n
to
id
en
tif
y
th
e
p
o
s
itio
n
s
o
f
t
h
e
n
o
s
e
tip
an
d
n
o
s
e
r
i
d
g
e.
T
h
ey
p
r
esen
ted
an
ef
f
ec
tiv
e
en
er
g
y
id
ea
to
s
h
o
w
th
e
lo
ca
l
d
is
tr
ib
u
tio
n
o
f
n
eig
h
b
o
r
in
g
p
o
in
ts
an
d
d
etec
t th
e
ca
n
d
id
ate
n
o
s
e
tip
s
.
I
n
th
e
en
d
,
a
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM)
class
if
ier
was u
s
ed
to
ch
o
o
s
e
th
e
r
ig
h
t
n
o
s
e
tip
s
.
Dib
ek
lio
˘
g
lu
e
t a
l.
[
2
,
3
]
d
is
p
l
ay
ed
s
tr
ateg
ies f
o
r
r
ec
o
g
n
izin
g
f
ac
ial
in
f
o
r
m
atio
n
o
n
3D
f
ac
ial
d
atasets
to
em
p
o
wer
p
o
s
tu
r
e
am
e
n
d
m
e
n
t u
n
d
e
r
im
p
o
r
tan
t p
o
s
tu
r
e
v
ar
ieties.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2
2
5
2
-
8
8
1
4
P
o
w
erfu
l p
r
o
ce
s
s
in
g
to
th
r
ee
-
d
imen
s
io
n
a
l fa
cia
l reco
g
n
itio
n
u
s
in
g
…
(
Mo
h
a
mma
d
K
a
r
imi Mo
r
id
a
n
i
)
327
T
h
ey
p
r
esen
te
d
a
s
tatis
tical
s
t
r
ateg
y
to
d
is
tin
g
u
is
h
f
ac
ial
elem
en
ts
u
s
in
g
th
e
g
r
ad
ien
t
o
f
th
e
d
ep
th
m
ap
b
ased
o
n
p
r
ep
ar
in
g
a
m
o
d
el
o
f
lo
ca
l
elem
en
ts
.
T
h
e
tech
n
iq
u
e
was
tr
ied
ag
ain
s
t
th
e
FR
GC
v
1
an
d
th
e
B
o
s
p
h
o
r
u
s
d
atab
ases
; h
o
w
ev
er
,
in
f
o
r
m
atio
n
with
p
o
s
tu
r
e
v
ar
ieties wa
s
n
o
t ta
k
en
in
to
t
h
e
in
v
esti
g
atio
n
.
T
h
ey
lik
ewise
p
r
esen
ted
a
n
o
s
e
tip
lo
ca
lizatio
n
an
d
s
eg
m
en
tatio
n
tech
n
iq
u
e
u
tili
zin
g
cu
r
v
atu
r
e
-
b
ased
h
eu
r
is
tic
in
v
esti
g
atio
n
.
Als
o
,
ev
en
th
o
u
g
h
th
e
B
o
s
p
h
o
r
u
s
d
atab
ase
u
tili
ze
d
co
m
p
r
i
s
es
o
f
3
,
3
9
6
f
ac
ial
s
ca
n
s
,
th
ey
a
r
e
ac
q
u
ir
ed
f
r
o
m
8
1
s
u
b
jects.
At
last
,
n
o
c
o
r
r
ec
t
lo
ca
lizatio
n
d
is
tan
ce
er
r
o
r
r
esu
lts
wer
e
d
is
p
lay
ed
.
I
n
o
t
h
er
r
esear
ch
,
R
o
m
e
r
o
-
H
u
er
tas
et
al.
[
4
]
in
tr
o
d
u
ce
d
a
g
r
ap
h
m
atc
h
in
g
m
eth
o
d
to
d
eter
m
in
e
th
e
p
lace
s
o
f
th
e
n
o
s
e
tip
an
d
in
ter
n
al
ey
e
co
r
n
e
r
s
.
T
h
ey
p
r
e
s
en
ted
th
e
d
is
tan
ce
to
lo
ca
l
p
l
an
e
id
ea
t
o
p
o
r
tr
ay
th
e
lo
ca
l
d
is
tr
ib
u
tio
n
o
f
n
ei
g
h
b
o
r
in
g
p
o
in
ts
an
d
id
e
n
tify
co
n
v
ex
a
n
d
co
n
ca
v
e
r
eg
io
n
s
o
f
th
e
f
ac
e.
At
lo
n
g
last
,
af
ter
th
e
g
r
a
p
h
m
atch
in
g
m
et
h
o
d
h
as
wip
ed
o
u
t
f
alse
lan
d
m
ar
k
s
,
ac
co
r
d
i
n
g
to
th
e
m
in
i
m
u
m
Ma
h
alan
o
b
is
d
is
tan
ce
,
th
e
b
est
co
m
b
in
ati
o
n
o
f
lan
d
m
ar
k
s
p
o
in
ts
is
c
h
o
s
en
to
th
e
tr
ain
ed
lan
d
m
a
r
k
d
iag
r
am
m
o
d
el.
T
h
e
tech
n
iq
u
e
was
tr
ied
a
g
ai
n
s
t
FR
GC
v
1
an
d
FR
GC
v
2
d
atab
ases
,
in
clu
d
in
g
5
0
9
s
ca
n
s
an
d
3
2
7
1
s
ca
n
s
,
r
esp
ec
tiv
ely
.
T
h
ey
d
etailed
a
co
n
s
id
e
r
ab
le
r
ate
o
f
9
0
%
wit
h
th
r
esh
o
ld
s
f
o
r
th
e
n
o
s
e
tip
at
1
5
m
m
,
a
n
d
f
o
r
th
e
in
ter
n
al
ey
e
co
r
n
e
r
s
at
1
2
m
m
.
At
last
,
Per
ak
is
et
al.
[
5
,
6
]
d
is
p
lay
ed
tech
n
iq
u
es
f
o
r
d
is
tin
g
u
is
h
in
g
im
p
o
r
tan
t
f
ac
ial
p
o
i
n
ts
s
u
ch
as
ey
e
in
n
er
an
d
o
u
ter
co
r
n
er
s
,
m
o
u
th
co
r
n
er
s
,
an
d
n
o
s
e
an
d
ch
in
tip
s
)
b
ased
o
n
2
.
5
D
s
ca
n
s
.
Neig
h
b
o
r
h
o
o
d
s
h
ap
e
an
d
cu
r
v
atu
r
e
p
r
o
ce
s
s
in
g
u
s
in
g
s
h
ap
e
in
d
ex
,
e
x
p
u
ls
io
n
m
ap
s
,
an
d
tu
r
n
p
ictu
r
es
wer
e
u
tili
ze
d
to
f
in
d
ca
n
d
id
ate
lan
d
m
ar
k
p
o
in
ts
.
T
h
ese
ar
e
d
eter
m
i
n
ed
a
n
d
m
a
r
k
ed
b
y
co
o
r
d
in
atin
g
th
em
with
a
s
tatis
tic
al
f
ac
ial
lan
d
m
ar
k
m
o
d
el
[7
,
8]
.
T
h
e
g
o
al
o
f
th
is
p
ap
e
r
is
to
f
i
n
d
o
u
t
an
ef
f
icien
t
f
ac
e
r
e
g
is
tr
atio
n
m
eth
o
d
b
ased
o
n
t
r
ip
le
.
T
h
en
we
will in
tr
o
d
u
ce
th
e
p
r
o
b
lem
s
r
e
lated
to
th
e
ex
is
tin
g
s
y
s
tem
s
in
f
ac
e
r
ec
o
g
n
itio
n
a
n
d
p
o
s
s
ib
le
r
esear
ch
way
s
th
at
h
elp
to
s
o
lv
e
th
ese
is
s
u
es.
T
h
r
ee
d
im
en
s
io
n
al
m
eth
o
d
s
co
u
l
d
p
r
o
v
id
e
b
etter
r
o
b
u
s
tn
ess
to
c
r
ea
te
d
iv
er
s
ity
th
an
2D
-
b
ased
m
eth
o
d
s
.
Au
to
m
ati
c
f
ac
e
r
ec
o
g
n
itio
n
d
etec
tio
n
h
as
b
ec
o
m
e
o
n
e
o
f
th
e
m
o
s
t
im
p
o
r
tan
t
r
esear
ch
to
p
ics
in
im
ag
e
p
r
o
ce
s
s
in
g
d
u
e
to
its
ap
p
licatio
n
in
h
u
m
a
n
-
r
o
b
o
t
in
ter
ac
tio
n
s
,
p
ai
n
d
et
ec
tio
n
,
lie
d
etec
tio
n
,
an
d
o
th
e
r
p
s
y
ch
o
lo
g
ical
an
d
m
ed
ical
ap
p
licatio
n
s
.
2.
M
AT
E
R
I
AL
A
ND
M
E
T
H
O
D
I
n
th
is
m
eth
o
d
,
it
is
ass
u
m
ed
th
at
a
tr
u
e
d
is
p
lay
o
f
a
th
r
ee
-
d
im
en
s
io
n
al
im
ag
e
as
a
p
iece
o
f
th
e
im
ag
e
f
o
r
ea
c
h
lo
ca
tio
n
(
j,
i)
c
o
o
r
d
i
n
ates
(
X,
Y,
Z
)
th
e
th
r
ee
-
d
im
e
n
s
io
n
al
s
ce
n
e.
So
m
e
o
f
th
e
im
ag
es
r
ec
eiv
ed
f
r
o
m
th
e
d
ev
ice
illu
s
tr
ate
th
e
d
ata
o
n
a
f
o
r
m
o
f
p
o
ly
g
o
n
m
o
d
els,
u
s
u
ally
ca
lle
d
th
e
tr
ian
g
u
lar
m
o
d
el.
I
n
th
is
ca
s
e,
th
e
r
an
g
e
o
f
t
h
e
im
ag
e
ca
n
b
e
d
etec
ted
u
s
in
g
th
e
Z
-
b
u
f
f
er
al
g
o
r
ith
m
[
1
]
.
As
a
r
esu
lt,
th
e
r
esu
ltin
g
im
ag
e
m
ay
in
clu
d
e
a
n
y
n
u
m
b
er
o
f
f
ac
es;
in
th
at
ca
s
e,
it
d
ep
en
d
s
o
n
t
h
e
n
ee
d
an
d
also
th
e
im
a
g
in
g
d
e
v
ice.
T
o
av
o
id
th
e
less
co
m
p
u
tatio
n
al
p
r
o
b
le
m
,
we
f
ir
s
t seek
f
ac
ial
f
ea
tu
r
es su
ch
as e
y
es a
n
d
n
o
s
e.
As a
r
esu
lt,
th
is
is
th
e
f
ir
s
t
s
tep
in
im
ag
e
s
eg
m
en
tatio
n
in
r
elate
d
ar
ea
s
o
f
th
e
f
ac
e
f
ea
tu
r
es.
C
u
r
r
en
tly
,
o
u
r
g
o
al
is
to
d
is
tin
g
u
is
h
th
e
f
ac
e
t
r
ia
n
g
le
o
n
r
ea
l
f
ac
es.
T
h
is
m
eth
o
d
is
r
e
g
is
ter
ed
f
o
r
th
e
p
o
s
itio
n
an
d
s
tan
d
ar
d
o
r
i
en
tatio
n
o
f
th
e
f
ac
e
an
d
r
ed
u
ctio
n
o
f
th
e
v
ar
iab
ilit
y
o
f
th
e
d
ep
th
o
f
th
e
f
ac
e.
T
h
e
tr
ian
g
le
d
is
tan
ce
o
f
th
e
n
o
s
e
an
d
ey
es in
th
e
im
a
g
e
is
also
ca
lcu
lated
.
2
.
1
.
No
s
e
a
nd
ey
es ident
if
ica
t
io
n
T
h
e
HK
s
eg
m
en
tatio
n
m
eth
o
d
is
u
s
ed
to
f
i
n
d
t
h
e
n
o
s
e
an
d
th
e
in
n
er
c
o
r
n
e
r
s
o
f
th
e
ey
es
.
Fo
r
a
s
u
r
f
ac
e,
with
a
c
u
r
v
atu
r
e
H
m
ea
n
an
d
K
Gau
s
s
ian
cu
r
v
atu
r
e
,
th
e
f
ac
ial
ex
p
r
ess
io
n
ca
n
b
e
i
d
en
tifie
d
.
I
n
th
is
ca
s
e,
th
e
n
o
s
e
is
th
e
m
ax
im
u
m
p
o
i
n
t,
an
d
th
e
c
o
r
n
er
s
o
f
th
e
ey
es
ar
e
th
e
m
in
im
u
m
.
B
ased
o
n
t
h
e
av
er
a
g
e
c
u
r
v
atu
r
e
o
f
t
h
e
f
ac
e,
an
d
b
ec
a
u
s
e
o
f
th
e
lead
in
g
p
o
s
itio
n
an
d
h
illy
o
f
th
e
n
o
s
e,
it
h
as
t
h
e
h
ig
h
est
cu
r
v
atu
r
e
in
th
e
s
u
r
r
o
u
n
d
i
n
g
a
r
ea
,
s
o
r
eg
ar
d
in
g
th
e
f
ac
e,
th
e
p
o
in
ts
with
th
e
m
ax
im
u
m
cu
r
v
atu
r
e
ar
e
s
ea
r
ch
ed
.
Fo
r
th
is
p
u
r
p
o
s
e,
th
e
m
ea
n
s
u
r
f
ac
e
cu
r
v
atu
r
e
is
ca
lcu
lated
.
Af
ter
war
d
,
th
e
ar
ea
s
with
a
g
r
ea
ter
cu
r
v
atu
r
e
s
p
ec
if
ied
an
d
th
e
ar
ea
with
th
e
g
r
ea
test
cu
r
v
atu
r
e
will
b
e
t
h
e
tip
o
f
t
h
e
n
o
s
e.
I
f
th
e
cu
r
v
atu
r
e
o
f
an
ar
ea
is
g
r
ea
ter
,
th
er
ef
o
r
e
t
h
e
to
tal
am
o
u
n
t
o
f
cu
r
v
atu
r
e
in
t
h
e
a
r
ea
will
b
e
m
o
r
e
o
b
v
i
o
u
s
.
I
n
a
d
d
it
io
n
to
th
at,
b
ec
au
s
e
o
f
th
e
n
o
is
e,
C
u
r
v
atu
r
e
en
h
a
n
ce
s
in
o
n
e
o
r
s
ev
er
al
p
ix
els,
b
u
t
th
e
h
ig
h
cu
r
v
atu
r
e
o
f
t
h
e
s
ea
r
ch
ed
ar
ea
is
cu
r
v
ed
,
th
e
f
ilter
r
e
d
u
ce
s
th
e
n
o
is
e
an
d
av
er
a
g
es
th
e
e
f
f
ec
ts
o
f
p
r
ev
e
n
tin
g
th
e
wr
o
n
g
lo
ca
tio
n
b
y
ca
lcu
latin
g
t
h
e
m
ea
n
[
9
]
.
T
h
e
im
ag
e
o
f
th
e
m
ea
n
cu
r
v
atu
r
e,
b
ef
o
r
e
an
d
af
ter
th
e
f
ilter
in
g
is
s
h
o
wn
in
Fig
u
r
e
1.
T
h
is
m
eth
o
d
d
eter
m
i
n
es
th
e
p
o
s
itio
n
o
f
th
e
n
o
s
e
u
n
d
e
r
an
y
d
esire
d
an
g
le
ar
o
u
n
d
th
e
x
,
y
,
a
n
d
z.
T
h
e
o
n
ly
lim
itatio
n
is
m
o
r
e
l
ik
ely
if
m
o
r
e
th
an
h
alf
o
f
th
e
n
o
s
e
is
h
id
d
en
a
r
o
u
n
d
th
e
y
-
ax
is
.
T
h
is
m
eth
o
d
u
tili
ze
s
th
e
m
ea
n
cu
r
v
atu
r
e
o
f
th
e
f
ac
e
an
d
d
o
es
n
o
t
tak
e
ad
v
an
tag
e
o
f
th
e
e
d
u
ca
tio
n
al
an
d
class
if
ied
d
ata.
Fo
r
th
is
r
ea
s
o
n
,
th
is
m
et
h
o
d
h
as
a
h
ig
h
er
co
m
p
u
tatio
n
al
s
p
ee
d
th
a
n
th
e
m
eth
o
d
s
th
at
f
in
d
th
e
n
o
s
e
f
r
o
m
d
if
f
er
en
t a
n
g
les.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
: 2
2
5
2
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
9
,
No
.
4
,
Dec
em
b
e
r
2
0
2
0
:
3
2
6
–
3
3
2
328
Fig
u
r
e
1
.
T
h
e
f
r
o
n
t p
a
r
t o
f
t
h
e
f
ac
e
(
to
p
r
ig
h
t)
,
th
e
im
ag
e
o
f
a
v
er
ag
e
cu
r
v
atu
r
e
(
to
p
lef
t)
,
th
e
f
r
o
n
t
p
ar
t o
f
th
e
f
ac
e
with
m
o
r
e
lig
h
t in
ten
s
ity
(
b
o
tto
m
le
f
t)
,
th
e
im
a
g
e
o
f
th
e
av
er
ag
e
c
u
r
v
at
u
r
e
af
ter
f
ilter
in
g
(
b
o
tto
m
r
ig
h
t)
2
.
2
.
R
ec
o
g
nitio
n in im
a
g
e
ro
t
a
t
io
n
T
h
e
co
m
p
u
tatio
n
al
co
s
t
o
f
th
e
u
s
e
o
f
th
e
r
o
tatin
g
im
ag
e
is
h
ig
h
.
I
n
itially
,
a
s
er
ies
o
f
ar
e
as
th
at
ar
e
lik
ely
to
lo
ca
te
th
e
n
o
s
e
an
d
e
y
es
ar
e
s
p
ec
if
ied
,
th
e
n
to
lo
ca
te
th
e
ex
ac
t
ar
ea
,
th
e
r
o
tatin
g
im
ag
e
is
ex
tr
ac
ted
,
an
d
u
s
in
g
SVM
(
Su
p
p
o
r
t
Vec
to
r
Ma
ch
in
e)
class
if
icatio
n
,
th
e
ex
ac
t
lo
ca
ted
is
s
p
ec
if
ied
[
1
0
,
1
1
]
.
T
h
is
m
eth
o
d
n
ee
d
s
p
r
im
ar
y
d
ata
an
d
a
n
eu
r
al
n
etwo
r
k
.
T
o
d
eter
m
i
n
e
th
e
lo
ca
tio
n
o
f
th
e
n
o
s
e,
th
e
an
g
les
ar
o
u
n
d
th
e
y
-
ax
is
is
q
u
an
tized
,
an
d
th
en
th
e
im
a
g
es
r
o
tate
ar
o
u
n
d
th
e
y
-
ax
is
,
th
en
th
e
n
ea
r
est
p
o
in
t
to
th
e
ca
m
er
a
is
co
n
s
id
er
ed
as
a
ca
n
d
id
ate.
Af
ter
a
f
u
ll
r
o
tatio
n
ar
o
u
n
d
t
h
e
y
-
a
x
is
,
th
e
n
o
s
e
is
d
iag
n
o
s
ed
.
T
h
is
m
eth
o
d
ca
n
i
d
en
tif
y
th
e
n
o
s
e
p
o
s
itio
n
if
th
e
r
o
tati
o
n
is
ar
o
u
n
d
t
h
e
y
-
ax
is
,
an
d
th
is
m
eth
o
d
is
n
o
t
co
s
t
-
ef
f
ec
t
iv
e
if
th
e
r
o
tatio
n
h
ap
p
en
s
to
t
h
e
o
th
e
r
an
g
les.
2
.
3
.
T
hree
-
dim
ens
io
na
l f
a
ce
ro
t
a
t
io
n
T
h
r
ee
r
o
tatio
n
al
m
atr
ices
o
r
th
eir
m
u
ltip
lied
s
tr
u
ctu
r
e
ca
n
b
e
u
s
ed
to
r
o
tate
th
r
ee
-
d
im
en
s
io
n
al
im
ag
es
ar
o
u
n
d
th
e
ax
es
o
f
r
o
t
atio
n
[
1
2
-
1
4
]
.
I
f
we
ass
u
m
e
th
at
alp
h
a
is
r
o
tatio
n
ar
o
u
n
d
ax
i
s
X,
B
eta
is
r
o
tatio
n
ar
o
u
n
d
ax
is
Y,
a
n
d
L
am
b
d
a
is
r
o
tatio
n
ar
o
u
n
d
t
h
e
ax
is
o
f
Z
,
th
en
th
e
f
o
r
m
u
la
a
r
o
u
n
d
ea
c
h
ax
is
will
b
e
as
(
1
)
to
(
3
)
.
I
n
f
o
r
m
u
la
(
4
)
th
e
r
esu
l
t is th
e
b
y
m
u
ltip
ly
in
g
f
o
r
m
o
f
th
e
th
r
ee
m
atr
ices:
1
(
)
=
[
1
0
0
0
0
−
]
(
1
)
2
(
)
=
[
0
−
0
1
0
]
(
2
)
3
(
)
=
[
0
−
1
0
0
0
1
]
(
3
)
=
1
∗
2
∗
3
=
[
.
.
.
+
.
−
.
.
+
.
−
.
−
.
.
+
.
.
.
+
.
−
.
.
]
(
4
)
As
a
r
esu
lt,
v
ec
to
r
s
o
n
ea
ch
s
p
in
is
o
b
tain
ed
with
t
h
e
co
o
r
d
in
atio
n
o
f
th
e
r
o
ta
ted
im
ag
e.
T
h
e
in
ten
s
ity
is
ad
d
ed
to
u
s
e
th
is
m
atr
ix
,
th
r
ee
-
d
im
en
s
io
n
al
d
ata,
an
d
two
-
d
im
en
s
io
n
al
im
ag
e
r
o
tatio
n
.
So
,
th
e
im
ag
e
r
o
tatio
n
co
o
r
d
in
ati
o
n
is
f
o
u
n
d
.
T
h
e
d
ata
m
atr
ix
is
ch
an
g
ed
as
f
o
r
m
u
la
(
5
)
,
an
d
th
e
in
ten
s
ity
was
ad
d
ed
to
th
e
last
lin
e
o
f
im
a
g
e
d
ata.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2
2
5
2
-
8
8
1
4
P
o
w
erfu
l p
r
o
ce
s
s
in
g
to
th
r
ee
-
d
imen
s
io
n
a
l fa
cia
l reco
g
n
itio
n
u
s
in
g
…
(
Mo
h
a
mma
d
K
a
r
imi Mo
r
id
a
n
i
)
329
=
[
1
1
1
1
2
2
2
2
3
3
3
3
4
4
4
4
5
5
5
5
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
]
(
5
)
T
h
e
im
p
o
r
tan
t
p
o
in
t
h
er
e
is
th
at
two
an
d
th
r
ee
-
d
im
e
n
s
io
n
al
im
ag
es
o
f
th
e
s
am
e
s
ize,
an
d
b
o
th
m
u
s
t
b
e
ac
cu
r
ate
with
a
c
o
m
p
lete
p
ictu
r
e
o
f
t
h
e
i
n
ter
ac
tio
n
co
m
p
lian
ce
.
T
h
e
d
atab
ase
u
s
ed
;
t
h
e
two
-
d
im
en
s
io
n
al
im
ag
es
ar
e
lar
g
er
th
an
th
e
t
h
r
ee
-
d
im
en
s
io
n
al
im
a
g
es.
T
h
er
e
f
o
r
e,
f
ir
s
t
two
-
d
im
e
n
s
io
n
al
im
ag
es
ar
e
r
esized
to
b
e
th
e
s
am
e
s
ize
w
ith
d
im
en
s
io
n
s
o
f
th
r
ee
-
d
im
en
s
io
n
al
im
a
g
es.
Af
ter
t
h
at,
a
co
m
m
o
n
p
o
i
n
t
in
b
o
th
im
ag
es
is
m
ar
k
ed
,
a
n
d
two
im
ag
es
ar
e
ch
ar
ac
ter
ized
b
y
y
an
d
x
o
v
er
lap
,
th
e
n
th
e
d
ata
is
av
ailab
le
f
o
r
ea
ch
co
o
r
d
in
atio
n
[
1
5
,
1
6
]
.
Du
e
t
o
th
e
ch
a
n
g
e
o
f
th
e
d
ata
m
at
r
ix
,
a
r
o
tati
o
n
m
atr
i
x
also
n
e
ed
s
to
b
e
c
h
an
g
e
d
.
Sin
ce
th
e
in
ten
s
ity
d
ata
d
o
es n
o
t a
f
f
ec
t th
e
p
ix
el
lo
ca
tio
n
,
th
e
r
o
tatio
n
m
atr
i
x
will b
e
as f
o
r
m
u
la
(
6
)
.
W
h
er
e
th
e
a
v
ailab
le
ze
r
o
a
n
d
o
n
e
d
ata
ar
e
ad
d
ed
to
th
e
r
o
ta
tio
n
m
atr
ix
,
s
o
th
e
lig
h
t
in
te
n
s
ity
d
ata
is
tr
an
s
m
itted
with
o
u
t a
lter
in
g
.
[
0
0
0
0
0
0
1
]
(
6
)
T
h
en
th
e
co
llected
d
ata
in
a
t
wo
-
d
im
en
s
io
n
al
m
atr
ix
ar
e
tr
a
n
s
f
o
r
m
ed
in
to
a
two
-
d
im
e
n
s
io
n
al
r
o
tated
im
ag
e
[
1
7
]
.
2
.
4
.
F
a
ce
det
ec
t
io
n
T
h
e
u
s
e
o
f
in
f
o
r
m
atio
n
p
r
o
ce
d
u
r
e
h
as
s
ev
er
al
ad
v
a
n
tag
es.
First
o
f
all,
th
e
f
ir
s
t
d
ata
is
f
o
r
th
e
d
ep
th
,
n
o
t
th
e
in
ten
s
ity
o
f
th
e
lig
h
t,
s
o
in
d
e
p
en
d
e
n
t
o
f
t
h
e
s
ev
er
ity
o
f
th
e
lig
h
t
o
r
th
e
lig
h
t
r
ad
iati
o
n
a
n
g
le
o
r
th
e
f
ac
e
s
ize,
th
e
o
b
tain
ed
im
ag
e
is
f
ix
ed
.
Als
o
,
th
ese
d
ata
d
o
n
o
t
u
s
e
th
e
r
ef
lecte
d
lig
h
t
f
r
o
m
th
e
f
ac
e.
T
h
er
ef
o
r
e,
it
is
n
o
t
d
ep
en
d
en
t
o
n
th
e
s
k
in
co
lo
r
ch
an
g
es
af
f
ec
ted
b
y
th
e
m
ak
eu
p
o
r
s
u
n
lig
h
t
an
d
s
u
n
b
u
r
n
.
Gen
er
ally
,
th
e
th
r
ee
-
d
im
en
s
io
n
al
d
ata
c
an
b
e
m
o
v
ed
an
d
r
o
tated
i
n
th
e
d
esire
d
a
n
g
le
in
th
r
ee
-
d
im
en
s
io
n
al
s
p
ac
e;
ac
co
r
d
in
g
l
y
,
t
h
e
f
o
r
m
u
la
is
e
x
tr
ac
ted
u
n
d
er
th
e
d
esire
d
an
g
le.
T
h
e
r
ef
o
r
e
,
if
th
e
im
ag
e
is
n
o
t
at
th
e
d
esire
d
an
g
le,
it c
an
b
e
r
o
tated
to
ca
lc
u
late
th
e
im
ag
e
at
th
e
d
esire
d
an
g
le
[
1
8
-
2
0
]
.
T
h
e
n
ee
d
f
o
r
th
r
ee
-
d
im
e
n
s
io
n
al
im
ag
in
g
ca
m
er
as,
wh
ic
h
ar
e
m
o
r
e
ex
p
en
s
iv
e
th
an
two
-
d
i
m
en
s
io
n
al
im
ag
in
g
ca
m
er
as,
is
o
n
e
o
f
t
h
e
d
is
ad
v
a
n
tag
es
o
f
th
is
m
et
h
o
d
.
Facial
h
air
is
o
n
e
o
f
th
e
ca
s
es
wh
ich
m
ak
es
th
e
ex
tr
ac
tio
n
o
f
d
ep
th
d
ata
d
if
f
icu
lt
[
2
0
]
.
Du
e
to
th
e
d
et
ec
tio
n
o
f
f
ac
es
b
y
th
e
tr
in
ity
o
f
ey
es
a
n
d
n
o
s
e,
th
e
s
y
s
tem
ca
n
g
et
alo
n
g
with
p
r
o
b
lem
s
s
u
ch
as h
i
d
in
g
th
e
e
ar
,
h
air
,
e
y
eb
r
o
ws,
etc.
(
Fig
u
r
e
2
)
.
Fig
u
r
e
2
.
Face
d
etec
tio
n
u
s
in
g
n
o
s
e
an
d
e
y
es
T
h
e
co
m
p
lete
p
r
o
ce
d
u
r
e
ca
n
b
e
ex
tr
ac
ted
th
r
o
u
g
h
a
p
a
n
o
r
am
ic
ca
m
er
a
o
r
m
u
ltip
le
p
h
o
to
s
in
d
if
f
er
en
t
d
ir
ec
tio
n
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
: 2
2
5
2
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
9
,
No
.
4
,
Dec
em
b
e
r
2
0
2
0
:
3
2
6
–
3
3
2
330
2
.
5
.
E
x
perim
ent
a
l t
esting
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
im
p
lem
en
ted
o
n
th
e
B
o
s
p
h
o
r
u
s
d
atab
ase
o
n
1
5
0
f
ac
es
s
h
o
ws
th
at
t
h
is
m
eth
o
d
is
9
9
.
7
%
ac
cu
r
ate
in
th
e
d
iag
n
o
s
is
o
f
th
e
p
o
s
itio
n
o
f
th
e
n
o
s
e,
9
9
.
3
%
co
r
r
ec
t
d
iag
n
o
s
is
o
f
th
e
p
o
s
itio
n
o
f
th
e
ey
es,
an
d
9
6
.
7
%
ac
cu
r
ate
d
iag
n
o
s
is
o
f
e
y
e
p
o
s
itio
n
.
I
n
T
ab
le
1
,
th
e
r
esu
lts
d
eter
m
in
e
th
e
lo
ca
tio
n
o
f
th
e
ey
es,
n
o
s
e,
a
n
d
in
ter
n
al
p
a
r
ts
o
f
th
e
ey
es,
wh
ich
wer
e
s
h
o
wn
in
th
e
d
atab
ase.
T
ab
le
1
.
T
h
e
n
u
m
b
er
o
f
n
o
s
e
an
d
ey
es p
o
s
itio
n
s
wh
ich
wer
e
d
etec
ted
V
i
e
w
i
n
g
A
n
g
l
e
N
o
se
D
e
t
e
c
t
i
o
n
Le
f
t
Ey
e
D
e
t
e
c
t
i
o
n
R
i
g
h
t
E
y
e
D
e
t
e
c
t
i
o
n
F
r
o
n
t
P
h
o
t
o
1
4
3
1
4
1
1
3
9
R
o
t
a
t
e
2
5
d
e
g
r
e
e
s
a
r
o
u
n
d
Y
1
4
3
1
4
3
1
3
9
Lo
w
r
o
t
a
t
i
o
n
a
r
o
u
n
d
t
h
e
z
-
a
x
i
s
1
4
3
1
4
3
1
4
1
H
i
g
h
r
o
t
a
t
i
o
n
a
r
o
u
n
d
t
h
e
z
-
a
x
i
s
1
4
3
1
4
3
1
4
1
S
mi
l
e
P
i
c
t
u
r
e
1
4
3
1
4
2
1
4
0
O
p
e
n
M
o
u
t
h
1
4
1
1
4
1
1
4
0
Th
e
r
o
t
a
t
i
o
n
a
r
o
u
n
d
t
h
e
x
-
a
x
i
s (a
b
o
v
e
)
1
4
3
1
4
3
1
4
1
Th
e
r
o
t
a
t
i
o
n
a
r
o
u
n
d
t
h
e
x
-
a
x
i
s
(
d
o
w
n
)
1
4
2
1
4
2
3.
RE
SU
L
T
S
AND
D
I
SCU
SS
I
O
N
T
h
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
is
i
m
p
lem
en
ted
o
n
1
4
p
ictu
r
es
o
f
1
5
0
p
e
r
s
o
n
s
is
s
h
o
wn
in
T
ab
le
2
.
I
n
Fig
u
r
e
3
th
e
ac
c
u
r
ac
y
g
r
ap
h
is
p
lo
tted
f
o
r
d
if
f
er
e
n
t e
r
r
o
r
s
.
T
ab
le
2
.
E
s
tim
atio
n
ac
c
u
r
ac
y
o
f
an
g
le
f
o
r
v
ar
io
u
s
er
r
o
r
s
Th
e
e
r
r
o
r
z
-
A
x
i
s
y
-
A
x
i
s
x
-
A
x
i
s
1
0
°
e
r
r
o
r
.
2
1
9
8
%
9
9
.
5
2
%
9
8
.
8
1
%
6
°
Er
r
o
r
9
4
.
8
8
%
9
8
.
8
1
%
9
2
.
3
3
%
3
°
Er
r
o
r
7
7
.
4
6
%
9
5
.
0
0
%
6
7
.
2
4
%
Fig
u
r
e
3
.
Acc
u
r
ac
y
d
ia
g
r
am
f
o
r
d
if
f
e
r
en
t e
r
r
o
r
s
T
h
e
r
u
n
n
in
g
tim
e
o
f
th
is
Alg
o
r
ith
m
,
r
eg
ar
d
less
o
f
th
e
ti
m
e
n
ee
d
ed
to
d
ete
r
m
in
e
th
e
lo
ca
tio
n
o
f
th
e
n
o
s
e,
is
less
th
an
8
0
0
m
illi
s
ec
o
n
d
s
.
Mo
r
eo
v
er
,
t
h
is
m
eth
o
d
r
eq
u
ir
es
n
o
d
ata
f
o
r
tr
ain
in
g
.
B
ec
au
s
e
o
f
th
e
h
ier
ar
ch
ical
s
tr
u
ctu
r
e
i
n
c
ase
o
f
th
e
wr
o
n
g
esti
m
a
tio
n
o
f
th
e
an
g
les,
th
e
wr
o
n
g
esti
m
atio
n
o
f
th
e
o
th
er
an
g
les
is
also
p
o
s
s
ib
le.
Du
e
t
o
th
e
d
ata
m
ap
p
in
g
o
f
th
e
Z
=
0
,
th
e
ex
is
ten
ce
o
f
an
er
r
o
r
in
th
e
ca
lcu
latio
n
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2
2
5
2
-
8
8
1
4
P
o
w
erfu
l p
r
o
ce
s
s
in
g
to
th
r
ee
-
d
imen
s
io
n
a
l fa
cia
l reco
g
n
itio
n
u
s
in
g
…
(
Mo
h
a
mma
d
K
a
r
imi Mo
r
id
a
n
i
)
331
th
e
an
g
le
in
th
e
d
i
r
ec
tio
n
o
f
t
h
e
z
-
ax
is
is
p
o
s
s
ib
le.
I
f
we
ca
n
ca
lcu
late
p
ar
a
m
eter
s
u
s
in
g
t
h
e
ellip
s
e
in
th
r
ee
-
d
im
en
s
io
n
al
s
p
ac
e,
th
en
th
e
ac
cu
r
ac
y
ca
n
b
e
i
n
cr
ea
s
ed
.
Als
o
,
if
th
e
an
g
le
alo
n
g
t
h
e
x
-
ax
is
an
d
y
ca
n
b
e
d
eter
m
in
ed
f
r
o
m
th
e
o
r
ig
in
a
l
im
ag
e
alo
n
g
with
th
e
an
g
le
d
eter
m
in
atio
n
,
th
e
n
th
e
Sp
ee
d
an
d
e
r
r
o
r
o
f
th
e
m
eth
o
d
will b
e
d
ec
r
ea
s
ed
.
On
e
o
f
th
e
m
o
s
t
im
p
o
r
tan
t
ca
p
ab
ilit
ies
o
f
th
e
h
u
m
a
n
v
i
s
u
al
s
y
s
tem
is
th
e
ab
ilit
y
to
r
ec
o
g
n
ize
th
e
f
ac
e,
wh
ic
h
p
lay
s
a
n
im
p
o
r
tan
t
r
o
le
in
th
e
s
o
cial
life
o
f
e
ac
h
p
er
s
o
n
.
Ou
r
r
elatio
n
s
h
ip
with
o
th
er
p
eo
p
le
is
b
ased
o
n
o
u
r
a
b
ilit
y
to
id
en
tify
th
em
.
On
e
o
f
th
e
m
o
s
t
p
r
o
m
in
en
t
asp
ec
ts
o
f
th
e
h
u
m
a
n
f
ac
e
r
ec
o
g
n
itio
n
s
y
s
tem
is
its
h
ig
h
ac
cu
r
ac
y
an
d
r
eliab
ilit
y
,
b
u
t
all
h
u
m
an
b
ein
g
s
d
o
f
ac
e
r
ec
o
g
n
itio
n
ea
s
ily
an
d
with
o
u
t
m
u
ch
ef
f
o
r
t,
wh
ile
th
is
is
n
o
t
an
ea
s
y
task
f
o
r
co
m
p
u
ter
s
.
T
h
e
m
a
in
p
u
r
p
o
s
e
o
f
th
is
s
tu
d
y
is
to
s
tu
d
y
th
e
ef
f
ec
t
o
f
s
elec
tin
g
th
e
ap
p
r
o
p
r
iate
f
ea
tu
r
es
o
f
th
r
ee
-
d
im
e
n
s
io
n
al
im
ag
es
f
o
r
f
ac
e
r
ec
o
g
n
itio
n
.
T
h
e
r
esu
lts
s
h
o
w
th
e
s
u
cc
ess
o
f
th
is
m
eth
o
d
in
id
en
tify
in
g
f
ac
es
with
d
if
f
er
e
n
t
ch
ar
ac
ter
is
tics
.
Dif
f
er
en
t
a
p
p
r
o
ac
h
es
s
o
f
ar
ar
e
s
u
m
m
ar
ized
in
T
a
b
le
.
3.
T
ab
le
3
.
C
o
m
p
a
r
is
o
n
o
f
class
i
f
icatio
n
ac
cu
r
ac
y
of
d
if
f
er
e
n
t
m
eth
o
d
s
N
a
me
3D
f
a
c
e
d
a
t
a
b
a
s
e
M
a
x
i
mu
m
A
c
c
u
r
a
c
y
B
l
a
n
z
e
t
a
l
[
2
1
]
F
R
G
C
9
2
%
D
i
b
e
k
l
i
o
g
l
u
e
t
a
l
[
2
2
]
B
o
s
p
h
o
r
u
s
7
9
.
4
1
%
M
a
h
m
o
o
d
e
t
a
l
[
2
3
]
G
a
v
a
b
D
B
9
0
%
B
e
r
r
e
t
t
i
e
t
a
l
[
2
4
]
U
N
D
F
R
G
C
v
2
.
0
G
a
v
a
b
D
B
8
2
.
1
%
P
e
r
a
k
i
s e
t
a
l
[
2
5
]
UND
–
D
i
n
g
e
t
a
l
[
2
6
]
LFW
9
2
.
9
5
%
Pr
o
p
o
sed
m
e
t
h
o
d
B
o
sp
h
o
r
u
s
9
9
.
5
2
%
Du
e
to
th
e
s
en
s
itiv
e
n
atu
r
e
o
f
f
ac
e
d
etec
tio
n
,
h
i
g
h
p
r
ec
is
io
n
is
o
f
th
e
u
tm
o
s
t
i
m
p
o
r
tan
ce
.
Ob
tain
ed
r
esu
lts
s
h
o
w
th
at
th
e
o
p
en
m
o
u
t
h
in
cr
ea
s
es
th
e
p
o
s
s
ib
ilit
y
o
f
th
e
lo
s
s
o
f
th
e
n
o
s
e.
C
o
n
s
id
er
in
g
th
at
th
is
m
eth
o
d
is
n
o
t
o
n
l
y
to
id
e
n
tify
th
e
n
o
s
e
an
d
ey
es
id
en
ti
f
icatio
n
ar
e
also
co
n
s
id
er
e
d
,
s
o
th
is
wea
k
n
ess
is
co
v
er
ed
,
b
u
t
in
th
e
s
itu
atio
n
in
wh
ich
b
o
th
ey
es
ar
e
co
v
er
e
d
o
r
s
o
m
eh
o
w
th
at
p
ar
t
o
f
th
e
im
ag
e
is
lo
s
t
o
r
if
th
e
m
o
u
th
is
o
p
en
; th
e
i
d
en
tifi
ca
tio
n
will b
e
less
th
an
wh
en
t
h
e
m
o
u
t
h
is
clo
s
ed
.
4.
CO
NCLU
SI
O
N
Hu
m
an
f
ac
e
r
ec
o
g
n
itio
n
b
y
m
ac
h
in
e
th
r
o
u
g
h
3D
im
ag
es
h
as
b
ec
o
m
e
an
ac
tiv
e
r
esear
c
h
p
h
ase
in
im
ag
e
p
r
o
ce
s
s
in
g
c
o
m
m
u
n
ities
,
p
atter
n
r
ec
o
g
n
itio
n
,
n
eu
r
al
n
etwo
r
k
s
,
an
d
c
o
m
p
u
te
r
v
is
io
n
.
T
h
e
f
ac
e
p
lay
s
an
ess
en
tial
r
o
le
in
id
en
tify
in
g
p
eo
p
le
an
d
s
h
o
win
g
th
eir
f
ee
lin
g
s
at
th
e
co
m
m
u
n
ity
lev
el.
T
h
e
h
u
m
an
ab
ilit
y
to
r
e
co
g
n
ize
f
ac
es
is
r
em
ar
k
ab
le.
W
e
c
an
r
ec
o
g
n
ize
th
o
u
s
an
d
s
o
f
m
em
o
r
a
b
le
f
ac
es
in
o
u
r
life
tim
e
a
n
d
,
at
a
g
lan
ce
,
r
ec
o
g
n
ize
f
am
iliar
f
ac
es
ev
en
af
te
r
y
ea
r
s
o
f
s
ep
ar
atio
n
.
Face
r
ec
o
g
n
itio
n
h
as
b
ec
o
m
e
a
n
im
p
o
r
tan
t
i
s
s
u
e
in
ap
p
licatio
n
s
s
u
ch
as
s
ec
u
r
ity
s
y
s
tem
s
,
cr
ed
it
ca
r
d
co
n
tr
o
l
a
n
d
cr
im
e
id
en
tific
atio
n
.
I
n
th
is
ar
ticle,
a
n
ew
m
eth
o
d
t
o
f
ac
e
r
ec
o
g
n
itio
n
a
u
to
m
atica
l
ly
u
s
in
g
3D
f
ac
e
im
ag
es
was
p
r
esen
ted
.
T
h
e
g
eo
m
etr
ic
f
ea
tu
r
es
o
f
th
e
f
ac
e
wer
e
u
s
ed
to
id
en
tify
f
a
ce
s
with
th
e
h
elp
o
f
n
eu
r
al
n
e
two
r
k
s
.
T
h
e
r
esu
lts
s
h
o
wed
.
T
h
e
r
esu
lts
o
f
u
s
in
g
th
e
p
r
o
p
o
s
ed
m
eth
o
d
in
th
is
p
a
p
er
with
th
e
h
elp
o
f
B
o
s
p
h
o
r
u
s
d
atab
ase
s
h
o
wed
th
e
h
ig
h
est p
er
ce
n
tag
e
o
f
ac
cu
r
ac
y
co
m
p
ar
ed
t
o
p
r
e
v
io
u
s
m
e
th
o
d
s
.
ACK
NO
WL
E
DG
M
E
N
T
I
wo
u
ld
lik
e
to
ex
p
r
ess
m
y
d
ee
p
est
ap
p
r
ec
iatio
n
to
Pr
o
f
ess
o
r
B
u
len
t
E
s
n
ik
o
r
f
r
o
m
B
o
ğ
az
içi
Un
iv
er
s
ity
f
o
r
p
r
o
v
id
in
g
m
e
w
ith
s
ig
n
if
ican
t d
ata
th
at
g
r
ea
tly
ass
is
ted
th
e
r
esear
ch
.
RE
F
E
R
E
NC
E
S
[1
]
Xu
C.
,
Tan
T.
,
a
n
d
Wa
n
g
Y.,
Qu
a
n
L.
,
“
Co
m
b
in
i
n
g
l
o
c
a
l
fe
a
tu
re
s
fo
r
ro
b
u
st
n
o
se
lo
c
a
ti
o
n
in
3D
fa
c
ial
d
a
ta
,”
Pa
tt
e
rn
Rec
o
g
n
it
io
n
L
e
tt
e
rs
,
v
o
l
.
2
7
,
n
o
.
1
3
,
p
p
.
62
-
7
3
,
2
0
0
6
.
[2
]
Dib
e
k
li
o
g
l
u
H.
,
Pa
rt
-
b
a
se
d
3D
fa
c
e
re
c
o
g
n
it
io
n
u
n
d
e
r
p
o
se
a
n
d
e
x
p
re
ss
io
n
v
a
ria
t
io
n
s
,
M
a
ste
r'
s
th
e
sis,
Bo
_
g
a
z
i_
c
i
Un
iv
e
rsity
,
2
0
0
8
.
[3
]
Dib
e
k
li
o
g
l
u
H.,
S
a
lah
A
.
,
Ak
a
r
u
n
L.
,
“
3D
fa
c
ial
lan
d
m
a
rk
in
g
u
n
d
e
r
e
x
p
re
ss
io
n
,
p
o
se
,
a
n
d
o
c
c
lu
sio
n
v
a
riatio
n
s
,”
I
n
Pro
c
.
2
n
d
I
EE
E
I
n
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
B
io
me
trics
:
T
h
e
o
ry
,
Ap
p
li
c
a
ti
o
n
s
a
n
d
S
y
ste
ms
,
p
p
.
1
-
6
,
2
0
0
8
.
[4
]
Ro
m
e
ro
-
Hu
e
rtas
M
.
,
P
e
a
rs
N.
,
“
3D
fa
c
ial
lan
d
m
a
rk
l
o
c
a
li
z
a
ti
o
n
b
y
m
a
tch
in
g
sim
p
le
d
e
sc
rip
to
rs
,”
In
Pr
o
c
.
2
n
d
IEE
E
In
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
Bi
o
me
trics
:
T
h
e
o
ry
,
A
p
p
l
ica
ti
o
n
s a
n
d
S
y
ste
ms
,
2
0
0
8
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
: 2
2
5
2
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
9
,
No
.
4
,
Dec
em
b
e
r
2
0
2
0
:
3
2
6
–
3
3
2
332
[5
]
P
e
ra
k
is
P
.
,
P
a
ss
a
li
s
G
.
,
Th
e
o
h
a
ri
s
T.
,
To
d
e
rici
G
.
,
Ka
k
a
d
iaris
I.
,
“
P
a
rti
a
l
m
a
tch
in
g
o
f
in
terp
o
se
3D
fa
c
ial
d
a
ta
fo
r
fa
c
e
re
c
o
g
n
it
i
o
n
,”
In
Pro
c
.
3
rd
I
EE
E
In
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
Bi
o
me
trics
:T
h
e
o
ry
,
A
p
p
li
c
a
ti
o
n
s
a
n
d
S
y
ste
ms
,
p
p
.
4
3
9
-
4
4
6
,
2
0
0
9
.
[6
]
P
e
ra
k
is
P
.
,
Th
e
o
h
a
ris
T.
,
P
a
ss
a
li
s
G
.
,
Ka
k
a
d
iaris
I.
,
“
Au
to
m
a
ti
c
3D
fa
c
ial
re
g
io
n
re
tri
e
v
a
l
fr
o
m
m
u
lt
i
-
po
se
fa
c
ial
d
a
tas
e
ts
,
In
Pro
c
.
Eu
r
o
g
ra
p
h
ics
W
o
rk
sh
o
p
o
n
3D
Ob
jec
t
Retrie
v
a
l
,
p
p
.
3
7
-
4
4
,
2
0
0
9
.
[7
]
Xie
W,
S
h
e
n
L
,
Ya
n
g
M
,
Lai
Z.
,
“
Ac
ti
v
e
AU
b
a
se
d
p
a
tch
we
ig
h
ti
n
g
f
o
r
fa
c
ial
e
x
p
re
ss
io
n
re
c
o
g
n
it
io
n
,
”
S
e
n
so
rs
(Ba
se
l)
,
v
o
l.
1
7
,
p
p
.
2
7
5
,
2
0
1
7
.
[8
]
S
.
Be
rre
tt
i,
A.
De
l
Bim
b
o
,
P
.
P
a
la
,
“
S
p
a
rse
m
a
tch
in
g
o
f
sa
li
e
n
t
fa
c
ial
c
u
rv
e
s
fo
r
re
c
o
g
n
it
i
o
n
o
f
3D
fa
c
e
s
with
m
issin
g
p
a
rts,
”
IE
EE
T
r
a
n
sa
c
t
io
n
s o
n
I
n
fo
rm
a
ti
o
n
F
o
re
n
sic
s a
n
d
S
e
c
u
rity
,
v
o
l
.
8
,
n
o
.
2
,
p
p
.
3
7
4
-
3
8
9
,
2
0
1
3
.
[9
]
An
d
re
w
J.L
o
g
a
n
,
G
a
e
l
E.
G
o
rd
o
n
,
G
u
n
terLo
ffler,
“
Co
n
tri
b
u
t
io
n
s
o
f
in
d
iv
i
d
u
a
l
fa
c
e
fe
a
tu
re
s
to
fa
c
e
d
isc
rimin
a
ti
o
n
,”
Vi
sio
n
Res
e
a
rc
h
,
v
o
l.
1
3
7
,
p
p
.
29
-
3
9
,
2
0
1
7
.
[1
0
]
S
.
Ra
v
i
,
S
.
Wi
ls
o
n
,
“
F
a
c
e
d
e
tec
ti
o
n
wit
h
fa
c
ial
fe
a
tu
re
s
a
n
d
g
e
n
d
e
r
c
las
sifica
ti
o
n
b
a
se
d
o
n
su
p
p
o
rt
v
e
c
to
r
m
a
c
h
in
e
,”
2
0
1
0
IEE
E
I
n
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
C
o
mp
u
ta
t
io
n
a
l
In
telli
g
e
n
c
e
a
n
d
c
o
m
p
u
ti
n
g
R
e
se
a
rc
h
,
IS
BN:
9
7
8
8
1
8
3
7
1
3
6
2
7
,
2
0
1
0
[1
1
]
T.
Ve
rm
a
a
n
d
R.
K.
S
a
h
u
,
“
P
CA
-
LDA
b
a
se
d
fa
c
e
r
e
c
o
g
n
it
io
n
sy
ste
m
&
re
su
lt
s
c
o
m
p
a
riso
n
b
y
v
a
rio
u
s
c
las
sifica
ti
o
n
tec
h
n
i
q
u
e
s,
”
2
0
1
3
In
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
Gr
e
e
n
Hig
h
Per
fo
rm
a
n
c
e
Co
mp
u
ti
n
g
(ICGH
PC)
,
Na
g
e
rc
o
il
,
p
p
.
1
-
7
,
2
0
1
3
.
[1
2
]
Zh
o
u
,
S
.
,
Xia
o
,
S
.
“
3D
fa
c
e
re
c
o
g
n
it
io
n
:
a
s
u
rv
e
y
.
”
Hu
m.
Ce
n
t.
C
o
mp
u
t.
In
f.
S
c
i
.
v
o
l.
8
,
p
p
.
3
5
,
2
0
1
8
.
[1
3
]
Bo
ra
d
e
S
.
N.,
De
sh
m
u
k
h
R.
R.
,
S
h
rish
rima
l
P
.
,
“
Eff
e
c
t
o
f
d
istan
c
e
m
e
a
su
re
s
o
n
th
e
p
e
rf
o
rm
a
n
c
e
o
f
f
a
c
e
re
c
o
g
n
it
io
n
u
sin
g
p
ri
n
c
ip
a
l
c
o
m
p
o
n
e
n
t
a
n
a
ly
si
s
,”
In
:
Be
rre
tt
i
S
.
,
T
h
a
m
p
i
S
.
,
S
r
iv
a
sta
v
a
P
.
(e
d
s)
“
In
tel
li
g
e
n
t
s
y
ste
m
s
tec
h
n
o
l
o
g
ies
a
n
d
a
p
p
li
c
a
ti
o
n
s
,”
A
d
v
a
n
c
e
s in
In
telli
g
e
n
t
S
y
ste
ms
a
n
d
Co
m
p
u
ti
n
g
,
v
o
l.
3
8
4
,
2
0
1
6
.
[1
4
]
S
.
M
.
S
.
Isla
m
,
R.
Da
v
ies
,
M
.
Be
n
n
a
m
o
u
n
,
R
.
"
Ow
e
n
s
a
n
d
Ajm
a
l
M
ian
,
“
M
u
lt
ib
i
o
m
e
tri
c
h
u
m
a
n
re
c
o
g
n
i
ti
o
n
u
sin
g
3D
e
a
r
a
n
d
fa
c
e
fe
a
tu
re
s
,”
Pa
tt
e
r
n
Rec
o
g
n
it
i
o
n
,
v
o
l.
4
6
,
n
o
.
3
,
p
p
.
6
1
3
-
6
2
7
,
2
0
1
3
.
[1
5
]
X.Z
h
a
o
,
W.
Zh
a
n
g
,
G
.
Ev
a
n
g
e
lo
p
o
u
l
o
s,
D.Hu
a
n
g
,
S
.
S
h
a
h
,
Y.
Wan
g
,
I
.
Ka
k
a
d
iaris
a
n
d
L.
C
h
e
n
,
“
Be
n
c
h
m
a
rk
in
g
a
sy
m
m
e
tri
c
3D
-
2
D
fa
c
e
re
c
o
g
n
it
i
o
n
sy
ste
m
s,
”
Pro
c
.
3D
Fa
c
e
Bi
o
me
trics
W
o
rk
sh
o
p
,
in
c
o
n
j
u
c
ti
o
n
wit
h
IEE
E
In
t’l
Co
n
f.
o
n
A
u
to
m
a
ti
c
Fa
c
e
a
n
d
Ge
stu
re
Rec
o
g
n
it
io
n
(FG)
,
2
0
1
3
.
[1
6
]
Yu
e
m
in
g
Wan
g
,
G
a
n
g
P
a
n
,
Jia
n
z
h
u
a
n
Li
u
,
“
A
d
e
fo
rm
a
ti
o
n
m
o
d
e
l
to
re
d
u
c
e
th
e
e
ffe
c
t
o
f
e
x
p
re
ss
io
n
s
i
n
3D
fa
c
e
re
c
o
g
n
it
i
o
n
,
”
Vi
su
a
l
C
o
mp
u
ter
,
v
o
l.
27
,
n
o
.
5
,
p
p
.
3
3
3
-
3
4
5
,
2
0
1
1
.
[1
7
]
A.
Watt,
M
.
Watt,
A
d
v
a
n
c
e
d
A
n
ima
ti
o
n
,
a
n
d
Ren
d
e
rin
g
T
e
c
h
n
iq
u
e
s:
T
h
e
o
ry
a
n
d
Pr
a
c
ti
c
e
,
Ad
d
iso
n
-
Wes
ley
,
Re
a
d
in
g
,
M
A,
1
9
9
2
.
[1
8
]
A.
M
o
re
n
o
,
A.
S
a
n
c
h
e
z
,
J.
Ve
lez
,
F
.
Dia
z
,
“
F
a
c
e
re
c
o
g
n
it
io
n
u
si
n
g
3D
su
rfa
c
e
e
x
trac
ted
d
e
sc
rip
to
rs,
”
P
ro
c
e
e
d
in
g
s
o
f
t
h
e
Irish
M
a
c
h
i
n
e
Vi
si
o
n
a
n
d
I
ma
g
e
Pro
c
e
ss
in
g
,
2
0
0
4
.
[1
9
]
Na
e
e
m
Iq
b
a
l
Ra
ty
a
l,
Im
ti
a
z
Ah
m
a
d
Taj,
M
u
h
a
m
m
a
d
S
a
ji
d
,
No
u
m
a
n
Ali,
An
z
a
r
M
a
h
m
o
o
d
a
n
d
S
o
h
a
il
Ra
z
z
a
q
,
“
Th
re
e
-
d
ime
n
sio
n
a
l
fa
c
e
re
c
o
g
n
it
io
n
u
si
n
g
v
a
rian
c
e
-
b
a
se
d
r
e
g
istratio
n
a
n
d
su
b
jec
t
-
sp
e
c
ifi
c
d
e
sc
rip
to
rs,
”
In
ter
n
a
t
io
n
a
l
J
o
u
rn
a
l
o
f
A
d
v
a
n
c
e
d
Ro
b
o
ti
c
S
y
ste
ms
,
p
p
.
1
-
1
6
,
2
0
1
9
.
[2
0
]
M
il
ler,
P
.
,
a
n
d
L
y
le,
J.
T
h
e
e
ff
e
c
t
o
f
d
ist
a
n
c
e
me
a
su
re
s
o
n
th
e
re
c
o
g
n
it
i
o
n
ra
tes
o
f
PCA
a
n
d
L
D
A
b
a
se
d
fa
c
i
a
l
re
c
o
g
n
it
i
o
n
T
e
c
h
.
re
p
.
,
Clem
so
n
Un
iv
e
rsity
,
2
0
0
8
.
[2
1
]
Blan
z
V,
S
c
h
e
r
b
a
u
m
K,
S
e
id
e
l
HP,
“
F
it
ti
n
g
a
m
o
r
p
h
a
b
le
m
o
d
e
l
to
3D
sc
a
n
s
o
f
fa
c
e
s
,”
In
:
C
o
mp
u
ter
v
isio
n
,
p
p
.
1
-
8
,
2
0
0
7
.
[2
2
]
Dib
e
k
li
o
g
l
u
H,
S
a
lah
AA
,
A
k
a
ru
n
L
.
,
“
3D
fa
c
ial
lan
d
m
a
rk
i
n
g
u
n
d
e
r
e
x
p
re
ss
io
n
,
p
o
se
a
n
d
o
c
c
lu
si
o
n
v
a
riati
o
n
s
,”
I
n
:
Bi
o
me
trics
th
e
o
ry
,
a
p
p
li
c
a
t
io
n
s
a
n
d
sy
ste
ms
,
p
p
.
1
-
6
,
2
0
0
8
.
[2
3
]
M
a
h
m
o
o
d
S
A,
G
h
a
n
i
RF
,
Ke
ri
m
AA
.
,
“
3D
fa
c
e
re
c
o
g
n
it
i
o
n
u
sin
g
p
o
se
i
n
v
a
rian
t
n
o
se
re
g
io
n
d
e
tec
to
r
,”
In
:
Co
mp
u
ter
sc
ien
c
e
a
n
d
e
lec
tro
n
ic
e
n
g
in
e
e
rin
g
c
o
n
fer
e
n
c
e
,
2
0
1
4
.
[2
4
]
Be
rre
tt
i
S
,
De
l
Bim
b
o
P
P
A.
,
“
S
p
a
rse
m
a
tch
in
g
o
f
sa
li
e
n
t
fa
c
ial
c
u
rv
e
s
fo
r
re
c
o
g
n
it
io
n
o
f
3D
fa
c
e
s
with
m
issin
g
p
a
rts
,”
Fo
re
n
sic
s
S
e
c
u
,
v
o
l.
8
,
p
p
.
3
7
4
-
3
8
9
,
2
0
1
3
.
[2
5
]
P
e
ra
k
is
P
,
P
a
ss
a
li
s
G
,
Th
e
o
h
a
ris
T,
To
d
e
rici
G
,
Ka
k
a
d
iaris
IA.
,
“
P
a
rti
a
l
m
a
tch
in
g
o
f
in
terp
o
se
3D
fa
c
ial
d
a
ta
fo
r
fa
c
e
re
c
o
g
n
it
io
n
,”
I
n
:
Bi
o
me
trics
:
th
e
o
ry
,
a
p
p
li
c
a
ti
o
n
s
,
a
n
d
sy
ste
ms
,
p
p
.
4
3
9
-
4
4
6
,
2
0
0
9
.
[2
6
]
Hu
a
WG
.
,
“
Im
p
li
c
it
e
las
ti
c
m
a
tch
in
g
wit
h
ra
n
d
o
m
p
r
o
jec
ti
o
n
s
fo
r
p
o
se
-
v
a
rian
t
fa
c
e
re
c
o
g
n
it
i
o
n
,”
I
n
:
C
o
mp
u
t.
Vi
s
.
Pa
tt
e
rn
Rec
o
g
n
it
,
p
p
.
1
5
0
2
-
1
5
0
9
,
2
0
0
9
.
Evaluation Warning : The document was created with Spire.PDF for Python.