I
AE
S
I
nte
rna
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
5
, N
o.
2
,
J
u
n
e
2
0
1
6
,
p
p
.
8
0
~8
8
I
SS
N:
2252
-
8938
80
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
J
AI
Dev
elo
p
m
e
nt
o
f
an Effic
ient
F
a
ce
Recog
nition Sy
ste
m
Ba
sed
on
Linea
r and
No
nli
nea
r
Alg
o
rith
m
s
Ara
o
luw
a
Si
m
ileo
lu F
ila
ni
,
Adeba
y
o
O
lus
o
la
Adet
un
m
b
i
F
e
d
e
ra
l
Un
iv
e
rsity
o
f
T
e
c
h
n
o
lo
g
y
,
Ak
u
re
,
On
d
o
S
tate
,
Nig
e
ria
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
r
6
,
2
0
1
6
R
ev
i
s
ed
Ma
y
10
,
2
0
1
6
A
cc
ep
ted
Ma
y
2
7
,
2
0
1
6
T
h
is
p
a
p
e
r
p
re
se
n
ts
a
p
p
e
a
ra
n
c
e
b
a
se
d
m
e
th
o
d
s
f
o
r
f
a
c
e
re
c
o
g
n
it
io
n
u
si
n
g
li
n
e
a
r
a
n
d
n
o
n
li
n
e
a
r
tec
h
n
i
q
u
e
s.
T
h
e
li
n
e
a
r
a
lg
o
rit
h
m
s
u
se
d
a
re
P
rin
c
i
p
a
l
Co
m
p
o
n
e
n
t
A
n
a
l
y
sis
(P
CA
)
a
n
d
L
in
e
a
r
Dis
c
ri
m
in
a
n
t
A
n
a
l
y
sis
(
L
D
A
).
T
h
e
tw
o
n
o
n
li
n
e
a
r
m
e
th
o
d
s
u
se
d
a
re
th
e
Ke
rn
e
l
P
ri
n
c
ip
a
l
Co
m
p
o
n
e
n
t
s
A
n
a
l
y
sis
(KP
CA
)
a
n
d
Ke
rn
e
l
F
ish
e
r
A
n
a
l
y
si
s
(KFA
).
T
h
e
li
n
e
a
r
d
im
e
n
sio
n
a
l
re
d
u
c
ti
o
n
p
ro
jec
ti
o
n
m
e
th
o
d
s
e
n
c
o
d
e
p
a
tt
e
rn
in
f
o
rm
a
ti
o
n
b
a
se
d
o
n
se
c
o
n
d
o
rd
e
r
d
e
p
e
n
d
e
n
c
ies
.
T
h
e
n
o
n
li
n
e
a
r
m
e
th
o
d
s
a
re
u
se
d
to
h
a
n
d
le
re
latio
n
sh
ip
s
a
m
o
n
g
th
re
e
o
r
m
o
re
p
ix
e
ls.
In
th
e
f
in
a
l
sta
g
e
,
M
a
h
a
li
n
o
b
is
Co
sin
e
(
M
AHC
OS
)
m
e
tri
c
is
u
se
d
to
d
e
f
in
e
th
e
si
m
il
a
rit
y
m
e
a
su
re
b
e
tw
e
e
n
tw
o
im
a
g
e
s
a
f
ter
th
e
y
h
a
v
e
p
a
ss
e
d
th
ro
u
g
h
t
h
e
c
o
rre
sp
o
n
d
i
n
g
d
i
m
e
n
sio
n
a
l
re
d
u
c
ti
o
n
tec
h
n
iq
u
e
s.
T
h
e
e
x
p
e
rime
n
t
sh
o
w
e
d
th
a
t
L
DA
a
n
d
KFA
h
a
v
e
th
e
h
ig
h
e
st
p
e
rf
o
r
m
a
n
c
e
o
f
9
3
.
3
3
%
f
ro
m
th
e
CM
C
a
n
d
ROC
r
e
su
lt
s
w
h
e
n
u
se
d
w
it
h
G
a
b
o
r
w
a
v
e
lets
.
T
h
e
o
v
e
ra
ll
re
su
lt
u
sin
g
4
0
0
im
a
g
e
s
o
f
AT
&
T
d
a
t
a
b
a
se
sh
o
we
d
th
a
t
th
e
p
e
rf
o
rm
a
n
c
e
o
f
th
e
li
n
e
a
r
a
n
d
n
o
n
li
n
e
a
r
a
lg
o
rit
h
m
s
c
a
n
b
e
a
ffe
c
ted
b
y
th
e
n
u
m
b
e
r
o
f
c
la
ss
e
s
o
f
th
e
ima
g
e
s,
p
re
p
ro
c
e
ss
in
g
o
f
i
m
a
g
e
s,
a
n
d
t
h
e
n
u
m
b
e
r
o
f
f
a
c
e
i
m
a
g
e
s o
f
th
e
tes
t
se
ts
u
se
d
f
o
r
re
c
o
g
n
it
i
o
n
.
K
ey
w
o
r
d
:
A
p
p
ea
r
an
ce
-
B
ased
Me
th
o
d
s
D
i
m
e
n
s
io
n
al
R
ed
u
ctio
n
T
ec
h
n
iq
u
es
Face
R
ec
o
g
n
itio
n
Gab
o
r
W
av
elets
L
i
n
ea
r
An
d
No
n
li
n
ea
r
Face
R
ec
o
g
n
itio
n
Alg
o
r
it
h
m
s
Co
p
y
rig
h
t
©
2
0
1
6
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
:
A
r
ao
lu
w
a
Si
m
ileo
l
u
Fil
a
n
i
,
T
ea
ch
in
g
A
s
s
i
s
tan
t,
Dep
ar
t
m
e
n
t o
f
C
o
m
p
u
ter
Sc
ie
n
ce
,
Fed
er
al
Un
iv
er
s
it
y
o
f
T
ec
h
n
o
l
o
g
y
,
Ak
u
r
e,
On
d
o
State,
Ni
g
er
ia
.
E
m
ail:
a
f
ar
f
r
ee
3
4
0
@
g
m
ai
l.c
o
m
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
d
ev
elo
p
m
e
n
t
o
f
t
h
e
r
o
b
u
s
t
r
ec
o
g
n
itio
n
s
y
s
te
m
i
s
a
v
er
y
i
m
p
o
r
tan
t
ar
ea
u
n
d
er
d
is
cu
s
s
i
o
n
b
ec
au
s
e
o
f
th
e
w
id
e
r
an
g
e
o
f
ap
p
licati
o
n
s
i
n
d
if
f
er
e
n
t
s
p
h
er
e
s
,
s
u
ch
as
v
id
eo
s
u
r
v
ei
llan
ce
s
ec
u
r
i
t
y
s
y
s
te
m
s
,
co
n
tr
o
l
o
f
d
o
cu
m
en
ts
,
f
o
r
en
s
ics
s
y
s
te
m
s
etc.
[
1
]
.
I
n
th
e
la
s
t
t
h
r
ee
d
ec
ad
es
f
ac
e
id
en
ti
f
icat
io
n
p
r
o
b
le
m
h
as
e
m
er
g
ed
as
a
n
i
m
p
o
r
tan
t
r
esear
c
h
ar
ea
w
it
h
m
an
y
p
o
s
s
ib
le
ap
p
licatio
n
s
th
at
u
n
d
o
u
b
ted
l
y
alle
v
iate
a
n
d
ass
is
t
s
a
f
e
g
u
ar
d
o
u
r
ev
er
y
d
a
y
li
v
es
i
n
m
a
n
y
a
s
p
ec
t
s
[
2
]
.
Un
lik
e
m
a
n
y
o
t
h
er
id
en
t
if
icatio
n
m
et
h
o
d
s
,
f
ac
e
r
ec
o
g
n
itio
n
d
o
es
n
o
t
n
ee
d
to
m
ak
e
d
ir
ec
t c
o
n
tact
w
it
h
a
n
in
d
iv
id
u
al
i
n
o
r
d
er
to
v
alid
ate
th
eir
id
en
t
it
y
.
T
h
is
ca
n
b
e
u
s
e
f
u
l f
o
r
s
u
r
v
ei
llan
ce
o
r
tr
ac
k
in
g
a
n
d
in
d
etec
tio
n
s
y
s
te
m
s
.
Data
ac
q
u
i
s
itio
n
i
n
g
e
n
er
al
is
f
r
au
g
h
t
w
it
h
p
r
o
b
lem
s
f
o
r
o
th
er
b
io
m
etr
ic
s
: te
c
h
n
iq
u
e
s
t
h
at
r
el
y
o
n
h
a
n
d
s
a
n
d
f
in
g
er
s
ca
n
b
e
r
en
d
er
ed
u
s
eles
s
i
f
t
h
e
ep
id
er
m
is
tis
s
u
e
is
i
n
j
u
r
ed
in
s
o
m
e
w
a
y
(
i.e
.
,
b
r
u
is
ed
o
r
cr
ac
k
ed
)
.
A
lt
h
o
u
g
h
t
h
er
e
ar
e
a
n
u
m
b
er
o
f
f
ac
e
r
ec
o
g
n
itio
n
al
g
o
r
ith
m
s
w
h
ic
h
w
o
r
k
w
el
l
in
co
n
s
tr
ai
n
ed
en
v
ir
o
n
m
en
ts
,
f
ac
e
r
ec
o
g
n
itio
n
is
s
till
a
n
o
p
en
an
d
d
if
f
ic
u
lt
p
r
o
b
lem
i
n
r
ea
l
ap
p
licatio
n
s
[
3
]
.
I
n
f
ac
e
r
ec
o
g
n
itio
n
,
ap
p
ea
r
an
ce
-
b
ased
ap
p
r
o
ac
h
h
a
s
b
ee
n
w
id
el
y
u
s
ed
[
4
]
.
T
h
e
Face
i
m
ag
e
s
o
f
te
n
h
av
e
a
lar
g
e
n
u
m
b
er
o
f
p
ix
el
v
al
u
es
an
d
ar
e
r
ep
r
esen
ted
a
s
h
ig
h
-
d
i
m
e
n
s
io
n
al
v
ec
to
r
s
o
r
ar
r
a
y
s
.
Op
er
ati
n
g
d
ir
ec
tl
y
o
n
t
h
ese
v
ec
to
r
s
i
s
i
n
ef
f
icien
t,
w
o
u
ld
lead
to
h
ig
h
co
m
p
u
tatio
n
al
co
s
t
s
an
d
s
to
r
ag
e
s
p
ac
e,
an
d
p
o
s
es
th
e
c
u
r
s
e
o
f
d
i
m
e
n
s
io
n
alit
y
to
m
a
n
y
lear
n
i
n
g
tas
k
s
.
A
u
s
ef
u
l
s
u
b
s
p
ac
e
r
ep
r
ese
n
tatio
n
h
as
th
u
s
b
ec
o
m
e
d
esira
b
le
in
m
a
n
y
i
m
a
g
e
p
r
o
ce
s
s
i
n
g
ap
p
licatio
n
s
.
T
h
e
h
o
li
s
tic
an
d
co
m
p
o
n
e
n
t
-
b
a
s
ed
m
et
h
o
d
s
ar
e
t
w
o
m
ai
n
w
a
y
s
f
o
r
r
ep
r
esen
tin
g
th
e
f
ac
ial
ap
p
ea
r
an
ce
s
.
C
o
m
p
ar
ed
to
h
o
lis
tic
ap
p
r
o
ac
h
es,
f
ea
t
u
r
e
-
b
ased
m
et
h
o
d
s
ar
e
less
s
e
n
s
it
iv
e
to
v
ar
iatio
n
s
in
illu
m
in
a
tio
n
an
d
v
ie
w
p
o
i
n
t
[
2
]
.
I
n
h
o
lis
tic
r
ep
r
esen
tatio
n
,
a
f
ac
ial
i
m
ag
e
i
s
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
D
ev
elo
p
men
t o
f a
n
E
fficien
t
F
a
ce
R
ec
o
g
n
itio
n
S
ystem
B
a
s
ed
o
n
Lin
ea
r
…
(
A
r
a
o
lu
w
a
S
imileo
lu
F
ila
n
i
)
81
co
n
s
id
er
ed
as
a
v
ec
to
r
o
f
p
ix
e
ls
a
n
d
is
r
ep
r
esen
ted
as
a
s
in
g
le
p
o
in
t
i
n
t
h
e
h
i
g
h
d
i
m
en
s
io
n
al
s
p
ac
e.
S
u
b
s
p
ac
e
me
t
h
o
d
s
ar
e
th
e
n
ap
p
lied
to
r
e
d
u
ce
th
e
h
i
g
h
-
d
i
m
en
s
io
n
al
d
at
a
o
n
to
a
lo
w
er
d
i
m
e
n
s
io
n
al
s
p
ac
e
w
h
ile
r
etai
n
i
n
g
in
tr
i
n
s
ic
f
ea
tu
r
e
s
f
o
r
f
u
r
th
er
cl
ass
i
f
icatio
n
[
5
]
,
[
6
]
.
T
h
e
h
o
lis
tic
ap
p
ea
r
an
ce
b
ased
m
eth
o
d
s
ca
n
b
e
class
i
f
ied
as
lin
ea
r
m
eth
o
d
s
a
n
d
n
o
n
li
n
ea
r
m
et
h
o
d
s
.
L
i
n
ea
r
m
eth
o
d
s
e
x
p
licitl
y
tr
an
s
f
o
r
m
d
ata
f
r
o
m
h
i
g
h
d
i
m
e
n
s
io
n
a
l
s
u
b
s
p
ac
e
i
n
to
lo
w
d
i
m
e
n
s
io
n
a
l
s
u
b
s
p
ac
e
b
y
l
in
ea
r
m
ap
p
in
g
.
I
n
n
o
n
li
n
ea
r
tec
h
n
iq
u
e
s
,
ex
p
l
icit
p
r
o
j
ec
tio
n
s
ar
e
n
o
t
d
o
n
e.
I
n
s
tead
f
a
ith
f
u
l
lo
w
d
i
m
en
s
io
n
al
d
ata
m
atr
i
x
i
s
o
b
tain
ed
d
ir
ec
tl
y
f
r
o
m
h
i
g
h
d
i
m
e
n
s
io
n
al
d
ata
m
atr
i
x
.
T
h
e
s
u
cc
ess
f
u
l
f
ac
e
li
n
ea
r
o
r
n
o
n
lin
ea
r
m
et
h
o
d
s
u
s
e
d
d
ep
en
d
s
h
ea
v
il
y
o
n
th
e
p
ar
ti
cu
lar
ch
o
ice
o
f
th
e
f
ea
t
u
r
es
u
s
ed
b
y
th
e
p
atter
n
c
lass
i
f
ier
.
T
h
er
ef
o
r
e,
d
etailed
ev
alu
a
tio
n
a
n
d
b
en
c
h
m
ar
k
i
n
g
o
f
th
e
alg
o
r
it
h
m
s
i
s
cr
u
cial
f
o
r
late
r
u
s
e.
A
p
p
ea
r
an
ce
f
ac
e
r
ec
o
g
n
it
io
n
m
et
h
o
d
s
d
o
n
o
t
p
er
f
o
r
m
w
el
l
d
u
r
in
g
i
ll
co
n
d
itio
n
s
[
2
]
,
[
7
]
,
ev
en
t
h
e
m
o
s
t
r
ep
r
esen
tati
v
e
r
ec
o
g
n
iti
o
n
tech
n
iq
u
e
s
f
r
eq
u
en
tl
y
u
s
e
d
in
co
n
j
u
n
c
tio
n
w
i
th
f
ac
e
r
e
co
g
n
itio
n
co
u
ld
n
o
t
ac
h
iev
e
b
est
r
es
u
lt
[
7
]
,
[
8
]
.
O
n
e
o
f
th
e
m
o
s
t
s
u
cc
es
s
f
u
l
clas
s
if
ier
s
th
at
h
as
b
ee
n
u
s
ed
f
o
r
im
ag
e
r
ep
r
esen
tatio
n
is
Gab
o
r
W
av
elets.
T
h
is
is
b
e
ca
u
s
e
i
t
is
a
v
er
y
s
tr
o
n
g
p
r
ep
r
o
ce
s
s
in
g
a
n
d
ex
tr
ac
tio
n
al
g
o
r
ith
m
s
[
9
]
,
[
1
0
]
,
[
1
1
]
.
I
n
th
i
s
r
eg
ar
d
Gab
o
r
W
av
elet
s
is
c
h
o
s
en
f
o
r
th
is
w
o
r
k
to
p
r
o
v
id
e
r
o
b
u
s
t
f
ac
e
r
ec
o
g
n
iti
o
n
alg
o
r
it
h
m
s
.
T
h
e
r
e
m
ain
in
g
p
ar
ts
o
f
th
is
p
ap
er
ar
e
o
r
g
an
ized
as
f
o
llo
w
s
:
Sectio
n
I
I
g
iv
e
s
a
r
ev
ie
w
o
f
f
ac
e
r
ec
o
g
n
itio
n
m
et
h
o
d
o
lo
g
ies.
Sectio
n
I
I
I
d
escr
ib
es
th
e
m
et
h
o
d
o
lo
g
y
o
f
t
h
i
s
w
o
r
k
.
Sectio
n
I
V
p
r
esen
ts
t
h
e
ex
p
er
i
m
e
n
ts
a
n
d
th
e
r
esu
lts
.
Fi
n
a
ll
y
,
th
e
co
n
cl
u
s
io
n
o
f
t
h
e
w
o
r
k
is
d
r
a
w
n
i
n
s
ec
tio
n
V.
2.
RE
L
AT
E
D
S
T
UDI
E
S
A
m
aj
o
r
is
s
u
e
o
f
f
ac
e
r
ec
o
g
n
itio
n
is
h
o
w
to
i
m
p
r
o
v
e
t
h
e
o
v
er
all
p
er
f
o
r
m
a
n
ce
o
f
th
e
e
m
p
lo
y
ed
r
ec
o
g
n
itio
n
tech
n
iq
u
es
[
1
2
]
,
[
1
3
]
.
Mo
s
t
o
f
th
e
p
r
ev
io
u
s
m
et
h
o
d
s
w
er
e
m
a
in
l
y
f
o
cu
s
ed
o
n
f
r
o
n
tal
f
ac
e
i
m
a
g
es
o
r
s
in
g
le
-
v
ie
w
-
b
ased
f
ac
e
r
ec
o
g
n
itio
n
.
T
h
e
p
r
o
b
lem
w
it
h
t
h
ese
ea
r
l
y
s
o
lu
t
io
n
s
w
as
t
h
e
m
an
u
al
co
m
p
u
tatio
n
s
o
f
f
ea
t
u
r
es
m
ea
s
u
r
e
m
e
n
t
s
an
d
lo
ca
tio
n
s
[
1
4
]
.
T
h
e
n
o
tab
le
ea
r
lies
t
ap
p
r
o
ac
h
es
in
is
t
h
e
E
i
g
en
f
ac
es
[
2
]
.
T
h
e
eig
en
f
ac
e
s
tech
iq
u
es
w
a
s
d
ev
elo
p
ed
b
y
Siro
v
ic
h
an
d
Kir
b
y
(
1
9
8
7
)
an
d
u
s
ed
b
y
Ma
tt
h
e
w
T
u
r
k
a
n
d
Alex
P
en
tlan
d
in
f
ac
e
class
i
f
icat
io
n
[
1
3
]
,
[
1
5
]
b
y
u
s
in
g
s
ta
n
d
ar
d
lin
ea
r
alg
eb
r
a
tech
n
iq
u
e.
An
N
×
N
i
m
a
g
e
I
is
lin
ea
r
ized
in
a
2
v
ec
to
r
,
s
o
th
a
t
it
r
ep
r
esen
ts
a
p
o
in
t
i
n
a
2
-
d
i
m
e
n
s
io
n
al
s
p
ac
e.
R
ec
o
g
n
i
tio
n
o
f
a
p
r
o
b
e
i
m
a
g
e
i
s
p
er
f
o
r
m
ed
i
n
a
lo
w
e
r
d
i
m
en
s
io
n
al
s
p
ac
e
b
y
m
ea
n
s
o
f
a
d
i
m
e
n
s
io
n
alit
y
r
ed
u
c
ti
o
n
tech
n
iq
u
e
u
s
i
n
g
P
C
A
(
P
r
in
cip
al
C
o
m
p
o
n
e
n
t
An
al
y
s
i
s
)
.
Af
ter
t
h
e
li
n
ea
r
iza
tio
n
t
h
e
m
ea
n
v
ec
to
r
is
ca
lcu
la
te
d
.
T
h
e
co
v
ar
ian
ce
m
atr
i
x
is
th
e
n
co
m
p
u
ted
,
in
o
r
d
er
to
ex
tr
ac
t
a
lim
ited
n
u
m
b
er
o
f
it
s
eig
e
n
v
ec
to
r
s
,
co
r
r
esp
o
n
d
in
g
to
th
e
g
r
ea
test
ei
g
e
n
v
al
u
e
s
ca
lled
ei
g
en
f
ac
es.
As
t
h
e
P
C
A
i
s
p
er
f
o
r
m
ed
o
n
l
y
f
o
r
tr
ain
i
n
g
th
e
s
y
s
te
m
,
t
h
i
s
te
c
h
n
iq
u
e
ap
p
ea
r
s
to
b
e
v
er
y
f
a
s
t
w
h
e
n
te
s
ti
n
g
n
e
w
f
ac
e
i
m
a
g
es.
T
h
e
P
C
A
h
a
s
b
ee
n
i
n
ten
s
i
v
e
l
y
e
x
p
lo
ited
i
n
f
ac
e
r
ec
o
g
n
itio
n
ap
p
licatio
n
s
,
an
d
m
an
y
o
f
i
ts
v
ar
ia
tio
n
s
h
a
v
e
b
ee
n
d
ev
elo
p
ed
.
Ma
n
y
o
th
e
r
lin
ea
r
p
r
o
j
ec
tio
n
m
et
h
o
d
s
t
h
at
p
er
f
o
r
m
ed
b
ette
r
u
n
d
er
s
o
m
e
co
n
d
itio
n
s
h
a
v
e
b
ee
n
s
tu
d
ied
.
T
h
e
L
D
A
(
L
in
ea
r
Di
s
cr
i
m
in
a
n
t
An
al
y
s
i
s
)
[
6
]
h
as
b
ee
n
d
ev
elo
p
ed
as
a
b
etter
tech
n
iq
u
e
t
h
a
n
P
C
A
.
W
h
en
co
m
p
ar
ed
w
ith
P
C
A
,
L
D
A
g
i
v
e
s
a
h
ig
h
er
r
ec
o
g
n
itio
n
r
ate
w
h
en
a
w
id
e
tr
ain
i
n
g
s
et
is
a
v
ailab
l
e.
T
o
p
r
o
v
id
e
a
s
tr
o
n
g
er
s
y
s
t
e
m
,
P
C
A
h
a
s
b
ee
n
co
m
b
i
n
ed
w
i
th
L
D
A
[
1
6
]
b
u
t
it
h
as
b
ee
n
s
h
o
w
n
in
[
1
7
]
th
at,
co
m
b
in
i
n
g
P
C
A
an
d
L
DA
,
ca
n
n
o
t
al
w
a
y
s
p
r
o
d
u
ce
d
d
esire
d
r
esu
lt.
I
C
A
w
a
s
in
tr
o
d
u
ce
d
f
o
r
p
r
o
v
id
in
g
f
ac
e
r
ep
r
esen
tatio
n
s
w
i
th
h
i
g
h
-
o
r
d
er
d
ep
en
d
en
cies
th
at
ar
e
s
ep
ar
ated
in
to
in
d
i
v
id
u
al
co
ef
f
icie
n
ts
a
n
d
w
as e
x
p
ec
ted
to
g
iv
e
s
u
p
er
io
r
r
ec
o
g
n
itio
n
p
er
f
o
r
m
a
n
ce
th
a
n
P
C
A
w
h
ic
h
o
n
l
y
d
ep
en
d
o
n
s
ep
ar
ate
s
ec
o
n
d
-
o
r
d
er
r
ed
u
n
d
a
n
cies
[
1
8
]
.
Af
ter
w
ar
d
s
,
I
C
A
th
eo
r
y
w
a
s
co
n
tr
ad
icto
r
y
,
a
n
d
it
h
as
b
ee
n
s
h
o
w
n
t
h
at
I
C
A
d
o
es
n
o
t
al
w
a
y
s
p
er
f
o
r
m
b
etter
th
e
n
P
C
A
o
r
j
u
s
t
s
u
i
tab
le
f
o
r
a
s
p
ec
if
ic
tas
k
[
1
9
]
,
[
2
0
]
.
T
o
o
v
er
co
m
e
s
o
m
e
o
f
th
e
li
m
itat
io
n
s
o
f
th
e
m
e
n
tio
n
ed
,
o
th
er
h
y
b
r
id
s
o
f
P
C
A
,
L
D
A
an
d
I
C
A
al
g
o
r
ith
m
s
w
er
e
d
ev
elp
ed
.
Mo
s
t
o
f
th
ese
n
e
w
er
tech
n
iq
u
es
i
n
v
o
l
v
e
co
m
b
i
n
ati
o
n
o
f
o
n
e
o
r
m
o
r
e
alg
o
r
ith
m
s
[
8
]
,
[
2
1
]
,
[
6
]
.
As
a
P
C
A
a
n
d
L
D
A
f
a
il
to
d
is
co
v
er
t
h
e
u
n
d
er
l
y
i
n
g
s
tr
u
ct
u
r
e
o
f
f
ac
e
i
m
a
g
es
th
at
lie
o
n
a
h
id
d
e
n
n
o
n
li
n
ea
r
s
u
b
m
a
n
i
f
o
ld
,
a
lap
lacia
n
f
ac
e
ap
p
r
o
ac
h
w
as
p
r
o
p
o
s
ed
in
[
2
2
]
to
p
r
o
v
id
e
a
m
e
th
o
d
th
at
co
u
ld
d
etec
t
th
e
u
n
d
er
l
y
in
g
s
tr
u
ct
u
r
e
o
f
f
a
ce
s
th
at
lie
o
n
a
h
id
d
en
n
o
n
li
n
ea
r
s
u
b
m
ai
n
f
o
ld
th
at
P
C
A
a
n
d
L
D
A
co
u
ld
n
o
t
d
is
co
v
er
.
Du
r
in
g
th
e
tr
ain
in
g
s
tag
e,
th
e
i
m
a
g
es
w
er
e
f
ir
s
t
p
r
o
j
ec
te
d
to
a
P
C
A
s
u
b
s
p
ac
e
s
o
th
at
th
e
r
esu
lti
n
g
s
in
g
u
lar
m
a
tr
ix
is
n
o
n
s
in
g
u
l
ar
.
E
g
en
v
ec
to
r
s
an
d
E
i
g
e
n
v
a
lu
es
w
er
e
th
e
n
co
n
s
tr
u
c
ted
f
o
r
th
e
g
e
n
er
alize
d
eig
en
v
ec
to
r
p
r
o
b
lem
s
o
th
at
t
h
e
lin
ea
r
m
ap
p
in
g
b
est
p
r
eser
v
es
th
e
m
an
if
o
ld
’
s
esti
m
ated
i
n
tr
in
s
ic
g
eo
m
etr
y
i
n
a
lin
ea
r
s
en
s
e.
T
h
e
L
ap
lacia
n
f
ac
e
s
m
e
th
o
d
is
b
ased
o
n
L
o
ca
lit
y
P
r
eser
v
in
g
P
r
o
j
ec
tio
n
s
(
L
P
P
)
w
h
ic
h
is
a
lin
ea
r
m
et
h
o
d
an
d
m
a
y
n
o
t
d
e
tect
all
asp
ec
ts
o
f
t
h
e
in
tr
i
n
s
ic
n
o
n
li
n
ea
r
m
a
n
i
f
o
ld
s
tr
u
ct
u
r
e
b
y
p
r
eser
v
in
g
lo
ca
l
s
tr
u
ct
u
r
e.
A
n
o
v
el
ap
p
r
o
ac
h
b
ased
o
n
T
w
o
-
d
i
m
e
n
s
io
n
a
l
P
r
in
cip
al
C
o
m
p
o
n
e
n
t
An
al
y
s
i
s
(
2
DP
C
A
)
a
n
d
Ker
n
e
l
P
r
in
cip
al
C
o
m
p
o
n
e
n
t
An
al
y
s
i
s
(
KP
C
A
)
f
o
r
f
ac
e
r
ec
o
g
n
i
tio
n
w
a
s
d
ev
elo
p
ed
i
n
[
2
3
]
.
T
h
e
w
o
r
k
f
ir
s
t
p
er
f
o
r
m
s
T
w
o
-
d
i
m
e
n
s
io
n
al
P
r
in
cip
al
C
o
m
p
o
n
en
t
An
al
y
s
is
p
r
o
ce
s
s
t
o
p
r
o
j
ec
t
th
e
f
ac
es
o
n
to
t
h
e
f
ea
tu
r
e
s
p
ac
e.
I
t
th
e
n
p
er
f
o
r
m
s
Ker
n
el
P
r
in
cip
al
C
o
m
p
o
n
en
t
An
al
y
s
i
s
o
n
th
e
p
r
o
j
ec
ted
d
ata.
A
lth
o
u
g
h
i
t
s
h
o
wed
th
at
t
h
e
s
y
s
te
m
ac
h
iev
ed
a
h
i
g
h
r
ec
o
g
n
it
io
n
r
a
te,
it
o
n
l
y
f
o
c
u
s
ed
o
n
P
r
in
cip
a
l
C
o
m
p
o
n
e
n
t
An
al
y
s
i
s
i
m
p
r
o
v
ed
tech
n
iq
u
es
t
h
at
m
ax
i
m
izes
g
lo
b
al
v
ar
ia
n
ce
f
o
r
co
m
p
r
ess
io
n
an
d
n
o
t
f
o
r
class
i
f
icatio
n
lik
e
L
D
A
.
T
h
u
s
th
e
s
y
s
te
m
w
ill
p
er
f
o
r
m
ed
w
o
r
s
e
f
o
r
s
y
s
te
m
h
av
i
n
g
h
ig
h
n
u
m
b
er
o
f
clas
s
e
s
o
f
i
m
ag
e
s
.
A
Face
R
ec
o
g
n
i
t
io
n
alg
o
r
it
h
m
w
i
th
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
IJ
-
AI
Vo
l.
5
,
No
.
2
,
J
u
n
e
2
0
1
6
:
8
0
–
88
82
Su
p
p
o
r
t
Vec
to
r
Ma
ch
in
e
s
(
S
VM
)
w
a
s
p
r
esen
ted
i
n
[
2
4
]
.
T
h
e
p
ap
er
p
r
o
v
id
ed
a
f
ac
e
id
en
ti
f
icatio
n
s
y
s
te
m
b
ased
o
n
SVM
an
d
Di
s
cr
ete
W
av
elet
T
r
an
s
f
o
r
m
(
DW
T
)
.
T
h
e
DW
T
w
a
s
u
s
ed
to
av
o
id
in
cr
ea
s
e
i
n
th
e
co
m
p
u
tatio
n
al
ti
m
e.
Ho
w
e
v
e
r
it
h
a
s
b
ee
n
s
h
o
w
n
t
h
at
n
o
n
lin
ea
r
v
ec
to
r
li
k
e
S
VM
t
h
at
d
o
n
o
t
a
l
w
a
y
s
p
er
f
o
r
m
ed
b
etter
th
a
n
lin
ea
r
m
et
h
o
d
s
in
r
ea
l
-
w
o
r
ld
d
ata
s
ets
h
a
v
in
g
m
o
r
e
co
m
p
licated
d
is
tr
ib
u
tio
n
s
,
th
o
u
g
h
th
e
y
ea
s
il
y
d
e
m
o
n
s
tr
ate
t
h
eir
v
ir
t
u
e
o
n
ar
t
if
ic
ial
n
o
n
li
n
ea
r
d
ata
[
5
]
.
Face
R
ec
o
g
n
itio
n
S
y
s
te
m
B
ase
d
o
n
P
r
in
cip
al
C
o
m
p
o
n
e
n
t
An
al
y
s
i
s
(
P
C
A
)
w
it
h
B
ac
k
P
r
o
p
ag
atio
n
Neu
r
al
Net
w
o
r
k
s
(
B
P
NN)
w
a
s
d
ev
e
lo
p
ed
in
[
2
5
]
.
Su
p
p
o
r
t
Vec
to
r
Ma
ch
in
e
w
as
u
s
ed
f
o
r
f
ac
e
r
ec
o
g
n
i
tio
n
.
Si
m
i
lar
l
y
,
i
n
[
2
6
]
,
B
PNN
w
as
u
s
ed
.
T
h
e
f
ea
t
u
r
es
o
f
t
h
e
q
u
er
y
f
ac
e
i
m
a
g
e
an
d
d
atab
ase
f
ac
e
i
m
a
g
es
w
a
s
ex
tr
ac
ted
u
s
i
n
g
Gab
o
r
tr
an
s
f
o
r
m
a
n
d
tr
ain
ed
u
s
i
n
g
B
P
NN.
Gen
er
all
y
,
n
e
u
r
al
n
et
w
o
r
k
ar
e
n
o
n
li
n
ea
r
m
et
h
o
d
s
th
at
e
n
co
u
n
ter
p
r
o
b
le
m
s
w
h
e
n
t
h
e
n
u
m
b
er
o
f
class
es
in
cr
ea
s
es
ar
e
n
o
t
s
u
ita
b
le
f
o
r
a
s
i
n
g
le
m
o
d
el
i
m
a
g
e
r
ec
o
g
n
itio
n
ta
s
k
[
2
1
]
.
T
h
is
p
ap
er
in
v
o
l
v
es
th
e
u
s
e
o
f
Gab
o
r
W
av
elet
f
o
r
i
m
ag
e
r
ep
r
esen
tatio
n
w
it
h
li
n
ea
r
a
n
d
n
o
n
lin
ea
r
tech
n
iq
u
e
s
,
n
a
m
e
l
y
,
P
C
A
,
L
D
A
,
KP
C
A
an
d
KF
A
an
d
also
s
tu
d
ies t
h
e
r
esu
lt o
f
t
h
e
s
a
m
e
al
g
o
r
ith
m
s
w
it
h
o
u
t th
e
u
s
e
o
f
Gab
o
r
im
a
g
e
r
ep
r
esen
tatio
n
.
3.
M
E
T
H
O
DO
L
O
G
Y
Fig
u
r
e
1
s
h
o
w
t
h
e
Fra
m
e
w
o
r
k
u
s
ed
.
T
h
e
f
ir
s
t
s
tag
e
is
t
h
e
p
r
ep
r
o
ce
s
s
in
g
s
ta
g
e.
T
h
e
p
r
ep
r
o
ce
s
s
in
g
r
ep
r
esen
tatio
n
ad
d
r
ess
es
i
m
a
g
e
v
ar
iab
ilit
ie
s
ca
u
s
ed
b
y
il
l
u
m
in
at
io
n
,
f
ac
ia
l
ex
p
r
ess
io
n
an
d
o
th
er
d
if
f
er
e
n
t
i
m
a
g
e
i
m
p
er
f
ec
tio
n
s
.
T
h
is
w
a
s
ac
h
ie
v
ed
b
y
t
h
e
in
te
g
r
atio
n
o
f
Gab
o
r
w
a
v
elet
f
o
r
ex
tr
ac
ti
o
n
p
r
o
ce
s
s
to
f
o
r
m
th
e
Gab
o
r
im
a
g
e
r
ep
r
esen
tatio
n
o
f
th
e
i
m
ag
e
s
.
T
h
e
Gab
o
r
w
av
elets
(
k
er
n
els,
f
ilter
s
)
u
s
ed
is
d
ef
in
ed
as
s
h
o
w
n
in
E
q
u
atio
n
(
1
)
,
(
z)
=
‖
,
‖
2
2
−
‖
,
‖
2
‖
‖
2
2
2
[
,
−
−
2
2
]
(
1
)
w
h
er
e
an
d
b
d
ef
in
e
t
h
e
o
r
ien
tatio
n
a
n
d
s
ca
le
o
f
t
h
e
Ga
b
o
r
k
er
n
els,
=
(
,
)
,
‖
.
‖
d
en
o
tes
th
e
n
o
r
m
o
p
er
ato
r
,
an
d
th
e
w
a
v
e
v
ec
to
r
,
is
d
ef
in
ed
as
f
o
llo
w
s
:
,
=
∅
(
2
)
w
h
er
e
=
/
an
d
∅
=
/
8
.
is
th
e
m
ax
i
m
u
m
f
r
eq
u
e
n
c
y
,
an
d
f
is
th
e
s
p
ac
in
g
f
ac
to
r
b
et
w
e
e
n
k
er
n
el
s
i
n
t
h
e
f
r
eq
u
e
n
c
y
d
o
m
ain
.
T
h
e
e
f
f
ec
t
o
f
th
e
d
if
f
er
en
ce
o
f
co
n
v
e
x
f
u
n
ctio
n
s
ter
m
b
ec
o
m
e
s
n
eg
l
ig
ib
le
w
h
e
n
th
e
p
ar
a
m
eter
,
w
h
ic
h
d
eter
m
in
e
s
t
h
e
r
atio
o
f
t
h
e
Gau
s
s
ia
n
w
i
n
d
o
w
w
id
t
h
to
w
a
v
ele
n
g
t
h
,
h
as
s
u
f
f
icie
n
tl
y
lar
g
e
v
a
lu
e
s
.
As
i
n
m
o
s
t
ca
s
es
Gab
o
r
w
a
v
elet
s
u
s
ed
is
o
f
f
i
v
e
d
if
f
er
en
t
s
ca
le
s
,
ϵ
{
0
,
…
,
4
}
,
an
d
eig
h
t o
r
ien
tatio
n
s
,
ϵ
{
0
,
…
,
7
}
[
2
7
]
.
Fo
r
th
e
Gab
o
r
f
ea
tu
r
e
r
ep
r
esen
tat
io
n
,
th
e
Gab
o
r
w
av
el
et
r
ep
r
esen
tatio
n
o
f
a
n
i
m
a
g
e
is
t
h
e
co
n
v
o
lu
tio
n
o
f
th
e
i
m
a
g
e
w
it
h
th
e
f
a
m
il
y
o
f
Gab
o
r
k
er
n
el
s
as
d
ef
i
n
ed
b
y
(
1
)
.
L
et
I
(
x,
y
)
b
e
th
e
g
r
a
y
-
le
v
el
d
is
tr
ib
u
tio
n
o
f
an
i
m
a
g
e,
th
e
c
o
n
v
o
l
u
tio
n
o
f
i
m
a
g
e
I
is
d
ef
i
n
ed
as f
o
llo
w
s
:
,
(
)
=
I
(
z
)
*
,
(
)
(
3
)
w
h
er
e
th
e
f
o
llo
w
i
n
g
co
u
n
ts
=
(
,
)
,
∗
d
en
o
tes
th
e
co
n
v
o
l
u
tio
n
o
p
er
ato
r
,
an
d
,
(
)
is
th
e
co
n
v
o
lu
tio
n
r
esu
lt c
o
r
r
esp
o
n
d
in
g
to
th
e
Ga
b
o
r
k
er
n
el
at
o
r
ien
tatio
n
an
d
s
ca
le
.
W
h
ich
p
r
o
d
u
ce
th
e
s
et
S
=
{
,
(
)
:
∈
{
0
,
…
,
7
}
,
∈
{
0
,
…
,
4
}
}
(
4
)
E
q
u
atio
n
(
4
)
f
o
r
m
s
t
h
e
Gab
o
r
w
a
v
elet
r
ep
r
esen
tatio
n
o
f
th
e
i
m
a
g
e
I
(
z
)
.
T
o
en
co
m
p
a
s
s
d
if
f
er
en
t
s
p
atia
l
f
r
eq
u
en
c
ies
(
s
ca
le)
,
s
p
atial
lo
ca
liti
es,
an
d
o
r
ien
tatio
n
s
elec
tiv
ite
s
,
all
t
h
e
r
esp
r
esen
tatio
n
r
es
u
lt
s
ar
e
co
n
ca
ten
ated
an
d
a
n
au
g
m
en
te
d
f
ea
tu
r
e
v
ec
to
r
X
is
d
er
iv
ed
b
y
d
o
w
n
s
a
m
p
li
n
g
[
2
6
]
,
[
2
8
]
.
T
h
e
d
im
e
n
s
io
n
o
f
Gab
o
r
f
ac
e
s
is
v
er
y
h
ig
h
.
I
n
o
r
d
er
to
r
ed
u
ce
th
e
d
i
m
e
n
s
io
n
ali
t
y
,
at
t
h
e
s
a
m
e
ti
m
e
r
eser
v
e
t
h
e
i
n
tr
in
s
ic
p
ar
t
o
f
th
e
i
m
a
g
es,
t
h
e
d
i
m
e
n
s
io
n
al
r
ed
u
ctio
n
tech
n
iq
u
es
w
er
e
u
s
ed
i
n
t
h
e
s
ec
o
n
d
s
ta
g
e
.
Af
ter
p
as
s
i
n
g
th
r
o
u
g
h
t
h
e
p
r
e
p
r
o
ce
s
s
in
g
s
ta
g
es,
th
e
tr
ain
ed
i
m
a
g
es
u
n
d
er
g
o
d
i
m
e
n
s
io
n
al
r
ed
u
ctio
n
p
r
o
ce
s
s
e
s
u
s
i
n
g
t
h
e
li
n
ea
r
an
d
t
h
e
n
o
n
li
n
ea
r
tech
n
iq
u
e
s
:
P
C
A
,
L
D
A
,
KP
C
A
a
n
d
KF
A
b
y
p
r
o
j
ec
tin
g
t
h
e
i
m
ag
e
s
o
n
t
o
s
u
b
s
p
ac
e
a
n
d
s
to
r
in
g
th
eir
p
r
o
j
ec
tio
n
s
in
t
h
e
d
atab
ase.
Als
o
,
b
ef
o
r
e
th
e
m
atc
h
i
n
g
,
th
e
p
r
o
b
e
im
a
g
es
ar
e
p
r
ep
r
o
ce
s
s
ed
,
an
d
p
r
o
j
ec
ted
o
n
to
t
h
e
s
a
m
e
s
u
b
s
p
ac
e
a
s
t
h
e
g
aller
y
i
m
ag
e
u
s
in
g
a
s
i
m
ilar
alg
o
r
ith
m
.
T
h
e
f
in
a
l
s
tag
e
i
s
th
e
r
ec
o
g
n
itio
n
s
ta
g
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
D
ev
elo
p
men
t o
f a
n
E
fficien
t
F
a
ce
R
ec
o
g
n
itio
n
S
ystem
B
a
s
ed
o
n
Lin
ea
r
…
(
A
r
a
o
lu
w
a
S
imileo
lu
F
ila
n
i
)
83
T
h
e
test
i
m
ag
e
p
r
o
j
ec
tio
n
s
is
th
en
co
m
p
ar
ed
to
s
to
r
ed
g
aller
y
p
r
o
j
ec
tio
n
s
b
y
u
s
i
n
g
Ma
h
alan
o
b
i
s
C
o
s
i
n
e
d
is
ta
n
ce
m
etr
ic
b
y
ca
lcu
lati
n
g
t
h
e
d
is
ta
n
ce
s
f
r
o
m
a
p
r
o
b
e
im
a
g
e
p
r
o
j
ec
tio
n
to
all
g
aller
y
i
m
a
g
es
p
r
o
j
ec
tio
n
s
an
d
t
h
en
ch
o
o
s
i
n
g
th
e
m
i
n
i
m
u
m
d
i
s
tan
ce
as
a
s
i
m
ilar
it
y
m
ea
s
u
r
e.
T
h
e
id
en
tit
y
o
f
t
h
e
m
o
s
t
s
i
m
ilar
g
aller
y
i
m
a
g
e
is
t
h
e
n
ch
o
s
e
n
t
o
b
e
th
e
r
esu
lt
o
f
r
ec
o
g
n
itio
n
an
d
th
e
u
n
k
n
o
w
n
p
r
o
b
e
i
m
ag
e
is
i
d
en
ti
f
ied
.
T
h
e
d
ataset
ar
e
f
r
o
m
t
h
e
A
T
&
T
f
ac
e
d
atab
ase.
T
h
er
e
a
r
e
4
0
0
i
m
a
g
es
in
all.
T
h
er
e
ar
e
4
0
d
atasets
co
n
tai
n
i
n
g
d
if
f
er
e
n
t
s
u
b
j
ec
ts
.
E
ac
h
d
atas
ets
h
a
s
1
0
s
a
m
p
le
s
.
T
h
e
f
ir
s
t
th
r
ee
s
a
m
p
les
ar
e
s
elec
ted
f
o
r
tr
ain
i
n
g
,
t
h
e
n
ex
t
f
o
u
r
s
a
m
p
le
s
w
er
e
p
r
eser
v
ed
as
t
h
e
te
s
t
s
et,
t
h
e
r
e
m
ai
n
i
n
g
s
a
m
p
le
s
w
er
e
u
s
ed
a
s
t
h
e
ev
alu
a
tin
g
s
et.
T
h
e
alg
o
r
ith
m
s
u
s
ed
ar
e
d
escr
ib
ed
in
th
e
n
e
x
t sectio
n
.
3
.
1
.
P
rincipa
l C
o
m
po
ne
nt
Ana
ly
s
is
(
P
CA)
P
r
in
cip
al
C
o
m
p
o
n
en
t
A
n
al
y
s
i
s
(
P
C
A
)
is
u
s
ed
to
lin
ea
r
l
y
s
ep
ar
ate
th
e
i
m
a
g
e
v
ec
to
r
s
i
n
t
o
a
lo
w
e
r
d
ata.
I
f
v
-
d
i
m
e
n
s
io
n
al
v
ec
to
r
o
f
ea
ch
tr
ai
n
in
g
f
ac
e
o
f
s
et
h
as
M
i
m
a
g
es,
th
e
P
r
in
cip
al
C
o
m
p
o
n
en
t
An
al
y
s
is
(
P
C
A
)
al
g
o
r
ith
m
i
s
u
s
ed
to
f
i
n
d
a
t
-
d
i
m
e
n
s
io
n
al
s
u
b
s
p
ac
e
w
h
o
s
e
b
asi
s
v
ec
to
r
co
r
r
esp
o
n
d
to
th
e
m
a
x
i
m
u
m
v
ar
ian
ce
i
n
th
e
d
ir
ec
tio
n
in
t
h
e
o
r
ig
in
al
i
m
a
g
e
s
p
ac
e
s
,
w
h
er
e
(
t
<v
)
,
t
is
also
th
e
eig
en
v
ec
to
r
s
o
f
th
e
co
v
ar
ian
ce
m
atr
i
x
.
A
ll
i
m
a
g
es
o
f
k
n
o
w
n
f
ac
es
ar
e
p
r
o
j
ec
ted
o
n
to
th
e
f
ac
e
s
p
ac
e
to
f
in
d
s
ets
o
f
w
eig
h
t
s
t
h
at
d
escr
ib
e
th
e
co
n
tr
ib
u
tio
n
o
f
ea
ch
v
ec
to
r
.
B
y
co
m
p
ar
in
g
a
s
e
t
o
f
w
e
ig
h
t
s
f
o
r
t
h
e
u
n
k
n
o
w
n
f
ac
e
to
s
et
s
o
f
w
ei
g
h
t
s
o
f
k
n
o
w
n
f
ac
e
s
,
th
e
f
ac
e
ca
n
b
e
id
en
tif
ied
.
I
f
th
e
i
m
a
g
e
e
le
m
e
n
ts
ar
e
r
an
d
o
m
v
ar
iab
le
s
d
u
r
in
g
th
e
r
ec
o
g
n
itio
n
p
r
o
ce
s
s
,
th
e
P
r
in
cip
al
C
o
m
p
o
n
e
n
ts
A
n
a
l
y
s
is
(
P
C
A
)
b
asic
v
ec
to
r
s
ar
e
d
e
f
i
n
ed
as
ei
g
en
v
ec
to
r
s
o
f
t
h
e
s
ca
tter
m
a
tr
i
x
is
d
ef
in
ed
a
s
:
∑
(
−
)
.
(
−
)
=
1
(
5
)
w
h
er
e
is
th
e
m
ea
n
o
f
all
M
im
ag
e
s
in
th
e
tr
ai
n
in
g
s
et
o
r
th
e
m
ea
n
f
ac
e,
T
is
th
e
tr
an
s
p
o
s
e
o
f
its
p
r
o
p
er
ties
an
d
is
th
e
i
th
i
m
ag
e
w
it
h
it
s
co
lu
m
n
s
co
n
ca
te
n
ated
in
a
v
ec
to
r
.
Fig
u
r
e
2
s
h
o
w
s
t
h
e
m
ea
n
f
ac
e
o
f
th
e
w
o
r
k
a
n
d
th
e
E
i
g
e
n
f
ac
e
s
o
f
th
e
E
ig
e
n
Face
s
o
f
t
h
e
1
st
,
2
nd
,
1
0
0
th
,
an
d
th
e
3
0
0
th
s
u
b
j
e
cts.
T
h
e
P
r
in
cip
al
co
m
p
o
n
e
n
t
s
o
f
t e
i
g
en
v
ec
to
r
s
ar
e
t la
r
g
est eig
e
n
v
alu
e
s
,
cr
ea
tin
g
a
t d
i
m
en
s
io
n
a
l f
ac
e
s
p
ac
e
[
2
0
]
,
[
2
9
]
.
Fig
u
r
e
1.
T
h
e
Sy
s
te
m
Fra
m
e
W
o
r
k
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
IJ
-
AI
Vo
l.
5
,
No
.
2
,
J
u
n
e
2
0
1
6
:
8
0
–
88
84
Fig
u
r
e
2.
th
e
m
ea
n
f
ac
e
o
f
th
e
w
o
r
k
a
n
d
th
e
E
i
g
en
f
ac
es
3
.
2
.
L
inea
r
Dis
cr
i
m
ina
nt
A
na
ly
s
i
s
(
L
DA)
L
i
n
ea
r
Dis
cr
i
m
i
n
an
t
A
n
a
l
y
s
is
(
L
D
A
)
is
u
s
ed
to
p
r
o
v
id
e
a
b
etter
class
i
f
icatio
n
o
f
d
ata
w
h
er
e
th
e
d
at
a
co
n
tain
h
i
g
h
n
u
m
b
er
o
f
c
lass
es
.
T
h
is
is
ac
h
iev
ed
b
y
f
i
n
d
i
n
g
t
h
e
b
est
r
ep
r
ese
n
tatio
n
a
m
o
n
g
clas
s
es.
L
D
A
co
n
s
id
er
s
f
o
r
all
s
a
m
p
les
o
f
al
l
class
es,
th
e
b
et
w
ee
n
-
cla
s
s
s
c
atter
m
a
tr
ix
an
d
th
e
w
ith
in
-
cl
ass
s
ca
tter
m
atr
i
x
w
h
ic
h
ar
e
d
ef
in
ed
b
y
=
∑
.
(
−
)
.
(
−
)
=
1
(
6)
w
h
er
e
is
th
e
n
u
m
b
er
o
f
tr
ain
in
g
s
a
m
p
les
in
clas
s
i
,
c
is
th
e
n
u
m
b
er
o
f
d
is
tin
c
t
class
e
s
,
is
th
e
m
ea
n
v
ec
to
r
o
f
s
a
m
p
les
b
elo
n
g
in
g
t
o
class
i
an
d
r
ep
r
esen
ts
th
e
s
et
o
f
s
a
m
p
les
b
elo
n
g
in
g
to
class
i
w
i
th
b
ein
g
th
e
k
–
t
h
i
m
ag
e
o
f
t
h
at
c
lass
.
T
is
th
e
tr
an
s
p
o
s
e
o
f
it
s
p
r
o
p
er
ties
.
r
ep
r
esen
ts
th
e
s
ca
tter
o
f
f
ea
t
u
r
es
ar
o
u
n
d
th
e
o
v
er
all
m
ea
n
f
o
r
all
f
ac
e
class
es
a
n
d
r
ep
r
esen
ts
th
e
s
ca
tter
o
f
f
ea
tu
r
es
ar
o
u
n
d
th
e
m
e
an
o
f
ea
ch
f
ac
e
class
.
T
h
e
g
o
al
is
to
m
a
x
i
m
ize
w
h
ile
m
i
n
i
m
izi
n
g
,
in
o
th
er
w
o
r
d
s
,
m
ax
i
m
ize
th
e
r
atio
d
et
|
|
/
|
|
[
2
9
]
.
T
h
is
r
atio
is
m
a
x
i
m
ized
w
h
e
n
t
h
e
co
lu
m
n
v
ec
to
r
s
o
f
th
e
p
r
o
j
ec
tio
n
m
atr
ic
(
)
ar
e
th
e
ei
g
en
v
ec
to
r
s
o
f
−
1
.
.
I
n
o
r
d
er
to
p
r
ev
en
t
to
b
ec
o
m
e
=
∑
∑
(
−
)
.
(
−
)
∈
=
1
(
7
)
s
in
g
u
lar
,
P
C
A
i
s
u
s
ed
as a
p
r
ep
r
o
ce
s
s
in
g
s
tep
an
d
th
e
f
i
n
al
tr
an
s
f
o
r
m
atio
n
is
=
[
2
0
]
.
3
.
3
.
K
er
nel P
rincipa
l C
o
m
po
ne
n
t
Ana
ly
s
is
(
K
P
CA)
T
h
e
Ker
n
el
p
r
o
j
ec
tio
n
tech
n
iq
u
es
ar
e
u
s
ed
to
p
r
o
v
id
e
a
b
ett
er
d
is
cr
i
m
i
n
atio
n
a
m
o
n
g
n
o
n
l
i
n
ea
r
it
y
o
f
d
ata.
T
h
e
m
a
in
id
ea
is
to
m
a
p
in
p
u
t
d
ata
in
to
a
h
i
g
h
-
di
m
e
n
s
io
n
al
f
ea
t
u
r
e
s
p
ac
e
an
d
p
er
f
o
r
m
a
s
i
m
ilar
P
C
A
p
r
o
ce
s
s
ex
p
lain
ed
i
n
eq
u
atio
n
(
5
)
.
T
h
e
Ker
n
el
m
eth
o
d
s
i
s
u
s
ed
to
id
en
t
if
y
a
li
n
ea
r
s
u
b
s
p
ac
e
in
th
e
h
i
g
h
-
d
i
m
es
n
io
n
al
f
ea
t
u
r
e
s
p
ac
e
r
a
th
er
th
a
n
t
h
e
o
r
ig
i
n
al
i
n
p
u
t
s
p
ac
e
b
y
a
v
o
id
in
g
d
ir
ec
t
co
m
p
u
tat
io
n
o
f
th
e
n
o
n
li
n
ea
r
m
ap
p
in
g
Ф
t
h
r
o
u
g
h
“k
er
n
el
tr
ick
”
an
d
d
er
i
v
e
t
h
e
k
er
n
el
tr
an
s
f
o
r
m
atio
n
m
a
t
r
ices
b
ased
o
n
th
e
k
er
n
el
m
a
tr
ices
o
f
t
h
e
tr
ai
n
i
n
g
d
ata.
T
h
e
r
atio
n
ale
o
f
p
er
f
o
r
m
i
n
g
s
u
c
h
a
n
o
n
l
in
ea
r
m
a
p
p
in
g
co
m
es
f
r
o
m
C
o
v
er
’
s
th
eo
r
e
m
w
h
ic
h
s
tate
t
h
at
“A
co
m
p
le
x
p
atter
n
-
c
lass
if
icatio
n
p
r
o
b
lem
ca
s
t
in
a
h
ig
h
-
d
i
m
en
s
io
n
al
s
p
ac
e
n
o
n
li
n
ea
r
l
y
is
m
o
r
e
li
k
el
y
t
o
b
e
lin
ea
r
l
y
s
ep
ar
ab
le
th
a
n
in
a
lo
w
-
d
i
m
e
n
s
io
n
al
s
p
ac
e”
[
3
0
]
,
[
2
3
]
.
B
y
co
n
s
id
er
in
g
th
e
s
et
o
f
i
m
a
g
e
s
a
m
p
les
,
=
[
1
,
…
,
]
∈
(
8)
Ker
n
el
P
C
A
i
s
u
s
ed
b
y
p
r
o
j
ec
tin
g
ea
ch
v
ec
to
r
x
is
p
r
o
j
ec
ted
f
r
o
m
t
h
e
i
n
p
u
t
s
p
ac
e,
,
to
a
h
ig
h
d
i
m
e
n
s
io
n
a
l
f
ea
t
u
r
e
s
p
ac
e,
,
b
y
a
n
o
n
l
in
ea
r
m
ap
p
in
g
f
u
n
ctio
n
:
Ф
:
→
,
f
>
n
.
P
C
A
p
r
o
ce
s
s
is
t
h
e
n
ca
r
r
ied
o
u
t
o
n
th
e
k
er
n
el
s
u
b
s
p
ac
es b
y
s
o
l
v
i
n
g
th
e
co
r
r
esp
o
n
d
in
g
ei
g
en
v
al
u
e
p
r
o
b
lem
:
Ф
=
Ф
Ф
(
9
)
w
h
er
e
Ф
is
a
co
v
ar
ian
ce
m
atr
i
x
.
A
ll
s
o
lu
t
io
n
Ф
w
it
h
≠
0
lie
i
n
t
h
e
s
p
an
o
f
Ф
(
1
)
,
…,
Ф
(
)
[
2
9
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
D
ev
elo
p
men
t o
f a
n
E
fficien
t
F
a
ce
R
ec
o
g
n
itio
n
S
ystem
B
a
s
ed
o
n
Lin
ea
r
…
(
A
r
a
o
lu
w
a
S
imileo
lu
F
ila
n
i
)
85
3
.
4
.
K
er
nel F
is
her
Ana
ly
s
i
s
(
K
F
A)
KF
A
is
u
s
ed
to
r
ed
u
ce
t
h
e
d
ata
in
to
a
lo
w
er
s
u
b
s
p
ac
e
a
n
d
d
esig
n
ed
to
w
o
r
k
b
etter
th
a
n
th
e
li
n
ea
r
m
et
h
o
d
s
w
h
er
e
th
er
e
ar
e
co
m
p
lex
m
a
n
i
f
o
ld
o
f
d
ata
h
ig
h
n
u
m
b
er
o
f
cla
s
s
es.
I
t
is
p
er
f
o
r
m
ed
u
s
in
g
th
e
s
i
m
ilar
p
r
o
ce
d
u
r
e
o
f
KP
C
A
e
x
ce
p
t
t
h
at
Fi
s
h
er
L
in
ea
r
Di
s
cr
i
m
in
a
n
t
(
F
L
D)
i
s
co
n
s
id
er
ed
in
s
tea
d
o
f
P
C
A
af
ter
th
e
tr
an
s
f
o
r
m
atio
n
o
f
th
e
s
u
b
s
p
ac
e
to
h
ig
h
er
d
i
m
e
n
s
io
n
.
I
f
h
as
th
e
s
a
m
e
v
al
u
e
o
f
eq
u
atio
n
(
8
)
[
3
1
]
,
th
e
s
am
e
p
r
o
j
ec
tio
n
is
p
er
f
o
r
m
ed
o
n
t
h
e
v
ec
to
r
x
.
to
g
et
t
h
e
f
u
n
ctio
n
Ф
:
→
,
f
>
n
.
L
et
t
h
e
p
r
o
j
ec
ted
s
a
m
p
les Ф
(
x
)
b
e
ce
n
tr
ed
in
an
d
let
t
h
e
e
q
u
atio
n
s
t
h
at
u
s
e
d
o
t
p
r
o
d
u
c
ts
b
e
f
o
r
m
u
lated
f
o
r
Fis
h
er
l
in
ea
r
Di
s
cr
i
m
a
te
An
al
y
s
i
s
(
FL
D)
o
n
l
y
.
A
s
s
u
m
e
th
e
w
it
h
i
n
-
clas
s
an
d
b
et
w
ee
n
-
c
lass
s
ca
tter
m
atr
ices
b
e
Ф
an
d
Ф
,
to
a
p
p
ly
FL
D
i
n
k
er
n
el
s
p
ac
e,
t
h
e
s
o
lu
t
io
n
to
eig
en
v
al
u
es
a
n
d
eig
e
n
v
ec
to
r
s
Ф
of
Ф
Ф
=
Ф
Ф
(
1
0
)
ar
e
d
er
iv
ed
b
y
f
i
n
d
in
g
th
e
ei
g
en
v
ec
to
r
s
co
r
r
esp
o
n
d
in
g
to
lar
g
est
g
e
n
er
alize
d
ei
g
e
n
v
a
lu
e.
T
h
e
k
er
n
el
f
u
n
ctio
n
is
in
tr
o
d
u
ce
d
ef
in
ed
b
y
(
)
=
k
(
,
)
=
Ф
(
)
.
Ф
(
`
)
(
1
1
)
w
h
er
e
t
h
er
e
ex
is
t
s
a
c
-
clas
s
p
r
o
b
lem
an
d
a
r
-
th
s
a
m
p
le
o
f
cl
ass
t
an
d
t
h
e
s
-
th
s
a
m
p
le
o
f
cl
ass
u
b
e
an
d
r
esp
ec
tiv
el
y
(
w
h
er
e
clas
s
t
ha
s
s
a
m
p
le
s
an
d
cla
s
s
u
h
a
s
s
am
p
les).
T
h
en
f
in
al
l
y
p
r
o
j
ec
t
Ф
(
)
to
a
lo
w
er
d
i
m
en
s
io
n
al
s
p
ac
e
s
p
a
n
n
ed
b
y
th
e
ei
g
e
n
v
ec
to
r
s
Ф
in
a
w
a
y
s
i
m
ilar
to
Ker
n
el
P
C
A
u
s
in
g
F
i
s
h
er
f
ac
e
m
eth
o
d
f
o
r
f
ac
e
r
ec
o
g
n
i
tio
n
[
7
]
,
[
2
9
]
.
4.
M
AT
CH
I
NG
Fo
r
th
e
m
atc
h
in
g
tas
k
,
th
e
Ma
h
ali
n
o
b
is
C
o
s
in
e
(
M
A
HC
OS)
d
is
tan
ce
m
etr
ic
is
u
s
ed
.
T
h
is
i
s
b
ec
au
s
e
it
is
t
h
e
m
o
s
t
ac
cu
r
ate
a
n
d
ef
f
i
cien
t
i
n
ter
m
s
o
f
v
er
if
icatio
n
,
id
en
ti
f
icatio
n
a
n
d
r
o
b
u
s
t
n
ess
[
3
2
]
.
I
t
m
ea
s
u
r
e
th
e
co
s
in
e
o
f
th
e
p
r
o
j
ec
ted
in
to
th
e
r
ec
o
g
n
itio
n
s
p
ac
e
u
s
i
n
g
th
e
co
r
r
esp
o
n
d
in
g
d
i
m
e
n
s
io
n
al
r
e
d
u
ctio
n
tec
h
n
iq
u
es
.
Af
ter
tr
a
n
s
f
o
r
m
atio
n
s
ar
e
co
m
p
leted
,
Ma
h
ali
n
o
b
is
Di
s
tan
ce
m
ea
s
u
r
es
is
u
s
ed
to
clas
s
i
f
y
d
ata
p
o
in
ts
b
y
u
s
in
g
it
to
co
m
p
u
te
t
h
e
s
i
m
i
lar
it
y
b
et
w
ee
n
t
w
o
f
ac
e
s
f
ea
tu
r
e
s
.
Fo
r
i
m
a
g
es
u
a
n
d
v
w
it
h
co
r
r
esp
o
n
d
in
g
p
r
o
j
e
ctio
n
s
m
an
d
n
i
n
Ma
h
ali
n
o
b
is
s
p
ac
e,
w
h
er
e
m
an
d
n
ar
e
t
w
o
f
ea
tu
r
e
v
ec
to
r
s
tr
an
s
f
o
r
m
ed
in
to
Ma
h
ali
n
o
b
is
s
p
ac
e,
t
h
e
Ma
h
ali
n
o
b
is
C
o
s
i
n
e
is
[
3
3
]
:
ℎ
(
,
)
=
c
os
(
Ɵ
)
=
|
|
|
|
co
s
(
Ɵ
)
|
|
|
|
=
.
|
|
|
|
(
1
2
)
w
it
h
a
n
a
n
g
le
Ɵ
d
ef
in
ed
as
t
h
e
an
g
le
b
et
w
ee
n
t
h
e
i
m
a
g
es
a
f
te
r
th
e
y
h
a
v
e
b
ee
n
p
r
o
j
ec
ted
in
to
th
e
r
ec
o
g
n
itio
n
s
p
ac
e
as d
is
tan
ce
b
et
w
ee
n
p
r
o
jecte
d
i
m
ag
e
s
.
T
h
is
d
is
tan
ce
i
s
r
ef
er
ed
to
as th
e
Ma
h
C
o
s
i
n
e
d
is
tan
ce
.
5.
E
VA
L
UA
T
I
O
N
AND
R
E
SU
L
T
S.
I
n
o
r
d
er
to
test
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
ea
c
h
al
g
o
r
ith
m
t
h
r
ee
d
i
f
f
er
en
t
t
y
p
e
p
er
f
o
r
m
a
n
ce
m
etr
ic
s
ar
e
u
s
ed
w
it
h
a
n
d
w
it
h
o
u
t
th
e
u
s
e
o
f
G
ab
o
r
W
av
elets.
T
h
ey
ar
e
t
h
e:
(
a
)
Cu
m
ula
t
iv
e
M
a
t
ch
Sco
re
Curv
e
(
CM
C)
,
(
b)
Rec
eiv
er
O
pera
t
ing
Cha
ra
c
t
er
is
t
ic
(
RO
C)
Curv
e,
a
n
d
(
C)
E
x
pect
ed
P
er
f
o
r
m
a
nce
Curv
e
(
E
P
C)
.
T
h
e
C
u
m
u
lati
v
e
Ma
tc
h
C
u
r
v
e
s
(
C
MCs
)
is
u
s
ed
to
ca
lcu
late
th
e
r
ec
o
g
n
itio
n
r
ate.
T
h
e
h
o
r
izo
n
tal
ax
is
r
ep
r
esen
ts
t
h
e
r
an
k
a
n
d
th
e
v
er
tical
a
x
i
s
r
ep
r
esen
t
s
t
h
e
cu
m
u
lati
v
e
m
atc
h
s
co
r
e
co
r
r
esp
o
n
d
in
g
to
th
e
r
a
n
k
.
T
h
e
lo
w
er
c
u
r
v
e
co
r
r
esp
o
n
d
s
to
th
e
f
ac
e
r
e
co
g
n
itio
n
tech
n
iq
u
e
s
w
ith
a
lo
w
er
p
er
f
o
r
m
a
n
ce
.
T
h
e
R
ec
eiv
er
Op
er
atin
g
C
h
ar
ac
ter
is
tic
(
R
O
C
)
cu
r
v
e
is
a
m
o
r
e
g
en
er
al
c
u
r
v
e
u
s
ed
i
n
f
ac
e
r
ec
o
g
n
itio
n
p
er
f
o
r
m
an
ce
.
T
h
e
h
o
r
izo
n
tal
ax
i
s
r
ep
r
esen
ts
t
h
e
f
al
s
e
ac
ce
p
t
r
at
e
o
r
F
A
R
,
w
h
ile
t
h
e
v
er
tical
ax
is
co
r
r
esp
o
n
d
s
to
t
h
e
f
ac
e
v
er
if
ica
tio
n
r
ate
o
r
FVR
.
T
h
e
E
P
C
cu
r
v
e
s
h
o
w
s
class
if
ier
s
f
r
o
m
th
e
v
ie
w
p
o
in
t
o
f
t
h
e
tr
ad
eo
f
f
b
et
w
ee
n
f
a
ls
e
alar
m
a
n
d
f
alse
r
ej
ec
ts
p
r
o
b
ab
ilit
ies.
T
h
e
E
P
C
cu
r
v
e
s
ar
e
p
r
o
d
u
ce
u
s
i
n
g
a
n
ev
alu
a
tio
n
i
m
a
g
e
s
e
t
an
d
a
te
s
t
i
m
ag
e
s
et
w
h
ic
h
ar
e
r
eq
u
ir
ed
.
Fo
r
ea
ch
,
th
e
d
ec
is
io
n
t
h
r
esh
o
ld
t
h
at
m
i
n
i
m
i
ze
s
th
e
w
ei
g
h
ted
s
u
m
o
f
t
h
e
Fals
e
A
cc
ep
tan
c
e
R
ate
(
F
A
R
)
an
d
Fal
s
e
R
ej
ec
tio
n
R
ate
(
FR
R
)
is
co
m
p
u
ted
o
n
th
e
ev
a
lu
at
io
n
i
m
ag
e
s
et.
T
h
is
t
h
r
es
h
o
ld
is
th
e
n
u
s
ed
o
n
t
h
e
te
s
t
i
m
a
g
es
to
d
eter
m
i
n
e
th
e
v
al
u
e
o
f
th
e
h
al
f
to
tal
er
r
o
r
r
ates
(
HT
E
R
)
d
e
f
i
n
ed
as
HT
E
R
=
(
FAR
+F
R
R
)
/2
.
E
P
C
t
h
e
n
p
lo
t
th
e
h
al
f
to
tal
er
r
o
r
r
ate
(
HT
E
R
=0
.
5
(
FA
R
+
FR
R
)
)
ag
a
in
s
t
th
e
p
ar
a
m
eter
,
w
h
ic
h
co
n
tr
o
ls
th
e
r
elativ
e
i
m
p
o
r
tan
ce
o
f
th
e
t
w
o
er
r
o
r
r
ates
F
A
R
a
n
d
F
R
R
i
n
t
h
e
ex
p
r
es
s
io
n
:
F
A
R
+
(
1
−
)
FR
R
[
3
4
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
IJ
-
AI
Vo
l.
5
,
No
.
2
,
J
u
n
e
2
0
1
6
:
8
0
–
88
86
5
.
1
.
I
nte
rpre
t
a
t
io
n
o
f
Resul
t
s
T
ab
le
1
g
iv
es
th
e
d
etailed
s
u
m
m
ar
y
o
f
t
h
e
r
esu
lt
s
.
T
h
er
e
ar
e
4
0
0
im
a
g
es
co
n
tai
n
i
n
g
1
0
d
if
f
er
en
t
i
m
a
g
es
o
f
ea
ch
p
er
s
o
n
.
1
2
0
im
ag
e
s
ar
e
u
s
ed
f
o
r
tr
ain
i
n
g
,
1
6
0
im
ag
e
s
ar
e
u
s
ed
f
o
r
test
i
n
g
,
t
h
e
r
e
m
ai
n
i
n
g
i
m
a
g
es
s
er
v
e
as
t
h
e
e
v
alu
a
tio
n
s
et
s
.
Fro
m
t
h
e
r
es
u
lts
,
L
D
A
o
u
tp
er
f
o
r
m
ed
o
th
er
m
et
h
o
d
s
in
m
o
s
t
ca
s
es
a
n
d
h
av
e
a
v
er
y
clo
s
e
r
ec
o
g
n
i
tio
n
r
ate
w
it
h
K
F
A
.
B
o
th
L
D
A
a
n
d
KF
A
h
a
v
e
t
h
e
h
i
g
h
est
p
er
f
o
r
m
an
ce
o
f
9
3
.
3
3
%
f
r
o
m
t
h
e
C
M
C
an
d
R
O
C
r
es
u
lts
w
h
en
t
h
e
ev
a
lu
ati
n
g
s
e
t
is
u
s
ed
w
it
h
Gab
o
r
.
T
h
eir
p
er
f
o
r
m
an
ce
u
s
i
n
g
C
M
C
an
d
R
O
C
w
it
h
Gab
o
r
w
it
h
m
o
r
e
test
i
m
a
g
es
i
s
9
1
.
8
8
%
an
d
9
3
.
1
3
%
r
esp
ec
tiv
el
y
(
i.e
.
wh
en
t
h
e
te
s
t
s
et
i
s
u
s
ed
)
.
I
t
is
o
b
v
io
u
s
th
a
t
t
h
e
K
F
A
an
d
L
D
A
h
ig
h
p
er
f
o
r
m
an
ce
is
d
u
e
to
h
i
g
h
n
u
m
b
er
o
f
c
l
ass
es
o
f
t
h
e
i
m
a
g
es
o
f
t
h
e
s
y
s
te
m
(
t
h
e
d
atab
ase
h
as
a
to
tal
o
f
4
0
b
et
w
ee
n
-
c
la
s
s
(
m
atr
ices)
a
n
d
1
0
w
it
h
i
n
-
c
lass
i
m
a
g
es).
T
h
e
r
esu
lt
s
o
f
th
e
e
x
p
er
i
m
en
t
s
h
o
w
th
at
in
co
r
p
o
r
atin
g
Gab
o
r
i
m
ag
e
r
ep
r
esen
tat
io
n
m
a
k
es a
n
o
tab
le
co
n
tr
ib
u
tio
n
to
th
e
o
v
er
all
f
ac
e
r
ec
o
g
n
itio
n
p
er
f
o
r
m
an
ce
o
f
all
t
h
e
al
g
o
r
ith
m
s
.
T
h
is
is
t
h
e
Gab
o
r
w
av
ele
ts
r
ep
r
esen
tatio
n
s
ar
e
to
s
o
m
e
e
x
te
n
t
in
s
e
n
s
iti
v
e
to
i
m
ag
e
i
n
v
ar
iab
lili
tie
s
.
W
h
en
m
o
r
e
n
u
m
b
er
o
f
test
i
m
ag
e
s
ar
e
u
s
ed
w
it
h
th
e
Gab
o
r
w
av
e
lets
,
th
er
e
i
s
a
g
e
n
er
al
r
ed
u
ctio
n
i
n
p
er
f
o
r
m
a
n
ce
o
f
all
th
e
alg
o
r
it
h
m
s
a
n
d
t
h
er
e
is
m
o
r
e
r
ed
u
ctio
n
w
h
e
n
P
C
A
b
ased
al
g
o
r
ith
m
s
ar
e
u
s
ed
(
P
C
A
an
d
KP
C
A
)
t
h
an
t
h
e
L
D
A
b
ased
alg
o
r
it
h
m
s
(
L
D
A
a
n
d
KF
A
)
.
Fo
r
ex
a
m
p
le
w
i
th
Gab
o
r
W
a
v
elets,
t
h
e
R
OC
p
er
f
o
r
m
a
n
ce
o
f
b
o
th
P
C
A
an
d
KP
C
A
r
ec
o
g
n
itio
n
r
ate
f
o
r
th
e
ev
alu
a
tin
g
s
et(
w
h
ich
co
n
tai
n
1
2
0
p
r
o
b
e
im
a
g
e
s
)
ar
e
b
o
th
9
2
.
5
0
%
an
d
th
e
y
d
ec
r
ea
s
e
to
6
3
.
1
3
%
an
d
5
6
.
8
8
%
w
h
e
n
t
h
e
te
s
t se
t (
w
h
ich
co
n
ta
in
s
1
6
0
i
m
a
g
es)
i
s
u
s
ed
.
W
it
h
th
e
u
s
e
o
f
Gab
o
r
W
av
elets,
L
DA
an
d
KF
A
R
O
C
p
er
f
o
r
m
a
n
ce
ar
e
b
o
th
9
3
.
3
3
%
w
h
e
n
th
e
e
v
alu
at
in
g
s
et
is
u
s
ed
an
d
th
ey
j
u
s
t
d
ec
r
ea
s
e
to
9
1
.
8
8
%
an
d
9
3
.
1
3
%
w
h
e
n
th
e
tes
t
s
et
is
u
s
ed
.
T
h
is
s
h
o
w
s
t
h
at
L
D
A
b
ased
alg
o
r
ith
m
s
s
till
p
er
f
o
r
m
b
etter
wh
en
t
h
e
n
u
m
b
er
o
f
test
/p
r
o
b
e
is
in
cr
ea
s
ed
.
I
t
ca
n
also
b
e
s
ee
n
th
at
KP
C
A
p
er
f
o
r
m
w
o
r
s
t
h
a
v
i
n
g
t
h
e
h
i
g
h
e
s
t
er
r
o
r
r
ates
(
2
.
6
8
%
w
it
h
Gab
o
r
an
d
8
.
8
0
%
w
ith
o
u
t
Gab
o
r
)
.
I
t
also
h
as
th
e
lo
west
r
ec
o
g
n
itio
n
r
ate.
I
t
p
er
f
o
r
m
s
w
o
r
s
e
t
h
an
P
C
A
b
u
t
o
n
l
y
p
er
f
o
r
m
b
etter
t
h
an
P
C
A
(
f
r
o
m
t
h
e
C
M
C
r
es
u
lts
w
it
h
Gab
o
r
o
n
t
h
e
ev
al
u
ati
n
g
s
et)
w
h
e
n
t
h
e
Gab
o
r
f
ilter
s
i
s
u
s
ed
.
Ov
er
all
t
h
e
lin
e
ar
b
ased
alg
o
r
ith
m
s
till
p
er
f
o
r
m
s
b
e
tter
th
a
n
t
h
e
n
o
n
li
n
ea
r
o
n
es.
T
ab
le
1
.
R
ec
o
g
n
itio
n
r
ates
u
s
i
n
g
d
i
f
f
er
en
t Fac
e
R
ec
o
g
n
itio
n
p
er
f
o
r
m
a
n
ce
m
etr
ices
T
h
e
f
o
llo
w
i
n
g
co
n
c
lu
s
io
n
s
ar
e
d
r
aw
n
f
r
o
m
th
e
r
es
u
lt
s
o
b
tain
ed
f
r
o
m
t
h
e
ex
p
er
i
m
en
t
(
u
n
d
er
eq
u
al
w
o
r
k
i
n
g
co
n
d
itio
n
s
)
:
1
)
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
li
n
ea
r
an
d
n
o
n
li
n
ea
r
alg
o
r
it
h
m
s
d
ep
en
d
s
o
n
s
o
m
e
co
n
d
itio
n
s
.
T
h
ese
ar
e
ex
p
lai
n
ed
b
ello
w
:
a.
T
h
e
n
u
m
b
er
o
f
clas
s
es
o
f
a
f
ac
ial
r
ec
o
g
n
it
io
n
s
y
s
te
m
ca
n
af
f
ec
ts
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
t
y
p
e
o
f
lin
ea
r
a
n
d
n
o
n
l
in
ea
r
al
g
o
r
ith
m
u
s
ed
.
L
D
A
(
a
li
n
ea
r
al
g
o
r
ith
m
)
an
d
K
F
A
(
a
n
o
n
li
n
ea
r
alg
o
r
ith
m
)
ex
p
r
ess
l
y
p
r
o
v
id
es b
est d
is
cr
i
m
i
n
atio
n
a
m
o
n
g
clas
s
es.
b.
T
h
e
p
r
e
p
r
o
c
ess
in
g
u
s
i
n
g
Gab
o
r
f
ilter
s
in
cr
ea
s
es
t
h
e
r
ec
o
g
n
i
tio
n
r
ate
o
f
b
o
th
th
e
lin
ea
r
an
d
n
o
n
lin
ea
r
alg
o
r
ith
m
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
D
ev
elo
p
men
t o
f a
n
E
fficien
t
F
a
ce
R
ec
o
g
n
itio
n
S
ystem
B
a
s
ed
o
n
Lin
ea
r
…
(
A
r
a
o
lu
w
a
S
imileo
lu
F
ila
n
i
)
87
c.
W
h
en
m
o
r
e
test
i
m
a
g
es
ar
e
u
s
ed
af
ter
p
r
ep
r
o
ce
s
s
in
g
w
ith
Gab
o
r
w
av
e
lets
,
t
h
er
e
is
r
ed
u
ctio
n
i
n
r
ec
o
g
n
itio
n
r
ate
o
f
al
l
al
g
o
r
ith
m
s
h
o
w
e
v
er
t
h
e
r
ed
u
ctio
n
i
s
m
o
r
e
f
o
r
th
e
P
C
A
b
ased
a
lg
o
r
ith
m
s
t
h
an
th
e
L
D
A
b
ased
o
n
es.
T
h
is
s
h
o
w
s
t
h
at
th
e
i
n
cr
ea
s
e
i
n
n
u
m
b
er
o
f
test
i
m
a
g
es
ca
n
a
f
f
ec
ts
r
ec
o
g
n
itio
n
r
ate
(
o
f
all
t
h
e
alg
o
r
it
h
m
s
)
n
eg
ativ
el
y
b
u
t
th
e
L
D
A
b
ased
(
cl
ass
ed
b
ased
)
alg
o
r
ith
m
s
ar
e
le
s
s
a
f
f
ec
ted
th
an
t
h
e
P
C
A
b
ased
o
n
es.
2
)
Fro
m
t
h
e
o
v
er
all
r
esu
lts
t
h
e
lin
ea
r
alg
o
r
it
h
m
s
is
b
etter
th
a
n
th
e
n
o
n
li
n
ea
r
o
n
es.
6.
C
O
NCLU
SI
O
N
T
h
e
r
esu
lts
s
h
o
w
t
h
at
t
h
e
n
u
m
b
er
o
f
class
es
a
n
d
tes
t
i
m
ag
e
s
o
f
a
f
ac
ia
l
r
ec
o
g
n
itio
n
s
y
s
te
m
ca
n
h
a
v
e
an
ef
f
ec
t
o
n
th
e
r
ec
o
g
n
itio
n
r
ate
o
f
a
p
ar
ticu
l
ar
alg
o
r
ith
m
u
s
ed
.
I
n
co
r
p
o
r
atin
g
Gab
o
r
im
ag
e
r
ep
r
esen
tatio
n
w
it
h
li
n
ea
r
an
d
n
o
n
li
n
ea
r
alg
o
r
ith
m
s
in
cr
ea
s
e
s
t
h
eir
r
ec
o
g
n
itio
n
r
ate.
L
i
n
ea
r
s
u
b
s
p
ac
e
tech
n
iq
u
es
te
n
d
to
p
er
f
o
r
m
b
etter
th
a
n
th
e
n
o
n
li
n
ea
r
lin
ea
r
o
n
es
f
r
o
m
th
e
r
es
u
lt
o
f
th
e
w
o
r
k
ca
r
r
ied
o
u
t.
T
h
e
r
esear
ch
w
i
ll
b
e
o
f
o
u
t
m
o
s
t
i
m
p
o
r
ta
n
ce
to
a
n
y
o
r
g
an
iza
tio
n
th
at
w
is
h
es
to
d
ev
elo
p
a
f
ac
ial
r
ec
o
g
n
it
io
n
s
y
s
te
m
a
n
d
k
n
o
w
w
h
ic
h
o
f
th
e
f
ac
e
r
ec
o
g
n
i
tio
n
alg
o
r
it
h
m
s
h
a
v
e
a
b
etter
r
ec
o
g
n
itio
n
r
ate.
T
h
is
s
tu
d
y
w
ill
also
b
e
o
f
i
m
m
en
s
e
b
en
ef
it
to
p
r
o
s
p
ec
tiv
e
r
esea
r
ch
er
s
w
h
o
w
o
u
ld
li
k
e
to
u
n
d
er
ta
k
e
s
i
m
i
lar
s
tu
d
ies.
T
h
is
w
o
r
k
i
s
ab
le
to
co
m
p
a
r
e
lin
ea
r
a
n
d
n
o
n
li
n
ea
r
f
ac
e
r
ec
o
g
n
itio
n
al
g
o
r
ith
m
s
p
r
o
d
u
ce
d
.
T
h
e
r
esear
ch
is
o
n
l
y
co
n
ce
r
n
ab
o
u
t
2
D
h
o
lis
tic
f
ac
e
r
ec
o
g
n
itio
n
alg
o
r
ith
m
.
A
n
e
w
d
ev
elo
p
m
e
n
t
ca
n
m
a
k
e
u
s
e
o
f
2
D
lo
ca
l b
ased
ap
p
ea
r
an
ce
f
ac
e
r
ec
o
g
n
itio
n
al
g
o
r
ith
m
s
u
s
in
g
lin
ea
r
an
d
n
o
n
li
n
ea
r
alg
o
r
ith
m
s
.
RE
F
E
R
E
NC
E
S
[1
]
R
S
a
d
y
k
h
o
v
,
I
F
ro
lo
v.
T
h
e
d
e
v
e
l
o
p
me
n
t
fea
tu
re
s
o
f
th
e
f
a
c
e
re
c
o
g
n
it
i
o
n
sy
ste
m
.
I
n
IEE
E
p
r
o
c
e
e
d
in
g
s
o
f
th
e
In
t.
M
u
lt
ico
n
f
.
o
n
Co
m
p
.
S
c
i.
a
n
d
In
f
o
r.
T
e
c
h
.
IM
CS
IT
,
2
0
1
0
:
1
2
1
-
1
2
8
.
[2
]
W
Zh
a
o
,
R
Ch
e
ll
a
p
p
a
,
R
P
J
P
h
il
li
p
s,
A
Ro
se
n
fe
ld
.
F
a
c
e
re
c
o
g
n
it
io
n
:
A
li
tera
tu
re
su
rv
e
y
.
ACM
Co
mp
u
t
i.
S
u
rv
.
2
0
0
3
;
2
(4
)
:
3
9
9
-
4
5
8
.
[3
]
E.
G
.
Ortiz,
BC
Be
c
k
e
r.
F
a
c
e
re
c
o
g
n
it
io
n
f
o
r
w
e
b
-
sc
a
le
d
a
tas
e
t.
C
o
mp
.
Vi
s.
a
n
d
Im
a
g
e
U
n
d
e
rs
ta
n
d
in
g
.
2
0
1
3
;
1
1
8
:
153
-
1
7
0
.
[4
]
M
P
Be
h
a
m
,
S
M
M
Ro
o
m
i.
Fa
c
e
re
c
o
g
n
it
i
o
n
u
si
n
g
a
p
p
e
a
ra
n
c
e
b
a
se
d
a
p
p
ro
a
c
h
:
a
li
ter
a
tu
re
su
rv
e
y
.
In
P
r
o
c
.
IJC
A
In
t.
C
o
n
f
e
re
n
c
e
a
n
d
W
o
rk
sh
o
p
o
n
Re
c
e
n
t
T
re
n
d
s in
T
e
c
h
n
o
l
o
g
y
,
TCE
T
,
2
0
1
2
:
16
-
21.
[5
]
Hu
a
n
g
,
W
,
Yin
,
H
.
On
n
o
n
li
n
e
a
r
d
ime
n
sio
n
a
li
ty
re
d
u
c
ti
o
n
f
o
r
f
a
c
e
re
c
o
g
n
it
io
n
.
Ima
g
e
a
n
d
Vi
s
io
n
Co
m
p
u
ti
n
g
.
2
0
1
2
;
3
0
:
2
5
-
3
6
6
.
[6
]
J
L
u
,
KN
P
lata
n
io
ti
s,
A
N
V
e
n
e
tsa
n
o
p
o
u
l
o
s.
F
a
c
e
re
c
o
g
n
it
io
n
u
sin
g
L
D
A
-
b
a
se
d
a
lg
o
rit
h
m
s.
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
Ne
u
ra
l
Ne
two
rk
s.
2
0
0
3
:
1
4
(
1
):
1
9
5
–
2
0
0
.
[7
]
V
S
tr
u
c
,
F
M
il
h
e
li
c
,
N
P
a
v
e
sic
.
Co
mb
in
in
g
e
x
p
e
rts
fo
r
imp
ro
v
e
d
fa
c
e
v
e
rifi
c
a
ti
o
n
p
e
rfo
rm
a
n
c
e
.
in
P
ro
c
.
IE
EE
Co
n
f
.
ERK,
2
0
0
8
:
2
3
3
-
2
3
6
.
[8
]
V
Ba
lam
u
ru
g
a
n
,
M
S
ri
n
iv
a
sa
n
,
A
V
i
jay
a
n
a
r
a
y
a
n
a
n
,
L
Ba
i.
A
n
e
w
f
a
c
e
re
c
o
g
n
it
io
n
tec
h
n
i
q
u
e
s
u
sin
g
g
a
b
o
r
w
a
v
e
le
t
tran
sf
o
r
m
w
it
h
b
a
c
k
p
ro
p
a
g
a
ti
o
n
n
e
u
ra
l
n
e
tw
o
rk
.
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
C
o
mp
u
ter
Ap
p
li
c
a
ti
o
n
.
2
0
1
2
:
49
(3
):
4
1
-
46
.
[9
]
A
L
i,
S
S
h
a
n
,
W
G
a
o
.
Co
u
p
le
d
Bi
a
s
–
V
a
ria
n
c
e
T
ra
d
e
o
ff
fo
r
Cro
s
s
-
Po
se
Fa
c
e
Rec
o
g
n
it
i
o
n
.
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
Im
a
g
e
P
ro
c
e
ss
in
g
.
2
0
1
2
;
2
1
(1
):
3
0
5
,
3
1
5
.
[1
0
]
V
S
tr
u
c
,
N
P
a
v
e
sic
.
Ga
b
o
r
-
b
a
se
d
k
e
rn
e
l
p
a
rti
a
l
-
lea
st
-
sq
u
a
re
s
d
isc
rim
in
a
ti
o
n
f
e
a
tu
re
s
f
o
r
f
a
c
e
re
c
o
g
n
it
io
n
.
”
In
stit
u
te
o
f
M
a
t
h
e
ma
ti
c
s a
n
d
I
n
f
o
rm
a
ti
c
s
,
2
0
0
9
;
2
0
(
1
)1
1
5
-
1
3
8
.
[1
1
]
B
Zh
a
n
g
,
X
Ch
e
n
,
S
S
h
a
n
,
S
,
W
G
a
o
.
No
n
li
n
e
a
r
fa
c
e
re
c
o
g
n
it
i
o
n
b
a
se
d
o
n
ma
x
imu
m
a
v
e
ra
g
e
m
a
rg
in
c
riter
io
n
.
IEE
E
Co
n
f
.
Co
m
p
u
t.
V
is.
P
a
tt
e
rn
Re
c
o
g
.
,
2
0
0
5
;
1
(
4
):
5
5
4
–
5
5
9
.
[1
2
]
R
Ja
f
ri,
HR
A
r
a
b
n
ia.
A
su
rv
e
y
o
f
f
a
c
e
re
c
o
g
n
it
io
n
tec
h
n
i
q
u
e
s.
J
o
u
rn
a
l
o
f
In
f
o
rm
a
ti
o
n
Pro
c
e
ss
in
g
S
y
ste
ms
.
2
0
0
9
;
5
(2
)
.
[1
3
]
R
S
a
h
a
,
D
Bh
a
tt
a
c
h
a
rjee
.
F
a
c
e
Re
c
o
g
n
it
io
n
Us
i
n
g
E
ig
e
n
f
a
c
e
s
”
In
t.
J
.
Eme
rg
in
g
T
e
c
h
n
o
l
o
g
y
a
n
d
Ad
v
.
En
g
i
n
.
,
2
0
1
3
;
3
(5
):
9
0
-
93
.
[1
4
]
M
Na
sir
(2
0
1
2
,
Ja
n
u
a
ry
2
5
).
Ge
tt
in
g
Rea
d
y
f
o
r
F
a
c
e
Rec
o
g
n
it
io
n
-
L
e
v
e
l
4
a
t
u
to
ri
a
l
[
O
n
l
in
e
]
.
A
v
a
il
a
b
le:
h
tt
p
:
//
f
e
w
tu
to
rials.b
ra
v
e
sites
.
c
o
m
[1
5
]
M
T
u
rk
,
A
P
e
n
tl
a
n
d
.
Ei
g
e
n
f
a
c
e
s
fo
r
Re
c
o
g
n
it
i
o
n
.
J
.
Co
g
n
.
Ne
u
ro
sc
.
,
1
9
9
1
;
1
3
(
1
):
7
1
-
86
.
[1
6
]
A
Ba
n
sa
l,
K
M
e
h
ta,
S
A
ro
ra
,
Ba
n
sa
l,
A
M
e
h
ta,
K
A
ro
ra
,
S
.
Fa
c
e
Rec
o
g
n
i
ti
o
n
Us
in
g
PCA
a
n
d
L
D
A
Al
g
o
rit
h
m,
"
A
d
v
a
n
c
e
d
Co
m
p
u
ti
n
g
&
Co
m
m
u
n
ica
ti
o
n
T
e
c
h
n
o
lo
g
ies
(A
CCT
),
2
0
1
2
S
e
c
o
n
d
In
ter
n
a
ti
o
n
a
l
Co
n
f
e
re
n
c
e
o
n
,
2
0
1
2
:
2
5
1
,
2
5
4
,
7
-
8
.
[1
7
]
H
Yu
,
H,
J
Ya
n
g
.
A
d
irec
t
L
D
A
a
lg
o
rit
h
m
f
o
r
h
ig
h
-
d
im
e
n
sio
n
a
l
d
a
ta
w
it
h
a
p
p
li
c
a
ti
o
n
to
f
a
c
e
re
c
o
g
n
it
io
n
,
”
Pa
tt
e
rn
Rec
o
g
.
,
2
0
0
1
;
4
2
:
2
0
6
7
–
2
0
7
0
.
[1
8
]
MS
Ba
rtl
e
tt
,
J
R
M
o
v
e
ll
a
n
,
T
J
S
e
jn
o
w
sk
i.
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.
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
Ne
u
r
a
l
Ne
two
rk
s,
2
0
0
2
;
1
3
(6
):
1
4
5
0
-
1
4
6
4
.
[1
9
]
K De
lac
,
M
G
r
g
ic,
S
G
rg
i
c
.
S
ta
ti
s
ti
c
s in
fa
c
e
re
c
o
g
n
it
i
o
n
:
a
n
a
lyz
in
g
p
ro
b
a
b
il
it
y
d
istri
b
u
t
io
n
s o
f
PC
A,
ICA
a
n
d
L
DA
p
e
rfo
rm
a
n
c
e
re
su
lt
s
.
Im
a
g
e
a
n
d
S
ig
n
a
l
P
r
o
c
e
ss
in
g
a
n
d
A
n
a
l
y
si
s,
2
0
0
5
.
IS
P
A
2
0
0
5
.
P
r
o
c
e
e
d
in
g
s
o
f
th
e
4
th
In
tern
a
ti
o
n
a
l
S
y
m
p
o
siu
m
o
n
,
2
0
0
5
;
2
8
9
,
2
9
4
,
1
5
-
17
.
[2
0
]
K
De
lac
,
M
G
r
g
ic,
S
G
r
g
ic.
In
d
e
p
e
n
d
e
n
t
c
o
m
p
a
ra
ti
v
e
stu
d
y
o
f
P
CA
,
IC
A
,
a
n
d
L
D
A
o
n
F
ERE
T
d
a
ta
se
t.
In
t.
J
.
o
f
Ima
g
i
n
g
S
y
ste
ms
a
n
d
T
e
c
h
n
o
l
o
g
y
,
2
0
0
6
;
15
(
5
)
:
2
52
-
2
6
0
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
IJ
-
AI
Vo
l.
5
,
No
.
2
,
J
u
n
e
2
0
1
6
:
8
0
–
88
88
[2
1
]
A
F
A
b
a
te,
M
Na
p
p
i,
D
Ricc
o
,
G
S
a
b
a
ti
n
o
.
2
D
a
n
d
3
D
f
a
c
e
re
c
o
g
n
it
io
n
:
A
su
rv
e
y
.
Pa
tt
e
rn
Rec
o
g
n
it
.
2
0
0
7
;
28
:
1
8
8
6
-
1
9
0
.
[2
2
]
X
He
,
S
Ya
n
,
Y
Hu
,
P
Niy
o
g
i,
H
Zh
a
n
g
.
F
a
c
e
re
c
o
g
n
it
io
n
u
sin
g
Lap
lac
ian
f
a
c
e
s.
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
Pa
tt
e
r
n
A
n
a
lys
is
a
n
d
M
a
c
h
i
n
e
In
tell
ig
e
n
c
,
2
0
0
5
;
2
7
(3
)
:
3
2
8
-
3
4
0
.
[2
3
]
C
Ch
e
n
,
K
X
ie.
F
a
c
e
re
c
o
g
n
it
io
n
b
a
se
d
o
n
tw
o
-
d
im
e
n
sio
n
a
l
p
ri
n
c
ip
a
l
c
o
m
p
o
n
e
n
t
a
n
a
ly
sis
a
n
d
k
e
rn
e
l
p
rin
c
ip
a
l
c
o
m
p
o
n
e
n
t
a
n
a
ly
sis.
In
fo
rm
a
ti
o
n
T
e
c
h
n
o
lo
g
y
J
o
u
rn
a
l,
Asia
n
N
e
two
rk
fo
r
S
c
ien
ti
fi
c
In
f
o
rm
a
ti
o
n
IJ
CS
I,
2
0
1
2
;
11
(
12
)
:
1
7
8
1
-
1
7
8
.
[2
4
]
F
Ig
o
r
,
S
Ra
u
f
.
T
h
e
tec
h
n
iq
u
e
s
f
o
r
fa
c
e
re
c
o
g
n
it
i
o
n
wit
h
su
p
p
o
rt
v
e
c
to
r
ma
c
h
in
e
s
.
I
n
P
r
o
c
.
o
f
th
e
In
tern
a
ti
o
n
a
l
M
u
lt
ico
n
f
e
re
n
c
e
o
n
Co
m
p
u
ter S
c
i
e
n
c
e
a
n
d
In
f
o
rm
a
ti
o
n
T
e
c
h
n
o
l
o
g
y
IM
CS
IT
,
2
0
0
9
;
4
:
31
-
3
6
.
[2
5
]
M
A
Ka
sh
e
m
,
M
N AKh
ter,
S
Ah
med
,
A
M
A
la
m
.
Fa
c
e
Rec
o
g
n
it
io
n
S
y
ste
m B
a
se
d
o
n
Pri
n
c
ip
a
l
Co
mp
o
n
e
n
t
An
a
lys
is
(
PCA
)
wit
h
Ba
c
k
Pro
p
a
g
a
ti
o
n
Ne
u
ra
l
Ne
two
rk
s
(
BP
NN)
.
Ca
n
a
d
ian
Jo
u
rn
a
l
o
n
Im
a
g
e
P
ro
c
e
ss
in
g
a
n
d
Co
m
p
u
ter
V
isio
n
,
2
0
1
1
;
2
(
4
):
36
-
45
.
[2
6
]
LL
T
h
o
m
a
s,
C
G
o
p
a
k
u
m
a
r,
AA
T
h
o
m
a
s.
F
a
c
e
R
e
c
o
g
n
it
io
n
b
a
se
d
o
n
G
a
b
o
r
W
a
v
e
let
a
n
d
Ba
c
k
p
ro
p
a
g
a
ti
o
n
Ne
u
ra
l
Ne
tw
o
r.
J
.
S
c
i.
a
n
d
En
g
.
Res
e
a
rc
h
,
2
0
1
3
;
4
(6
)
:
2
1
1
4
-
2
1
1
9
.
[2
7
]
S
h
e
n
,
L
,
Ba
i,
L
F
a
irh
u
rst
,
M
.
(
2
0
0
6
).
G
a
b
o
r
w
a
v
e
lets
a
n
d
G
e
n
e
ra
l
Disc
ri
m
in
a
n
t
A
n
a
l
y
sis
f
o
r
fa
c
e
id
e
n
ti
f
ica
ti
o
n
a
n
d
ve
ri
f
ica
ti
o
n
.
Im
a
g
e
a
n
d
Visio
n
Co
m
p
u
ti
n
g
.
Ima
g
e
a
n
d
Vi
si
o
n
Co
mp
u
ti
n
g
,
2
0
0
7
;
25
:
5
5
3
–
5
6
3
.
[2
8
]
C.
L
iu
,
H
W
e
c
h
sle
r.
Ga
b
o
r
f
e
a
tu
r
e
b
a
se
d
c
las
si
f
ica
ti
o
n
u
sin
g
th
e
e
n
h
a
n
c
e
d
f
ish
e
r
li
n
e
a
r
d
isc
rim
in
a
n
t
m
o
d
e
l
f
o
r
f
a
c
e
re
c
o
g
n
it
io
n
,
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
Ima
g
e
Pro
c
e
ss
i
ng
,
2
0
0
2
;
1
1
(
4
):
4
6
7
-
4
7
6
.
[2
9
]
M
Ya
n
g
.
Ke
rn
e
l
e
ig
e
n
f
a
c
e
s
v
s.
k
e
rn
e
l
f
ish
e
rfa
c
e
s:
F
a
c
e
re
c
o
g
n
it
io
n
u
si
n
g
k
e
rn
e
l
m
e
th
o
d
s
.
i
n
P
ro
c
e
e
d
in
g
s
5
t
h
IE
EE
In
t.
C
o
n
f
.
o
n
A
u
to
m
a
ti
c
F
a
c
e
a
n
d
G
e
stu
re
Re
c
o
g
.
,
2002
:
2
1
5
-
2
2
0
,
2
1
-
21
.
[3
0
]
C
L
iu
.
Ca
p
it
a
li
z
e
o
n
Di
m
e
n
sio
n
a
li
ty
In
c
re
a
sin
g
T
e
c
h
n
iq
u
e
s
f
o
r
I
m
p
ro
v
in
g
F
a
c
e
R
e
c
o
g
n
it
io
n
.
IEE
E
T
ra
n
sa
c
ti
o
n
s o
n
Pa
tt
e
rn
A
n
a
lys
is
a
n
d
M
a
c
h
i
n
e
In
t
e
ll
ig
e
n
c
e
,
2
0
0
6
;
2
8
(
5
):
7
2
5
-
7
3
7
.
[3
1
]
A
S
M
o
o
n
,
R
S
riv
a
sta
v
a
,
Y
P
a
n
d
e
y
,
(2
0
1
3
).
Im
p
a
c
t
o
f
k
e
rn
e
l
f
ish
e
r
a
n
a
l
y
sis
m
e
th
o
d
o
n
f
a
c
e
re
c
o
g
n
it
io
n
.
In
ter
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
E
n
g
in
e
e
rin
g
a
n
d
A
d
v
a
n
c
e
d
T
e
c
h
n
o
l
o
g
y
,
IJ
EA
T
,
2
0
1
3
;
2
(
3
)
.
[3
2
]
RM
Ib
ra
h
im
,
F
EZ
A
b
o
u
-
Ch
a
d
i,
A
S
S
a
m
r
a
.
P
las
ti
c
S
u
rg
e
r
y
F
a
c
e
R
e
c
o
g
n
it
io
n
:
A
c
o
m
p
a
ra
ti
v
e
S
tu
d
y
o
f
P
e
rf
o
rm
a
n
c
e
.
In
ter
n
a
ti
o
n
a
l
J
o
u
r
n
a
l
o
f
Co
m
p
u
ter
S
c
ien
c
e
IJ
CS
I
.
2
0
1
3
;
5
(2
)
.
[3
3
]
NSS
Ma
r,
CB
F
o
o
k
e
s,
P
KD
V
Y
a
rlag
a
d
d
a
,
(2
0
1
2
).
S
o
l
d
e
r
jo
i
n
t
d
e
fec
ts
c
la
ss
if
ica
ti
o
n
u
sin
g
t
h
e
L
o
g
-
Ga
b
o
r
Fi
lt
e
r,
th
e
Disc
re
te
W
a
v
e
let
T
ra
n
sfo
rm
a
n
d
t
h
e
Disc
re
te
Co
sin
e
T
r
a
n
sf
o
rm
,
”
In
In
ter
n
a
ti
o
n
a
l
Co
n
f
e
re
n
c
e
o
n
A
d
v
a
n
c
e
s
in
M
e
c
h
a
n
ica
l
a
n
d
B
u
il
d
in
g
,
2
0
1
2
:
9
-
11
.
[3
4
]
W
Be
n
g
io
,
J
M
a
rit
h
o
z
.
T
h
e
Exp
e
c
ted
Per
fo
rm
a
n
c
e
Cu
rv
e
:
A
Ne
w
Ass
e
ss
e
me
n
t
M
e
a
su
r
e
fo
r
Per
so
n
a
l
Au
th
e
n
ti
c
a
ti
o
n
.
I
n
P
ro
c
e
e
d
i
n
g
s o
f
Od
y
ss
e
y
.
T
h
e
S
p
e
a
k
e
r
a
n
d
L
a
n
g
u
a
g
e
Re
c
o
g
n
it
io
n
W
o
rk
sh
o
p
,
2
0
0
4
:
2
7
9
-
2
8
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.