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.
14
,
No
.
2
,
J
u
n
e
2
0
2
5
,
p
p
.
430
~
4
3
8
I
SS
N:
2252
-
8
8
1
4
,
DOI
:
1
0
.
1
1
5
9
1
/ijaas
.
v
14
.
i
2
.
pp
430
-
4
3
8
430
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
a
s
.
ia
esco
r
e.
co
m
Co
ntent
bas
ed im
a
g
e re
trieva
l usin
g
vis
ua
l geo
metric
g
ro
up1
9
with
J
a
ccard s
imi
la
rity mea
sure
Ra
j
a
t
h Ar
a
k
er
e
Na
ra
y
a
na
s
wa
m
y
1,
2
,
Vidy
a
la
k
s
hm
i K
ris
hn
e
G
o
wda
2
1
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
G
S
S
S
I
n
st
i
t
u
t
e
o
f
En
g
i
n
e
e
r
i
n
g
a
n
d
Te
c
h
n
o
l
o
g
y
f
o
r
W
o
me
n
,
M
y
s
u
r
u
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
d
a
n
E
n
g
i
n
e
e
r
i
n
g
(
D
a
t
a
S
c
i
e
n
c
e
)
,
S
r
i
S
i
d
d
h
a
r
t
h
a
I
n
st
i
t
u
t
e
o
f
T
e
c
h
n
o
l
o
g
y
(
A
C
o
n
s
t
i
t
u
e
n
t
C
o
l
l
e
g
e
o
f
S
r
i
S
i
d
d
h
a
r
t
h
a
A
c
a
d
e
m
y
o
f
H
i
g
h
e
r
E
d
u
c
a
t
i
o
n
)
,
T
u
m
a
k
u
r
u
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Oct
23
,
2
0
2
4
R
ev
is
ed
Ap
r
6
,
2
0
2
5
Acc
ep
ted
Ap
r
23
,
2
0
2
5
Co
n
ten
t
-
b
a
se
d
ima
g
e
re
tri
e
v
a
l
(CBIR)
is
a
n
imp
o
rtan
t
re
se
a
rc
h
a
re
a
th
a
t
fo
c
u
se
s
o
n
e
m
e
rg
i
n
g
tec
h
n
iq
u
e
s
fo
r
h
a
n
d
li
n
g
larg
e
ima
g
e
c
o
ll
e
c
ti
o
n
s
a
n
d
e
n
a
b
li
n
g
e
fficie
n
t
re
tri
e
v
a
l.
T
h
e
m
a
in
c
h
a
ll
e
n
g
e
o
f
ima
g
e
re
tri
e
v
a
l
is
to
e
x
trac
t
re
lev
a
n
t
fe
a
tu
re
v
e
c
to
rs
fo
r
ima
g
e
d
e
sc
rip
ti
o
n
.
Th
e
re
f
o
re
,
v
isu
a
l
g
e
o
m
e
tri
c
g
r
o
u
p
1
9
(VG
G
1
9
)
wi
th
Ja
c
c
a
rd
is
p
ro
p
o
se
d
in
t
h
is
re
s
e
a
rc
h
fo
r
CBIR.
Th
e
VG
G
1
9
a
ll
o
ws
to
c
a
p
tu
re
o
f
h
iera
rc
h
ica
l
fe
a
tu
re
s,
a
n
d
it
is
a
p
p
ro
p
riate
fo
r
tex
t
u
re
a
n
d
fi
n
e
d
e
tail
c
h
a
ra
c
teristics
.
It
e
n
a
b
les
to
p
ro
d
u
c
ti
o
n
o
f
r
o
b
u
st
a
n
d
d
isc
rimin
a
ti
v
e
fe
a
tu
re
re
p
re
se
n
tatio
n
s
b
e
c
a
u
se
o
f
n
u
m
e
ro
u
s
c
o
n
v
o
l
u
ti
o
n
a
l
lay
e
rs.
Th
e
Ja
c
c
a
rd
is
u
ti
li
z
e
d
a
s
a
sim
il
a
rit
y
m
e
a
su
re
a
m
o
n
g
fe
a
tu
re
v
e
c
to
rs
b
y
c
o
m
p
a
rin
g
t
h
e
u
n
io
n
a
n
d
in
ters
e
c
ti
o
n
o
f
fe
a
tu
re
se
ts.
It
is
h
e
lp
fu
l
t
o
m
a
n
a
g
e
sp
a
rse
a
n
d
h
i
g
h
e
r
-
d
ime
n
sio
n
a
l
d
a
ta
th
a
t
p
ro
v
id
e
s
a
fa
st
a
n
d
a
c
c
u
ra
te
ima
g
e
re
tri
e
v
a
l
p
ro
c
e
ss
.
T
h
e
Ca
lt
e
c
h
2
5
6
a
n
d
Co
re
l
1
K
d
a
tas
e
ts
a
re
c
o
n
sid
e
re
d
a
n
d
it
is
p
re
p
r
o
c
e
ss
e
d
b
y
ima
g
e
re
siz
in
g
a
n
d
n
o
rm
a
li
z
a
ti
o
n
.
T
h
e
ra
w
ima
g
e
s
a
re
re
siz
e
d
to
e
n
su
re
d
a
tas
e
t
sim
il
a
rit
y
a
n
d
n
o
rm
a
li
z
e
d
i
n
to
th
e
ra
n
g
e
o
f
0
a
n
d
1
.
T
h
e
m
e
tri
c
s
su
c
h
a
s
re
c
a
ll
,
f
-
m
e
a
su
re
,
a
n
d
a
v
e
ra
g
e
p
re
c
isio
n
a
re
u
se
d
to
c
a
lcu
late
th
e
VG
G
1
9
-
Ja
c
c
a
rd
p
e
rfo
rm
a
n
c
e
.
Th
e
VG
G
1
9
-
Ja
c
c
a
rd
a
c
h
iev
e
s
a
v
e
ra
g
e
p
re
c
isio
n
o
f
9
9
.
0
a
n
d
9
9
.
8
%
f
o
r
Ca
lt
e
c
h
2
5
6
a
n
d
Co
r
e
l
1
K
d
a
tas
e
ts
c
o
m
p
a
re
d
to
th
e
two
-
sta
g
e
CBIR
tec
h
n
iq
u
e
.
K
ey
w
o
r
d
s
:
C
o
n
ten
t
-
b
ased
im
ag
e
r
etr
iev
al
Hig
h
er
-
d
im
e
n
s
io
n
al
d
ata
J
ac
ca
r
d
No
r
m
aliza
tio
n
Vis
u
al
g
eo
m
etr
ic
g
r
o
u
p
1
9
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
:
R
ajath
Ar
ak
er
e
Nar
ay
an
aswa
m
y
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
in
ee
r
in
g
GSSS
I
n
s
titu
te
o
f
E
n
g
in
ee
r
in
g
an
d
T
ec
h
n
o
lo
g
y
f
o
r
W
o
m
en
KR
S
R
o
ad
,
Me
tag
alli,
My
s
u
r
u
-
5
7
0
0
1
6
,
I
n
d
ia
E
m
ail:
r
ajath
1
9
1
9
8
5
@
g
m
ail.
c
o
m
1.
I
NT
RO
D
UCT
I
O
N
C
o
n
ten
t
-
b
ased
im
ag
e
r
etr
ie
v
al
(
C
B
I
R
)
is
th
e
p
r
o
ce
s
s
o
f
r
etr
i
ev
in
g
im
ag
es
r
elev
a
n
t
to
a
q
u
er
y
im
ag
e
f
r
o
m
a
n
ex
ten
s
iv
e
n
u
m
b
e
r
o
f
v
is
u
al
co
n
ten
ts
[
1
]
.
I
t
is
u
s
ed
t
o
s
ea
r
ch
f
o
r
f
ea
tu
r
e
r
ep
r
esen
tatio
n
o
f
a
n
im
ag
e
to
r
etr
iev
e
r
an
k
ed
im
a
g
e
co
llectio
n
b
ased
o
n
its
s
im
ilar
ity
to
t
h
e
q
u
er
y
in
s
tan
ce
[
2
]
.
B
ased
o
n
th
e
v
is
u
al
co
n
ten
t
o
f
a
q
u
er
y
im
ag
e,
C
B
I
R
s
ea
r
ch
es
f
o
r
r
elate
d
im
ag
es,
with
ap
p
licatio
n
s
in
ar
e
as
s
u
ch
a
s
o
n
lin
e
s
h
o
p
p
in
g
,
r
em
o
te
d
iag
n
o
s
tics
,
an
d
f
ac
ial
r
ec
o
g
n
itio
n
[
3
]
.
Du
e
to
th
e
g
r
o
win
g
d
e
m
an
d
f
o
r
im
ag
e
-
b
as
ed
d
ata
r
etr
iev
al
s
y
s
tem
s
,
C
B
I
R
h
as
b
ec
o
m
e
a
lead
in
g
tech
n
o
l
o
g
y
in
th
is
f
ield
[
4
]
.
I
t
tak
es
m
o
s
t
s
ig
n
if
ican
t
an
d
b
r
o
ad
ap
p
licatio
n
s
in
n
u
m
er
o
u
s
ar
e
as
s
u
ch
as
ar
ch
itectu
r
e
d
esig
n
,
ed
u
ca
tio
n
,
m
ed
ical
s
cien
ce
,
th
e
Dep
ar
tm
en
t
o
f
J
u
s
tice
,
a
n
d
Militar
y
af
f
air
s
[
5
]
.
I
m
ag
e
s
im
ilar
ity
p
la
y
s
a
s
i
g
n
if
ican
t
r
o
le
in
th
e
C
B
I
R
s
y
s
tem
wh
ich
e
n
s
u
r
es
ac
cu
r
ate
m
at
ch
i
n
g
an
d
en
h
a
n
ce
s
th
e
o
v
er
all
s
ea
r
ch
ex
p
er
i
en
ce
[
6]
.
T
h
e
u
s
er
s
u
b
m
its
th
e
q
u
er
y
,
an
d
th
e
s
y
s
tem
r
etr
iev
es
im
ag
es
th
r
o
u
g
h
th
e
s
im
ilar
ity
cr
iter
io
n
in
d
escen
d
in
g
way
[
7
]
.
T
h
e
d
u
al
f
u
n
d
am
e
n
tal
s
tep
s
o
f
ea
ch
C
B
I
R
ar
e
f
ea
tu
r
e
ex
tr
ac
t
io
n
an
d
s
im
ilar
ity
m
ea
s
u
r
em
e
n
ts
[
8
]
.
T
h
e
C
B
I
R
h
as
n
u
m
er
o
u
s
u
p
g
r
ad
es
wh
ic
h
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2252
-
8
8
1
4
C
o
n
ten
t b
a
s
ed
ima
g
e
r
etri
ev
a
l
u
s
in
g
visu
a
l g
eo
metric g
r
o
u
p
1
9
…
(
R
a
ja
th
A
r
a
ke
r
e
N
a
r
a
ya
n
a
s
w
a
my
)
431
ar
e
d
esig
n
ed
to
e
n
h
an
ce
its
ef
f
icien
cy
an
d
r
etr
iev
al
p
e
r
f
o
r
m
an
ce
[
9
]
.
T
h
e
C
B
I
R
s
ch
em
es
co
n
ce
n
tr
ate
o
n
p
r
o
m
in
e
n
t
im
ag
e
f
ea
t
u
r
es
wh
ich
en
h
an
ce
s
th
e
r
etr
iev
al
ac
cu
r
ac
y
an
d
r
elev
a
n
cy
ac
co
r
d
i
n
g
to
q
u
er
y
s
ea
r
c
h
[
1
0
]
.
T
h
e
m
ajo
r
p
r
o
b
lem
r
elat
ed
to
C
B
I
R
is
to
g
et
s
ig
n
if
ica
n
t
d
ata
f
r
o
m
r
aw
d
ata
f
o
r
r
em
o
v
in
g
th
e
s
em
a
n
tic
g
ap
[
1
1
]
.
I
t
d
em
o
n
s
tr
ates
th
e
d
if
f
er
en
ce
b
etwe
en
h
ig
h
-
le
v
el
r
ep
r
esen
tatio
n
s
o
f
im
ag
es
a
n
d
l
o
w
-
lev
el
co
n
ce
p
ts
.
T
h
e
v
is
u
al
f
ea
tu
r
e
r
ep
r
esen
ts
th
e
ex
tr
ac
ted
im
ag
es
wh
ich
ar
e
m
atch
ed
th
r
o
u
g
h
o
t
h
er
s
in
d
ata
to
id
en
tify
s
im
ilar
im
ag
es
[
1
2
]
.
T
h
e
im
a
g
e
f
ea
tu
r
es
r
ep
r
esen
t
t
h
e
co
l
o
r
o
r
s
h
ap
e
d
is
tr
ib
u
tio
n
o
f
t
h
e
im
ag
e.
T
h
e
s
h
ap
e
-
b
ased
im
ag
e
r
etr
iev
al
u
tili
ze
s
m
o
m
en
t o
r
e
d
g
es wh
ile
co
lo
r
-
b
ased
r
etr
iev
al
u
tili
ze
s
a
h
is
to
g
r
am
o
f
im
a
g
e
p
ix
el
v
alu
es
[
1
3
]
.
R
ec
en
tly
v
a
r
io
u
s
g
lo
b
al,
lo
ca
l
,
an
d
d
ee
p
lear
n
in
g
(
DL
)
f
ea
tu
r
es
h
av
e
b
ee
n
d
ev
elo
p
ed
f
o
r
C
B
I
R
b
y
co
n
s
id
er
in
g
v
ar
i
o
u
s
m
id
-
le
v
el
p
r
esen
tatio
n
s
an
d
v
is
u
al
p
r
o
p
er
ties
lik
e
co
lo
r
,
s
h
a
p
e
,
an
d
te
x
tu
r
e.
T
h
e
m
ain
p
r
o
b
lem
in
C
B
I
R
i
s
to
ca
lcu
l
ate
th
e
r
elev
an
ce
o
f
q
u
er
y
im
ag
es
b
ased
o
n
d
ataset
im
ag
es
[
1
4
]
,
[
1
5
]
.
Mo
s
t
o
f
th
e
ex
is
tin
g
r
esear
ch
m
o
s
tly
f
o
cu
s
ed
o
n
m
atch
in
g
s
im
ilar
im
ag
es
th
r
o
u
g
h
lev
er
a
g
i
n
g
m
u
lti
-
v
is
u
al
f
ea
tu
r
es
[
1
6
]
.
I
n
th
is
co
n
tex
t,
s
ev
er
al
ap
p
r
o
ac
h
es
r
elate
d
to
s
o
m
e
im
ag
e
r
etr
iev
al
m
eth
o
d
s
th
at
h
av
e
b
ee
n
u
s
ed
in
r
ec
en
t
y
ea
r
s
ar
e
d
is
cu
s
s
ed
alo
n
g
with
t
h
eir
lim
itatio
n
s
.
T
a
h
er
i
et
a
l
.
[
1
7
]
s
u
g
g
ested
a
c
o
m
b
in
atio
n
o
f
lo
w
-
lev
el
f
ea
tu
r
es
an
d
a
d
ee
p
B
o
lt
zm
an
n
m
ac
h
in
e
f
o
r
C
B
I
R
(
L
B
-
C
B
I
R
)
.
T
h
e
f
ea
tu
r
e
v
ec
to
r
i
n
teg
r
ated
lo
w
-
le
v
el
an
d
m
i
d
-
lev
el
f
ea
tu
r
es.
T
h
e
lo
w
-
lev
el
f
ea
tu
r
e
ex
tr
ac
tio
n
co
n
tain
s
s
h
a
p
e,
c
o
lo
r
,
an
d
tex
tu
r
e
wh
ich
a
r
e
ac
co
m
p
lis
h
ed
th
r
o
u
g
h
a
u
to
-
c
o
r
r
elo
g
r
am
,
Gab
o
r
wa
v
elet
tr
an
s
f
o
r
m
,
a
n
d
m
u
lti
-
lev
el
f
r
ac
tal
d
im
en
s
io
n
.
T
h
e
m
id
-
lev
el
f
ea
tu
r
es
ar
e
e
x
tr
ac
ted
b
y
a
d
ee
p
B
o
ltzm
an
n
m
ac
h
in
e
.
T
h
e
L
B
-
C
B
I
R
h
as
b
etter
p
er
f
o
r
m
an
ce
b
ec
au
s
e
o
f
its
lo
w
-
lev
el
f
e
atu
r
e
ex
tr
ac
tio
n
.
Ho
wev
e
r
,
it
d
o
es
n
o
t
co
n
ce
n
tr
ate
an
d
ex
tr
ac
ts
th
e
h
ig
h
-
lev
e
l
f
ea
tu
r
es
wh
ich
af
f
ec
t
th
e
m
o
d
el’
s
f
lex
ib
ilit
y
.
Fad
ae
i
et
a
l
.
[
1
8
]
p
r
esen
ted
a
m
u
lti
-
s
ca
le
av
er
ag
in
g
lo
ca
l
b
in
ar
y
p
atter
n
s
f
o
r
C
B
I
R
.
I
n
itially
,
th
e
im
ag
e
was
cr
ea
ted
in
v
ar
i
o
u
s
s
ca
les
,
an
d
tex
tu
r
e
f
ea
tu
r
es we
r
e
ex
tr
ac
ted
f
r
o
m
th
e
im
ag
e
s
ca
le.
L
ast
ly
,
ex
tr
a
cted
f
ea
tu
r
es
o
f
v
ar
io
u
s
r
a
n
g
e
s
ar
e
in
teg
r
ated
to
c
r
ea
te
th
e
last
f
ea
tu
r
e
v
ec
to
r
.
E
v
er
y
lo
ca
l
p
atter
n
s
u
f
f
er
s
f
r
o
m
m
ajo
r
d
r
awb
ac
k
s
in
th
at
i
t
d
o
es
n
o
t
h
an
d
le
b
asic
im
ag
e
d
ata.
I
t
ad
d
r
ess
ed
o
v
er
f
itti
n
g
an
d
g
e
n
er
aliza
tio
n
is
s
u
es,
b
u
t
th
e
s
im
ilar
ity
was
n
o
t
m
ea
s
u
r
e
d
t
h
er
eb
y
r
ed
u
cin
g
th
e
C
B
I
R
p
er
f
o
r
m
an
ce
.
W
an
g
et
a
l
.
[
1
9
]
in
tr
o
d
u
ce
d
a
two
-
s
tag
e
C
B
I
R
b
y
s
p
ar
s
e
r
ep
r
esen
tatio
n
an
d
f
ea
tu
r
e
f
u
s
io
n
.
T
h
e
g
en
er
alize
d
s
ea
r
ch
tr
ee
(
G
I
ST)
f
ea
tu
r
es
ar
e
p
r
im
a
r
ily
ap
p
lied
to
r
etr
ie
v
e
im
ag
es
with
t
h
e
s
am
e
s
ce
n
e
d
ata
b
y
ca
lcu
latin
g
C
an
b
er
r
a
d
is
tan
c
e.
T
h
en
,
s
p
ar
s
e
co
d
in
g
an
d
p
o
o
lin
g
f
ea
tu
r
es
ar
e
ap
p
lied
to
attain
a
s
p
ar
s
e
p
r
esen
tatio
n
o
f
lo
ca
l
f
ea
tu
r
es
f
r
o
m
r
etr
iev
al
r
esu
lts
.
At
last
,
th
e
E
u
clid
ea
n
d
is
tan
ce
is
ap
p
lied
to
ca
lc
u
late
s
p
ar
s
e
f
ea
tu
r
e
v
ec
to
r
s
im
ilar
ity
to
o
b
tain
b
etter
r
etr
iev
al
r
es
u
lts
.
Nev
er
th
eless
,
it
f
ail
s
to
d
etec
t
th
e
ir
r
eg
u
lar
s
h
ap
e
im
ag
es
an
d
r
ed
u
ce
s
th
e
in
ter
p
r
etab
ilit
y
wh
i
ch
af
f
ec
ts
th
e
p
er
f
o
r
m
an
ce
o
f
C
B
I
R
.
Gee
th
a
et
a
l
.
[
2
0
]
d
ev
elo
p
e
d
an
im
ag
e
r
etr
iev
al
f
o
r
ex
tr
ac
tin
g
lo
ca
l
f
ea
tu
r
es
th
at
d
ep
en
d
ed
o
n
a
co
m
b
in
atio
n
o
f
s
ca
le
-
in
v
a
r
ian
t
f
ea
tu
r
e
tr
an
s
f
o
r
m
(
SIFT
)
an
d
K
AZ
E
.
T
h
e
s
tr
en
g
th
o
f
th
e
l
o
ca
l
f
ea
tu
r
e
d
escr
ip
to
r
SIFT
was
co
m
p
lem
en
ted
th
r
o
u
g
h
th
e
g
lo
b
al
f
ea
tu
r
e
d
e
s
cr
ip
to
r
KAZ
E
.
T
h
e
SIFT
f
o
cu
s
ed
o
n
th
e
wh
o
le
im
ag
e
ar
ea
b
y
f
ea
tu
r
i
n
g
f
in
e
p
o
in
ts
an
d
p
o
n
d
er
s
o
f
KAZ
E
o
n
b
o
u
n
d
a
r
y
d
etails.
T
h
e
c
o
m
b
in
atio
n
o
f
lo
ca
l
an
d
g
lo
b
al
f
ea
tu
r
e
d
escr
ip
to
r
s
en
h
an
ce
d
a
n
im
ag
e
r
etr
iev
al
t
h
at
h
as
a
d
iv
er
s
e
class
if
icatio
n
o
f
s
em
an
tics
an
d
s
u
p
p
o
r
ts
t
o
attain
g
o
o
d
r
esu
lts
in
h
u
g
e
-
s
ca
le
r
etr
iev
al.
Ho
we
v
er
,
th
e
m
eth
o
d
r
eq
u
ir
ed
a
h
ig
h
am
o
u
n
t
o
f
o
b
jects
to
s
ea
r
ch
.
R
ah
b
ar
an
d
T
ah
er
i
[
2
1
]
im
p
lem
en
ted
a
tr
ip
let
lo
s
s
f
u
n
ctio
n
-
b
ased
b
in
a
r
y
c
r
o
s
s
-
en
tr
o
p
y
f
o
r
C
B
I
R
.
I
t
en
ab
les
d
ee
p
m
etr
ic
lear
n
in
g
an
d
cr
ea
tes
d
is
cr
im
in
ativ
e
f
e
atu
r
e
s
p
ac
e
with
th
e
h
ig
h
est
d
is
cr
im
in
ativ
e
am
o
n
g
class
es
an
d
d
is
tan
ce
.
T
h
e
im
ag
e
f
ea
tu
r
es
ar
e
ex
tr
ac
ted
th
r
o
u
g
h
p
r
e
-
tr
ain
ed
C
NN.
T
h
en
,
t
h
e
Siam
ese
n
etwo
r
k
w
as
tr
ain
ed
t
o
b
u
ild
d
is
cr
im
in
ativ
e
f
ea
tu
r
e
s
p
ac
e
b
y
d
ee
p
m
etr
ic
n
etwo
r
k
.
T
h
e
m
o
d
el
d
o
es
n
o
t
ex
tr
ac
t
m
a
n
y
in
f
o
r
m
ativ
e
an
d
h
ier
ar
ch
ical
p
atter
n
s
d
u
e
to
v
a
n
is
h
in
g
g
r
ad
ien
t
is
s
u
es.
Ho
wev
er
,
its
g
en
er
aliza
tio
n
p
er
f
o
r
m
an
ce
was
r
ed
u
ce
d
,
an
d
b
r
i
d
g
in
g
th
e
s
em
an
tic
g
a
p
p
r
o
b
lem
o
cc
u
r
r
ed
am
o
n
g
lo
w
-
lev
el
an
d
h
ig
h
-
le
v
el
s
em
an
tic
f
ea
tu
r
es.
Fro
m
th
e
ab
o
v
e
an
al
y
s
is
,
th
e
ex
is
tin
g
tech
n
iq
u
es
h
av
e
li
m
itatio
n
s
s
u
ch
as
n
o
t
co
n
c
en
tr
atin
g
an
d
ex
tr
ac
tin
g
th
e
h
ig
h
-
lev
el
f
ea
t
u
r
es
wh
ich
af
f
ec
t
th
e
m
o
d
el’
s
f
lex
ib
ilit
y
.
T
h
e
s
im
ilar
ity
was
n
o
t
m
ea
s
u
r
ed
th
er
eb
y
r
e
d
u
cin
g
th
e
C
B
I
R
p
er
f
o
r
m
a
n
ce
.
Stru
g
g
les
to
d
ete
ct
ir
r
eg
u
lar
s
h
ap
e
im
ag
es
r
ed
u
ce
in
ter
p
r
etab
ilit
y
an
d
r
eq
u
ir
e
a
lar
g
e
n
u
m
b
er
o
f
o
b
jects
to
s
ea
r
ch
.
T
h
e
g
en
er
aliza
tio
n
p
er
f
o
r
m
a
n
ce
was
r
ed
u
ce
d
an
d
b
r
id
g
in
g
th
e
s
em
an
tic
g
ap
p
r
o
b
lem
o
cc
u
r
r
ed
a
m
o
n
g
lo
w
-
lev
el
an
d
h
ig
h
-
lev
el
s
em
an
tic
f
ea
tu
r
es.
T
o
tack
le
th
ese
lim
itatio
n
s
,
th
e
v
is
u
al
g
eo
m
e
tr
ic
g
r
o
u
p
1
9
(
VGG1
9
)
with
J
ac
ca
r
d
is
p
r
o
p
o
s
ed
in
t
h
is
r
esear
ch
.
T
h
e
k
ey
co
n
tr
ib
u
tio
n
s
ar
e
s
tated
s
f
o
llo
ws
:
th
e
VGG1
9
i
s
s
e
lecte
d
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
wh
ic
h
ef
f
ec
tiv
ely
ca
p
tu
r
es
a
h
ier
ar
ch
ical
f
ea
tu
r
e
with
m
o
r
e
in
f
o
r
m
atio
n
in
th
e
im
ag
e
b
ec
au
s
e
o
f
n
u
m
er
o
u
s
co
n
v
o
l
u
tio
n
al
lay
er
s
.
I
t
is
u
s
ef
u
l
in
p
r
o
d
u
cin
g
f
in
e
an
d
lay
er
ed
v
is
io
n
ch
ar
ac
ter
is
tics
wh
ich
is
h
elp
f
u
l
in
m
atch
in
g
im
ag
es
at
C
B
I
R
.
J
ac
ca
r
d
is
u
s
ed
to
co
m
p
ar
e
th
e
s
im
ilar
ity
an
d
im
ag
e
d
iv
er
g
en
ce
o
f
d
if
f
er
e
n
t
s
am
p
les
d
u
e
to
its
ef
f
ec
tiv
en
ess
.
I
t
en
ab
les
b
etter
-
d
is
cr
im
in
atin
g
im
ag
es
wh
ich
h
av
e
s
e
m
an
tic
co
n
ten
t
at
a
h
ig
h
er
lev
el.
T
h
is
r
esear
ch
p
ap
er
is
s
y
s
tem
atize
d
as:
s
ec
tio
n
2
d
escr
ib
es
th
e
d
etails
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
.
Sectio
n
3
ex
p
lain
s
th
e
r
esu
lts
an
d
d
is
cu
s
s
io
n
. T
h
e
c
o
n
clu
s
io
n
o
f
th
is
r
esear
ch
p
ap
e
r
is
g
iv
e
n
in
s
ec
tio
n
4
.
2.
RE
S
E
ARCH
M
E
T
H
O
D
T
h
e
VGG1
9
with
J
ac
ca
r
d
is
p
r
o
p
o
s
ed
f
o
r
C
B
I
R
s
y
s
tem
.
Data
s
et
s
u
ch
as
C
altec
h
2
5
6
an
d
C
o
r
el
1
K
ar
e
p
r
ep
o
s
s
ess
ed
b
y
u
s
in
g
im
a
g
e
r
esizin
g
an
d
n
o
r
m
aliza
tio
n
.
I
t
is
ex
tr
ac
ted
b
y
u
s
in
g
VGG1
9
wh
ich
g
en
e
r
ates
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.
14
,
No
.
2
,
J
u
n
e
2
0
2
5
:
4
3
0
-
438
432
a
f
ea
tu
r
e
v
ec
to
r
th
at
p
r
esen
ts
th
e
k
ey
v
is
u
al
ch
a
r
ac
ter
is
tics
o
f
im
ag
es.
T
h
en
,
th
e
s
im
ilar
it
y
is
ca
lcu
lated
b
y
J
ac
ca
r
d
,
an
d
t
h
e
im
ag
es a
r
e
r
et
r
iev
ed
.
Fig
u
r
e
1
d
en
o
tes th
e
p
r
o
ce
s
s
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
.
Fig
u
r
e
1
.
Pro
ce
s
s
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
l
o
g
y
2
.
1
.
Da
t
a
s
et
T
h
is
r
esear
ch
u
tili
ze
s
th
e
C
a
ltech
2
5
6
a
n
d
C
o
r
el
1
K
d
at
asets
wh
ich
ar
e
d
etailed
wit
h
th
e
to
tal
n
u
m
b
er
o
f
class
es
an
d
i
m
ag
e
s
.
T
h
e
C
altec
h
2
5
6
d
ataset
[
2
2
]
h
as
3
0
,
6
0
7
im
ag
es
a
cr
o
s
s
2
5
6
class
es,
ev
er
y
class
h
as
a
m
in
im
u
m
o
f
8
0
im
ag
es.
T
h
e
d
ataset
in
clu
d
es
v
ar
io
u
s
class
es
s
u
ch
as
r
in
g
s
,
b
r
a
in
s
,
an
d
d
ia
m
o
n
d
s
.
T
h
e
C
o
r
el
1
K
d
ataset
[
2
3
]
h
a
s
1
0
,
0
0
0
im
ag
es
ac
r
o
s
s
1
,
0
0
0
class
es,
ev
er
y
class
h
as
1
0
0
i
m
ag
es.
T
h
e
d
ataset
in
clu
d
es
v
ar
io
u
s
class
es
s
u
ch
as
ca
k
e,
f
lo
wer
,
an
d
bus
.
T
h
e
d
ataset
is
d
iv
id
ed
in
to
tr
ain
in
g
an
d
test
in
g
r
atio
o
f
7
0
:3
0
.
T
ab
le
1
r
ep
r
esen
ts
a
d
at
aset d
escr
ip
tio
n
.
T
ab
le
1
.
Data
s
et
d
escr
ip
tio
n
D
a
t
a
s
e
t
N
u
mb
e
r
o
f
c
l
a
sse
s
N
u
mb
e
r
o
f
i
m
a
g
e
s
I
mag
e
si
z
e
C
a
l
t
e
c
h
2
5
6
2
5
6
3
0
,
6
0
7
3
0
0
×
2
0
0
p
i
x
e
l
s
C
o
r
e
l
1
K
1
,
0
0
0
1
0
,
0
0
0
1
2
8
×
1
2
8
p
i
x
e
l
s
2
.
2
.
P
re
pro
ce
s
s
ing
T
h
e
C
altec
h
2
5
6
an
d
C
o
r
el
1
K
d
atasets
ar
e
p
r
ep
r
o
ce
s
s
ed
b
y
u
s
in
g
im
ag
e
r
esizin
g
an
d
n
o
r
m
aliza
tio
n
.
E
ac
h
d
ataset
h
as
a
d
if
f
er
en
t
s
ize
o
f
im
ag
es,
s
o
it
is
r
esized
an
d
n
o
r
m
alize
d
in
to
t
h
e
s
ca
le
o
f
0
an
d
1
.
T
h
e
ex
p
lan
atio
n
o
f
im
ag
e
r
esizin
g
an
d
n
o
r
m
aliza
tio
n
is
g
i
v
e
n
as
f
o
ll
ow
s
,
2
.
2
.
1
.
I
m
a
g
e
re
s
izing
T
h
e
DL
alg
o
r
ith
m
q
u
ick
ly
tr
ain
s
with
s
m
all
im
ag
e
s
izes.
As
th
e
r
aw
in
p
u
t
im
ag
es
v
ar
y
in
s
ize
v
ar
io
u
s
DL
alg
o
r
ith
m
s
r
eq
u
ir
e
th
e
im
ag
e
to
r
esize
in
th
e
s
am
e
d
im
en
s
io
n
.
I
t
is
r
eq
u
ir
ed
to
s
tan
d
ar
d
ize
th
e
im
ag
e
s
ize
b
e
f
o
r
e
tr
ain
in
g
.
T
h
er
ef
o
r
e
,
r
aw
im
ag
es
a
r
e
r
esi
z
ed
in
to
2
5
6
×2
5
6
d
im
en
s
io
n
s
to
e
n
s
u
r
e
d
ataset
s
im
ilar
ity
.
T
h
e
p
ix
el
v
alu
es
ar
e
ad
ju
s
ted
to
s
m
all
s
ize
an
d
in
ap
p
r
o
p
r
iate
r
eg
io
n
s
ar
e
r
em
o
v
ed
d
u
r
i
n
g
th
e
r
esizin
g
p
r
o
ce
s
s
.
2
.
2
.
2
.
No
rma
liza
t
io
n
T
h
e
n
o
r
m
aliza
tio
n
is
u
s
ed
to
u
n
if
o
r
m
th
e
v
alu
es
f
r
o
m
d
if
f
e
r
en
t
q
u
er
y
r
esu
lts
wh
ich
g
e
n
e
r
ate
v
alu
es
in
0
to
1
r
a
n
g
es
[
2
4
]
.
T
h
e
n
o
r
m
alize
d
s
co
r
es
o
n
ev
er
y
f
ea
t
u
r
e
ar
e
m
u
ltip
lied
th
r
o
u
g
h
w
eig
h
ts
to
ac
h
iev
e
a
to
tal
s
co
r
e.
I
f
is
an
im
ag
e
v
al
u
e,
is
a
m
ax
im
u
m
s
co
r
e
an
d
t
h
e
n
o
r
m
alize
d
s
co
r
e
′
is
esti
m
a
ted
b
y
(
1
)
.
′
=
1
−
(
1
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2252
-
8
8
1
4
C
o
n
ten
t b
a
s
ed
ima
g
e
r
etri
ev
a
l
u
s
in
g
visu
a
l g
eo
metric g
r
o
u
p
1
9
…
(
R
a
ja
th
A
r
a
ke
r
e
N
a
r
a
ya
n
a
s
w
a
my
)
433
2
.
3
.
F
e
a
t
ure
ex
t
r
a
ct
io
n
T
h
e
VGG1
9
h
as
th
r
ee
f
u
lly
c
o
n
n
ec
ted
(
FC
)
an
d
s
ix
teen
2
D
co
n
v
o
lu
tio
n
(
co
n
v
)
lay
e
r
s
[
2
5
]
.
I
n
th
e
tr
ain
in
g
p
h
ase,
th
e
co
n
v
lay
er
is
u
s
ed
f
o
r
f
ea
t
u
r
e
ex
tr
ac
tio
n
,
an
d
th
e
m
ax
-
p
o
o
lin
g
lay
er
wi
th
f
ew
co
n
v
lay
er
s
is
u
s
e
d
to
r
ed
u
ce
th
e
f
ea
tu
r
e
d
im
en
s
io
n
ality
.
T
h
e
VGG1
9
is
s
elec
ted
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
b
ec
au
s
e
o
f
its
ef
f
ec
tiv
en
ess
in
c
ap
tu
r
in
g
f
ea
t
u
r
es
with
m
u
ch
in
f
o
r
m
atio
n
.
I
t
is
ef
f
icien
t
in
p
r
o
d
u
cin
g
f
i
n
e
an
d
lay
er
e
d
v
is
io
n
ch
ar
ac
ter
is
tics
wh
ich
ar
e
u
s
ef
u
l
in
m
atch
i
n
g
im
a
g
es
at
C
B
I
R
.
I
n
itially
,
th
e
2
D
co
n
v
lay
e
r
is
ap
p
lied
d
is
tin
ctly
to
ev
er
y
in
p
u
t
im
ag
e
b
y
th
e
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U)
ac
tiv
atio
n
f
u
n
ctio
n
f
o
r
s
p
atial
f
e
atu
r
e
ex
tr
ac
ti
o
n
.
I
t
h
as
6
4
f
ilter
s
th
r
o
u
g
h
t
h
e
R
eL
U
f
u
n
ctio
n
.
T
o
cr
ea
te
c
o
n
v
r
esu
lts
with
less
co
m
p
l
ex
ity
,
a
m
ax
-
p
o
o
lin
g
lay
e
r
is
ap
p
lied
with
a
2
×2
m
atr
ix
w
h
ich
p
er
f
o
r
m
s
a
d
o
wn
-
s
am
p
li
n
g
p
r
o
ce
d
u
r
e.
T
h
e
n
,
th
r
ee
d
u
al
co
n
v
lay
er
s
h
a
v
e
1
2
8
f
ilter
s
with
th
e
m
atr
i
x
o
f
3
×3
wh
ich
u
tili
ze
d
th
e
R
eL
U
f
u
n
ctio
n
.
T
h
ese
in
cl
u
d
ed
lay
e
r
s
allo
w
VGG1
9
to
d
if
f
er
en
tiate
h
ig
h
er
-
lev
el
f
ea
t
u
r
es
th
at
ha
ve
b
ee
n
lo
s
t
in
p
as
t
co
n
v
lay
er
s
.
T
h
e
n
,
m
ax
-
p
o
o
l
in
g
b
y
2
×
2
p
o
o
lin
g
is
f
o
llo
wed
b
y
f
o
u
r
2
D
co
n
v
lay
er
s
wh
ich
h
av
e
th
e
f
o
r
m
atio
n
o
f
2
5
6
f
ilter
s
with
a
m
atr
ix
o
f
3
×3
th
at
is
f
o
llo
wed
b
y
m
ax
-
p
o
o
lin
g
an
d
th
e
2
D
co
n
v
lay
er
h
as
a
f
o
r
m
atio
n
o
f
5
1
2
f
ilter
s
with
m
ax
-
p
o
o
lin
g
.
T
h
en
,
f
o
u
r
m
o
r
e
2
D
co
n
v
lay
er
s
h
av
e
a
f
o
r
m
atio
n
o
f
1
,
0
2
4
f
ilter
s
wh
ich
is
f
o
llo
wed
b
y
m
ax
-
p
o
o
lin
g
.
T
wo
f
u
lly
co
n
n
ec
ted
(
FC
)
lay
er
s
ar
e
o
r
g
an
ized
th
r
o
u
g
h
4
,
0
9
6
n
eu
r
o
n
s
an
d
R
eL
U
f
u
n
ctio
n
wh
ich
is
f
o
llo
wed
b
y
FC
lay
er
with
n
eu
r
o
n
s
o
f
1
,
0
0
0
.
L
astl
y
,
th
e
o
u
tp
u
t
is
m
in
im
ized
to
d
u
al
class
es
th
r
o
u
g
h
th
e
s
o
f
tm
ax
ac
tiv
atio
n
f
u
n
ctio
n
.
T
h
e
v
ar
ia
n
ce
am
o
n
g
tr
u
e
an
d
p
r
ed
icted
s
co
r
es
f
o
r
VGG1
9
is
attain
ed
b
y
u
s
in
g
b
in
ar
y
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
f
u
n
ctio
n
wh
ich
is
esti
m
ated
b
y
(
2
)
, w
h
er
e,
an
d
̂
ar
e
tr
u
e
an
d
p
r
ed
icted
lab
els.
=
−
[
(
̂
)
+
(
1
−
)
(
1
−
̂
)
]
(
2
)
2
.
3
.
1
.
Co
nv
o
lutio
n la
y
er
T
h
e
co
n
v
o
lu
tio
n
la
y
er
is
a
m
ain
elem
en
t
in
VGG1
9
wh
ic
h
u
tili
ze
s
a
g
r
o
u
p
o
f
in
p
u
t
i
m
ag
es
an
d
f
ilter
s
to
d
ev
elo
p
2
D
lay
e
r
s
.
T
h
e
VGG1
9
en
a
b
les
th
e
m
o
d
el
to
r
ec
o
r
d
im
a
g
e
f
u
n
ctio
n
d
u
e
to
th
e
weig
h
t
d
is
tr
ib
u
tio
n
with
less
co
s
t.
I
n
th
e
C
B
I
R
,
it
h
as
s
ix
teen
co
n
v
l
ay
e
r
s
an
d
a
3
×3
f
ilter
.
E
v
er
y
c
o
n
v
lay
er
is
ap
p
lied
as
a
s
et
o
f
k
er
n
els
to
th
e
in
p
u
t
wh
ic
h
ca
p
tu
r
es
s
p
atia
l
h
ier
ar
ch
ies
s
u
ch
as
tex
tu
r
es,
ed
g
es
,
an
d
in
t
r
icate
s
tr
u
ctu
r
es.
T
h
e
VGG1
9
s
tac
k
s
n
u
m
er
o
u
s
co
n
v
lay
er
s
b
ef
o
r
e
m
a
x
-
p
o
o
lin
g
w
h
ich
en
ab
les
d
ee
p
f
ea
tu
r
e
ex
tr
ac
t
io
n
am
o
n
g
v
ar
io
u
s
lev
e
ls
.
2
.
3
.
2
.
M
a
x
po
llin
g
la
y
er
T
h
e
m
ax
p
o
o
lin
g
is
u
s
ed
f
o
r
i
m
ag
e
f
ea
tu
r
e
ex
tr
ac
tio
n
wh
ich
is
th
e
m
o
s
t
u
s
u
al
f
o
r
m
o
f
p
o
o
lin
g
lay
er
.
I
t
u
s
ed
a
2
×2
f
ilter
f
o
r
s
elec
tin
g
th
e
co
n
v
lay
er
o
f
th
e
ac
tiv
atio
n
m
ap
.
T
h
e
p
o
o
lin
g
lay
e
r
m
in
im
izes
th
e
s
ize
an
d
ex
ec
u
tio
n
p
ar
am
eter
s
i
n
th
e
n
etwo
r
k
th
at
h
an
d
le
o
v
er
f
itti
n
g
an
d
en
h
a
n
ce
s
th
e
n
et
wo
r
k
p
r
o
ce
s
s
.
T
h
e
VGG1
9
r
ed
u
ce
s
th
e
c
o
m
p
u
ta
tio
n
al
lo
ad
a
n
d
o
v
er
co
m
es
th
e
o
v
er
f
itti
n
g
is
s
u
es
by
r
ed
u
cin
g
th
e
n
u
m
b
e
r
o
f
p
ar
am
eter
s
.
T
h
e
m
ax
p
o
o
lin
g
is
ap
p
lied
af
ter
e
ac
h
two
o
r
th
r
ee
co
n
v
lay
e
r
s
wh
ich
co
n
t
r
ib
u
te
to
ca
p
tu
r
in
g
h
ier
ar
ch
ical
p
atter
n
s
in
d
ata.
2
.
3
.
3
.
Rec
t
if
ied linea
r
un
it
T
h
e
R
eL
U
is
a
ty
p
e
o
f
ac
ti
v
atio
n
f
u
n
ctio
n
th
at
ca
lcu
late
s
th
e
in
p
u
t
v
alu
es
a
n
d
it
r
eq
u
ir
es
less
co
m
p
u
ter
r
eso
u
r
ce
s
co
m
p
ar
e
d
to
o
th
er
ac
t
iv
atio
n
f
u
n
ctio
n
s
.
T
h
e
n
eu
r
al
n
etwo
r
k
s
ar
e
tr
ain
ed
th
r
o
u
g
h
VGG1
9
wh
ich
h
as
b
ac
k
p
r
o
p
a
g
atio
n
s
tag
es
an
d
m
o
d
if
ied
th
e
weig
h
t
s
to
r
e
d
u
ce
lo
s
s
es
f
o
r
ev
er
y
e
p
o
ch
.
I
f
th
e
n
u
m
b
er
o
f
lay
er
s
is
en
h
an
ce
d
,
th
e
g
r
a
d
ien
ts
ar
e
d
ec
r
ea
s
ed
lin
ea
r
ly
wh
ich
r
ed
u
ce
s
th
e
g
r
ad
ien
t
s
c
o
r
e.
I
f
th
e
g
r
a
d
ien
t
is
ze
r
o
b
ec
a
u
s
e
o
f
ac
tiv
atio
n
it
p
u
s
h
es
th
e
s
co
r
e
to
0
wh
ic
h
is
k
n
o
wn
as
a
v
an
is
h
in
g
g
r
a
d
ien
t
i
s
s
u
e.
T
h
e
R
eL
U
is
g
iv
en
in
(
3
)
.
(
)
=
(
0
,
)
=
{
0
,
≤
0
,
>
0
(
3
)
T
h
e
R
eL
U
is
u
s
ed
af
ter
e
v
er
y
co
n
v
a
n
d
FC
lay
er
wh
ic
h
p
r
esen
ts
n
o
n
-
lin
ea
r
ity
th
r
o
u
g
h
co
n
v
er
tin
g
all
n
eg
ativ
e
s
co
r
es
in
th
e
f
ea
tu
r
e
m
ap
to
0
.
I
t
ass
is
t
s
in
tr
ain
in
g
d
ee
p
n
etwo
r
k
s
b
y
r
ed
u
cin
g
v
an
is
h
in
g
g
r
a
d
ien
t
is
s
u
es.
T
h
is
ass
i
s
ts
in
b
o
o
s
tin
g
th
e
tr
ain
in
g
p
r
o
ce
s
s
d
u
e
to
ef
f
icien
cy
wh
ich
r
eq
u
ir
e
s
s
im
p
le
th
r
esh
o
l
d
o
p
er
atio
n
.
I
n
VGG1
9
,
R
eL
U
en
h
an
ce
s
th
e
ca
p
ab
ilit
y
to
lear
n
in
tr
icate
p
atter
n
s
wh
ich
e
n
ab
les
it
to
p
r
o
v
id
e
b
etter
p
er
f
o
r
m
an
ce
in
C
B
I
R
.
2
.
3
.
4
.
F
ull
y
c
o
nn
ec
t
ed
la
y
er
I
n
VGG1
9
,
th
e
FC
lay
er
is
u
s
e
d
to
t
r
an
s
f
o
r
m
h
ig
h
er
-
lev
el
f
ea
tu
r
es
ex
tr
ac
ted
b
y
c
o
n
v
lay
er
s
in
to
f
in
al
class
if
ier
o
u
tp
u
ts
.
Af
ter
co
n
v
,
p
o
o
lin
g
la
y
er
s
,
th
e
o
u
t
p
u
t
is
f
latten
ed
an
d
f
e
d
in
to
th
e
FC
lay
er
.
T
h
e
VGG1
9
h
as
th
r
ee
FC
lay
er
s
s
u
ch
as
t
h
e
f
ir
s
t
two
h
a
v
e
4
0
9
6
u
n
its
,
an
d
th
e
last
o
n
e
h
as
1
,
0
0
0
u
n
its
.
T
h
ese
lay
er
s
a
r
e
f
o
llo
wed
b
y
R
eL
U
an
d
th
e
f
in
al
FC
lay
er
g
en
er
ates
cla
s
s
p
r
o
b
ab
ilit
ies.
Fig
u
r
e
2
d
is
p
lay
s
th
e
VGG1
9
ar
ch
itectu
r
e.
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.
14
,
No
.
2
,
J
u
n
e
2
0
2
5
:
4
3
0
-
438
434
Fig
u
r
e
2
.
Ar
c
h
itectu
r
e
o
f
VGG1
9
2
.
4
.
Sim
ila
ri
t
y
m
ea
s
ure
T
h
e
s
im
ilar
ity
is
d
eter
m
in
ed
b
y
co
m
p
ar
in
g
im
ag
es
an
d
th
e
s
im
ilar
ity
am
o
n
g
im
ag
e
im
ag
es
is
d
etec
ted
wh
en
its
n
eig
h
b
o
r
s
ar
e
s
im
ilar
.
T
h
e
J
ac
ca
r
d
is
u
s
ed
to
co
m
p
ar
e
th
e
s
im
ilar
ity
an
d
im
ag
e
elem
e
n
t
d
iv
er
g
en
ce
s
in
v
ar
io
u
s
s
am
p
l
es
d
u
e
to
its
ef
f
icien
cy
.
I
t
is
h
elp
f
u
l
in
C
B
I
R
b
ec
au
s
e
it
r
ef
lects
th
e
n
u
m
b
er
o
f
in
ter
s
ec
tio
n
s
ets
wh
en
co
m
p
a
r
in
g
d
is
s
im
ilar
an
d
s
im
ilar
im
ag
es.
I
t
en
a
b
les
b
etter
d
is
cr
im
in
ativ
e
im
ag
es
th
at
h
av
e
s
em
an
tic
co
n
ten
t
at
a
h
ig
h
er
lev
el,
an
d
it
is
b
en
ef
icial
in
f
o
cu
s
in
g
o
n
th
e
s
em
an
tic
g
ap
.
T
h
e
s
im
ilar
ity
is
ca
lcu
lated
b
y
u
s
in
g
th
e
J
a
cc
ar
d
s
im
ilar
ity
m
ea
s
u
r
e
as (
4
)
.
(
,
)
=
|
⋂
|
|
⋃
|
=
|
⋂
|
|
|
+
|
|
−
|
⋂
|
(
4
)
W
h
er
e
is
J
ac
ca
r
d
s
im
ilar
i
ty
,
|
|
an
d
|
|
ar
e
n
u
m
b
er
o
f
elem
en
ts
in
an
d
,
th
e
|
⋂
|
is
th
e
s
ize
o
f
t
h
e
in
ter
s
ec
tio
n
o
f
s
et
an
d
,
|
⋃
|
is
th
e
s
ize
o
f
th
e
u
n
io
n
o
f
s
et
an
d
.
I
n
itially
,
th
e
s
im
ilar
ity
s
co
r
e
am
o
n
g
im
ag
e
g
r
ap
h
s
is
in
itialized
to
NUL
L
wh
ich
d
en
o
tes
n
o
im
a
g
es.
T
h
en
,
th
e
m
o
d
el
ca
lcu
lat
es
a
s
im
ilar
ity
s
co
r
e
f
o
r
ev
e
r
y
p
air
am
o
n
g
im
ag
es
an
d
an
d
in
clu
d
es
th
e
r
esu
lt
in
th
e
s
im
ilar
ity
s
co
r
e.
L
astl
y
,
it
r
etu
r
n
s
th
e
f
i
n
al
s
im
ilar
ity
s
co
r
e.
T
h
er
ef
o
r
e
,
th
e
s
im
ilar
ity
b
etwe
en
th
e
ac
tu
al
im
ag
e
an
d
q
u
er
y
im
a
g
e
r
esu
lt
s
in
b
etter
r
etr
iev
al
im
ag
es.
T
h
e
s
u
itab
le
im
ag
es
r
elate
d
to
q
u
er
y
im
ag
es
ar
e
ac
co
m
p
lis
h
ed
th
r
o
u
g
h
f
ea
t
u
r
e
ex
tr
ac
tio
n
a
n
d
s
im
ilar
ity
ca
lcu
latio
n
.
Fig
u
r
e
3
s
h
o
ws
an
ex
am
p
le
o
f
im
ag
e
r
etr
iev
al
f
r
o
m
th
e
C
altec
h
2
5
6
d
ataset
o
f
th
e
q
u
er
y
im
a
g
e
(
Fig
u
r
e
3
(
a
)
)
a
n
d
th
e
r
et
r
iev
ed
im
a
g
e
(
Fig
u
r
e
3
(
b
)
)
.
Fig
u
r
e
4
s
h
o
ws
an
ex
am
p
le
o
f
im
ag
e
r
etr
iev
al
f
r
o
m
th
e
C
o
r
el
1
K
d
a
taset o
f
th
e
q
u
er
y
im
ag
e
(
Fig
u
r
e
4
(
a)
)
an
d
th
e
r
etr
iev
e
d
im
ag
e
(
Fig
u
r
e
4
(
b
)
)
.
(
a)
(
b
)
Fig
u
r
e
3
.
Sam
p
le
o
f
im
a
g
e
r
etr
iev
al
f
r
o
m
C
altec
h
2
5
6
d
ataset
(
a)
q
u
er
y
im
a
g
e
an
d
(
b
)
r
etr
ie
v
ed
im
ag
es
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2252
-
8
8
1
4
C
o
n
ten
t b
a
s
ed
ima
g
e
r
etri
ev
a
l
u
s
in
g
visu
a
l g
eo
metric g
r
o
u
p
1
9
…
(
R
a
ja
th
A
r
a
ke
r
e
N
a
r
a
ya
n
a
s
w
a
my
)
435
(
a)
(
b
)
Fig
u
r
e
4
.
Sam
p
le
o
f
im
a
g
e
r
etr
iev
al
f
r
o
m
C
o
r
el
1
K
d
ataset
(
a
)
q
u
er
y
im
ag
e
a
n
d
(
b
)
r
et
r
iev
e
d
im
ag
es
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
e
p
r
o
p
o
s
ed
VGG1
9
-
J
ac
ca
r
d
is
im
p
lem
en
ted
in
th
e
en
v
ir
o
n
m
en
t
o
f
Py
t
h
o
n
3
.
1
0
wit
h
s
y
s
tem
co
n
f
ig
u
r
atio
n
o
f
W
in
d
o
ws
1
0
OS,
I
n
tel
C
o
r
e
i5
,
an
d
8
GB
R
AM
.
T
h
e
m
etr
ics
s
u
ch
as
r
ec
all,
f
-
m
ea
s
u
r
e
,
a
n
d
av
er
ag
e
p
r
ec
is
io
n
a
r
e
u
s
ed
to
ca
lcu
late
th
e
VGG1
9
-
J
ac
ca
r
d
p
er
f
o
r
m
an
ce
.
T
h
ese
m
etr
ics ar
e
f
o
r
m
u
lated
in
(
5
)
-
(
7
)
.
W
h
er
e
an
d
ar
e
tr
u
e
p
o
s
itiv
e
an
d
f
alse
n
e
g
ativ
e
,
an
d
ar
e
r
ec
all
a
n
d
p
r
ec
is
io
n
at
th
th
r
esh
o
ld
,
−
1
is
a
r
ec
all
at
−
1
th
th
r
e
s
h
o
ld
.
=
+
×
100
(
5
)
−
=
2
×
×
+
×
100
(
6
)
=
∑
(
−
−
1
)
×
100
(
7
)
T
ab
le
2
d
is
p
lay
s
th
e
d
if
f
er
e
n
t
f
ea
tu
r
e
e
x
tr
ac
tio
n
r
esu
lts
with
m
etr
ics
o
f
r
ec
all,
f
-
m
e
asu
r
e
,
an
d
av
er
ag
e
p
r
ec
is
io
n
f
o
r
C
altec
h
2
5
6
a
n
d
C
o
r
el
1
K
d
atasets
.
T
h
e
I
n
ce
p
tio
n
V1
,
M
o
b
ileNetV3
,
an
d
R
esNet1
5
1
ar
e
tak
en
as
ex
is
tin
g
tech
n
iq
u
es
to
co
m
p
ar
e
th
e
r
esu
lt
v
a
lu
es
o
f
VGG1
9
.
T
h
e
VGG1
9
ac
h
iev
es
9
9
.
5
%,
9
9
.
3
%
,
an
d
9
9
.
0
%
o
f
r
ec
all,
f
-
m
ea
s
u
r
e
,
a
n
d
a
v
er
ag
e
p
r
e
cisi
o
n
f
o
r
th
e
C
altec
h
2
5
6
d
ataset.
T
h
e
VGG1
9
ac
h
iev
es 9
8
.
8
,
9
8
.
8
,
a
n
d
9
9
.
8
% o
f
r
ec
all,
f
-
m
ea
s
u
r
e
,
an
d
av
er
ag
e
p
r
ec
is
io
n
f
o
r
t
h
e
C
o
r
el
1
K
d
ataset.
T
ab
le
3
d
is
p
lay
s
th
e
d
if
f
er
e
n
t
lo
s
s
f
u
n
ctio
n
r
esu
lts
with
m
etr
ics
o
f
r
ec
all,
f
-
m
ea
s
u
r
e
,
an
d
av
er
a
g
e
p
r
ec
is
io
n
f
o
r
C
altec
h
2
5
6
an
d
C
o
r
el
1
K
d
atasets
.
T
h
e
m
ea
n
s
q
u
ar
e
er
r
o
r
(
MSE
)
,
m
ea
n
a
b
s
o
lu
te
er
r
o
r
(
MA
E
)
,
an
d
c
r
o
s
s
-
en
tr
o
p
y
a
r
e
ta
k
en
as
ex
is
tin
g
tech
n
iq
u
es
to
co
m
p
ar
e
t
h
e
r
esu
lt
v
alu
es
o
f
J
ac
ca
r
d
.
T
h
e
J
ac
ca
r
d
ac
h
iev
es
9
9
.
5
,
9
9
.
3
,
an
d
9
9
.
0
%
o
f
r
ec
all,
f
-
m
ea
s
u
r
e
,
a
n
d
a
v
er
ag
e
p
r
ec
is
io
n
f
o
r
t
h
e
C
altec
h
2
5
6
d
ataset.
T
h
e
J
ac
ca
r
d
ac
h
iev
es 9
8
.
8
,
9
8
.
8
,
a
n
d
9
9
.
8
% o
f
r
ec
all,
f
-
m
ea
s
u
r
e
,
an
d
av
e
r
ag
e
p
r
ec
is
io
n
f
o
r
th
e
C
o
r
el
1
K
d
ataset.
T
ab
le
4
d
is
p
lay
s
th
e
av
er
a
g
e
p
r
ec
is
io
n
b
y
d
if
f
er
en
t
d
is
tan
ce
m
ea
s
u
r
e
r
esu
lts
in
ter
m
s
o
f
with
an
d
with
o
u
t
VGG1
9
f
o
r
C
altec
h
2
5
6
an
d
C
o
r
el
1
K
d
atasets
.
C
o
s
in
e
an
d
ch
i
-
s
q
u
ar
e
ar
e
tak
en
as
ex
is
tin
g
tech
n
iq
u
es
to
co
m
p
ar
e
t
h
e
r
es
u
lt
v
alu
es
o
f
J
ac
ca
r
d
.
T
h
e
J
ac
ca
r
d
ac
h
ie
v
es
3
2
.
1
7
an
d
4
9
.
2
8
in
te
r
m
s
o
f
with
an
d
with
o
u
t
VGG1
9
f
o
r
th
e
C
altec
h
2
5
6
d
a
taset.
T
h
e
J
ac
ca
r
d
ac
h
iev
es
3
1
.
5
7
an
d
5
0
.
7
0
in
ter
m
s
o
f
with
an
d
with
o
u
t V
GG1
9
f
o
r
th
e
C
o
r
el
1
K
d
ataset.
T
ab
le
5
d
is
p
lay
s
th
e
p
r
o
p
o
s
e
d
m
eth
o
d
o
lo
g
y
r
esu
lts
with
m
etr
ics
o
f
r
ec
all,
f
-
m
ea
s
u
r
e
,
an
d
av
er
a
g
e
p
r
ec
is
io
n
f
o
r
C
altec
h
2
5
6
an
d
C
o
r
el
1
K
d
atasets
.
T
h
e
E
f
f
icien
tNet,
Den
s
eNe
t1
2
1
,
a
n
d
C
NN
ar
e
tak
en
as
ex
is
tin
g
tech
n
iq
u
es
to
co
m
p
a
r
e
th
e
r
esu
lt
v
alu
es
o
f
VGG1
9
-
J
ac
ca
r
d
.
T
h
e
VGG1
9
-
J
ac
ca
r
d
ac
h
iev
es
9
9
.
5
,
9
9
.
3
,
a
n
d
9
9
.
0
%
o
f
r
ec
all,
f
-
m
ea
s
u
r
e
,
an
d
a
v
er
ag
e
p
r
ec
is
io
n
f
o
r
th
e
C
altec
h
2
5
6
d
ataset.
T
h
e
VGG1
9
-
J
ac
ca
r
d
ac
h
iev
es
9
8
.
8
,
9
8
.
8
,
a
n
d
9
9
.
8
% o
f
r
ec
all,
f
-
m
ea
s
u
r
e
,
an
d
av
er
ag
e
p
r
ec
is
io
n
f
o
r
t
h
e
C
o
r
el
1
K
d
ataset.
T
ab
le
2
.
R
esu
lts
f
o
r
d
if
f
er
en
t f
ea
tu
r
e
ex
tr
ac
tio
n
D
a
t
a
s
e
t
M
e
t
h
o
d
R
e
c
a
l
l
(
%)
F
-
mea
su
r
e
(
%)
A
v
e
r
a
g
e
p
r
e
c
i
s
i
o
n
(
%)
C
a
l
t
e
c
h
2
5
6
I
n
c
e
p
t
i
o
n
V
1
9
6
.
5
9
6
.
7
9
6
.
8
M
o
b
i
l
e
N
e
t
V
3
9
7
.
2
9
7
.
5
9
7
.
7
R
e
s
n
e
t
1
5
1
9
8
.
3
9
8
.
4
9
8
.
6
V
G
G
1
9
9
9
.
5
9
9
.
3
9
9
.
0
C
o
r
e
l
1
K
I
n
c
e
p
t
i
o
n
V
1
9
7
.
3
9
7
.
0
9
7
.
6
M
o
b
i
l
e
N
e
t
V
3
9
8
.
0
9
8
.
0
9
8
.
0
R
e
s
n
e
t
1
5
1
9
5
.
8
9
6
.
0
9
6
.
1
V
G
G
1
9
9
8
.
8
9
8
.
8
9
9
.
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.
14
,
No
.
2
,
J
u
n
e
2
0
2
5
:
4
3
0
-
438
436
T
ab
le
3
.
R
esu
lt f
o
r
d
if
f
er
e
n
t lo
s
s
f
u
n
ctio
n
D
a
t
a
s
e
t
M
e
t
h
o
d
R
e
c
a
l
l
(
%)
F
-
mea
su
r
e
(
%)
A
v
e
r
a
g
e
p
r
e
c
i
s
i
o
n
(
%)
C
a
l
t
e
c
h
2
5
6
M
S
E
9
7
.
0
9
6
.
8
9
6
.
5
M
A
E
9
6
.
5
9
6
.
6
9
6
.
8
C
r
o
ss
e
n
t
r
o
p
y
9
6
.
8
9
6
.
7
9
6
.
6
Jac
c
a
r
d
9
9
.
5
9
9
.
3
9
9
.
0
C
o
r
e
l
1
K
M
S
E
9
7
.
5
9
7
.
3
9
7
.
1
M
A
E
9
6
.
8
9
7
.
5
9
6
.
9
C
r
o
ss
e
n
t
r
o
p
y
9
6
.
9
9
7
.
1
9
7
.
0
Jac
c
a
r
d
9
8
.
8
9
8
.
8
9
9
.
8
T
ab
le
4
.
R
esu
lt f
o
r
a
v
er
ag
e
p
r
ec
is
io
n
with
d
if
f
er
en
t
d
is
tan
ce
m
ea
s
u
r
e
D
a
t
a
s
e
t
M
e
t
h
o
d
W
i
t
h
V
G
G
1
9
W
i
t
h
o
u
t
V
G
G
1
9
C
a
l
t
e
c
h
2
5
6
C
o
s
i
n
e
3
5
.
5
6
5
6
.
3
9
C
h
i
-
sq
u
a
r
e
3
5
.
2
5
5
7
.
6
3
Jac
c
a
r
d
3
2
.
1
7
4
9
.
2
8
C
o
r
e
l
1
K
C
o
s
i
n
e
3
6
.
9
2
6
9
.
7
5
C
h
i
-
sq
u
a
r
e
4
0
.
1
7
6
8
.
0
8
Jac
c
a
r
d
3
1
.
5
7
5
0
.
7
0
T
ab
le
5
.
R
esu
lt
of
th
e
p
r
o
p
o
s
e
d
m
eth
o
d
o
lo
g
y
D
a
t
a
s
e
t
M
e
t
h
o
d
R
e
c
a
l
l
(
%)
F
-
mea
su
r
e
(
%)
A
v
e
r
a
g
e
p
r
e
c
i
s
i
o
n
(
%)
C
a
l
t
e
c
h
2
5
6
Ef
f
i
c
i
e
n
t
N
e
t
9
8
.
6
9
8
.
6
9
8
.
8
D
e
n
seN
e
t
1
2
1
9
8
.
4
9
8
.
4
9
8
.
6
C
N
N
9
7
.
9
9
7
.
9
9
8
.
1
V
G
G
1
9
-
Ja
c
c
a
r
d
9
9
.
5
9
9
.
3
9
9
.
0
C
o
r
e
l
1
K
Ef
f
i
c
i
e
n
t
N
e
t
9
8
.
2
9
8
.
2
9
8
.
3
D
e
n
seN
e
t
1
2
1
9
7
.
9
9
7
.
9
9
8
.
0
C
N
N
9
7
.
5
9
7
.
5
9
7
.
7
V
G
G
1
9
-
Ja
c
c
a
r
d
9
8
.
8
9
8
.
8
9
9
.
8
3
.
1
.
Co
m
pa
ra
t
iv
e
a
na
ly
s
is
I
n
th
is
co
n
te
x
t,
th
e
c
o
m
p
ar
ati
v
e
an
aly
s
is
with
th
e
p
r
o
p
o
s
ed
VGG1
9
-
J
ac
ca
r
d
is
p
r
esen
ted
i
n
ter
m
s
o
f
v
ar
io
u
s
m
etr
ics s
u
ch
as r
ec
all,
f
-
m
ea
s
u
r
e
,
an
d
av
e
r
ag
e
p
r
ec
is
io
n
.
T
h
e
ex
is
tin
g
m
eth
o
d
s
s
u
ch
as L
B
-
C
B
I
R
[
1
7
]
an
d
T
wo
-
s
tag
e
C
B
I
R
[
1
9
]
ar
e
co
n
s
id
er
ed
wh
ich
is
co
m
p
a
r
ed
with
th
e
p
r
o
p
o
s
ed
VGG1
9
-
J
ac
ca
r
d
f
o
r
b
o
t
h
d
atasets
s
u
c
h
as
C
atec
h
-
2
5
6
a
n
d
C
o
r
el
1
K.
T
h
e
VGG1
9
-
J
ac
ca
r
d
ac
h
iev
es
9
9
.
5
,
9
9
.
3
,
an
d
9
9
.
0
%
o
f
r
ec
all,
f
-
m
ea
s
u
r
e
,
an
d
av
er
a
g
e
p
r
ec
is
i
o
n
f
o
r
th
e
C
altec
h
2
5
6
d
ataset.
T
h
e
VGG1
9
-
J
ac
ca
r
d
ac
h
ie
v
es
9
8
.
8
,
9
8
.
8
,
an
d
9
9
.
8
%
o
f
r
ec
all,
f
-
m
ea
s
u
r
e
,
a
n
d
av
er
a
g
e
p
r
ec
is
io
n
f
o
r
th
e
C
o
r
el
1
K
d
ataset.
T
ab
le
6
d
is
p
la
y
s
th
e
co
m
p
ar
ativ
e
an
aly
s
is
.
T
ab
le
6
.
C
o
m
p
a
r
ativ
e
an
aly
s
is
D
a
t
a
s
e
t
M
e
t
h
o
d
R
e
c
a
l
l
(
%)
F
-
mea
su
r
e
(
%)
A
v
e
r
a
g
e
p
r
e
c
i
s
i
o
n
(
%)
C
a
l
t
e
c
h
2
5
6
LB
-
C
B
I
R
[
1
7
]
NA
NA
3
2
.
1
2
Tw
o
-
st
a
g
e
C
B
I
R
[
1
9
]
0
.
2
4
0
.
3
8
NA
V
G
G
1
9
-
Ja
c
c
a
r
d
9
9
.
5
9
9
.
3
9
9
.
0
C
o
r
e
l
1
K
LB
-
C
B
I
R
[
1
7
]
NA
1
6
.
8
3
9
9
.
2
V
G
G
1
9
-
Ja
c
c
a
r
d
9
8
.
8
9
8
.
8
9
9
.
8
3
.
2
.
Dis
cus
s
io
n
T
h
e
lim
itatio
n
s
o
f
ex
is
tin
g
m
eth
o
d
s
an
d
ad
v
a
n
tag
es
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
ar
e
d
is
cu
s
s
ed
in
th
is
s
ec
tio
n
.
T
h
e
ex
is
tin
g
tec
h
n
iq
u
es
s
u
ch
as
L
B
-
C
B
I
R
[
1
7
]
d
o
n
o
t
co
n
ce
n
t
r
ate
an
d
ex
tr
ac
t
th
e
h
ig
h
-
lev
el
f
ea
tu
r
es
th
at
af
f
ec
t
th
e
m
o
d
el’
s
f
lex
i
b
ilit
y
.
Mu
lti
-
s
ca
le
av
er
ag
in
g
lo
c
al
b
in
ar
y
p
atter
n
s
[
1
8
]
d
o
es
n
o
t
co
n
ce
n
tr
ate
an
d
ex
tr
ac
ts
th
e
h
ig
h
-
le
v
el
f
ea
tu
r
es
th
at
af
f
ec
t
th
e
m
o
d
el
’
s
f
le
x
ib
ilit
y
.
T
wo
-
s
tag
e
C
B
I
R
[
1
9
]
f
ails
to
d
etec
t
t
h
e
ir
r
eg
u
lar
s
h
a
p
e
im
ag
es
a
n
d
r
e
d
u
ce
s
th
e
in
ter
p
r
etab
ilit
y
wh
i
ch
af
f
ec
ts
th
e
p
er
f
o
r
m
an
ce
o
f
C
B
I
R
.
T
h
e
SIFT
-
KAZ
E
[
2
0
]
r
eq
u
ir
ed
a
h
ig
h
n
u
m
b
er
o
f
o
b
jects
to
s
ea
r
ch
.
I
n
th
e
tr
ip
let
lo
s
s
f
u
n
ctio
n
[
2
1
]
,
g
en
er
aliza
tio
n
p
er
f
o
r
m
an
ce
was
r
ed
u
ce
d
,
an
d
b
r
id
g
in
g
th
e
s
em
a
n
tic
g
ap
p
r
o
b
lem
o
cc
u
r
r
e
d
am
o
n
g
l
ow
-
l
ev
el
an
d
h
i
g
h
-
lev
e
l
s
em
an
tic
f
ea
tu
r
es.
T
o
o
v
e
r
co
m
e
th
ese
lim
itatio
n
s
,
th
is
r
es
ea
r
ch
p
r
o
p
o
s
es
a
VGG1
9
-
J
ac
ca
r
d
f
o
r
C
B
I
R
.
T
h
e
VGG1
9
en
ab
les
to
ca
p
tu
r
e
o
f
h
ier
ar
ch
ical
f
ea
tu
r
es
an
d
is
s
u
itab
le
f
o
r
ch
a
r
ac
ter
izin
g
tex
t
u
r
e
an
d
f
i
n
e
d
etails.
T
h
e
p
r
e
-
t
r
ain
ed
weig
h
ts
o
f
V
GG1
9
allo
w
to
g
en
e
r
at
io
n
o
f
d
is
cr
im
in
ati
v
e
an
d
r
o
b
u
s
t
f
ea
t
u
r
e
r
ep
r
esen
tatio
n
s
th
er
eb
y
en
h
an
cin
g
C
B
I
R
p
er
f
o
r
m
an
ce
.
T
h
e
J
ac
ca
r
d
is
u
s
ed
as
a
s
im
ilar
ity
m
ea
s
u
r
e
am
o
n
g
f
ea
tu
r
e
v
ec
t
o
r
s
b
y
co
m
p
ar
in
g
th
e
in
ter
s
ec
tio
n
an
d
u
n
io
n
o
f
f
ea
tu
r
e
s
ets.
I
t h
elp
s
m
an
ag
e
h
ig
h
-
d
im
e
n
s
io
n
al
an
d
s
p
ar
s
e
d
ata
wh
ich
p
r
o
v
id
es q
u
ick
a
n
d
ac
cu
r
ate
im
ag
e
r
etr
iev
al.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2252
-
8
8
1
4
C
o
n
ten
t b
a
s
ed
ima
g
e
r
etri
ev
a
l
u
s
in
g
visu
a
l g
eo
metric g
r
o
u
p
1
9
…
(
R
a
ja
th
A
r
a
ke
r
e
N
a
r
a
ya
n
a
s
w
a
my
)
437
4.
CO
NCLU
SI
O
N
T
h
is
r
esear
ch
p
r
o
p
o
s
es
a
VGG1
9
with
J
ac
ca
r
d
f
o
r
C
B
I
R
.
T
h
e
VGG1
9
is
u
s
ed
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
th
at
ef
f
icien
tly
ca
p
tu
r
es
a
h
ier
ar
ch
ical
f
ea
tu
r
e
with
m
o
r
e
in
f
o
r
m
atio
n
in
th
e
im
ag
e
d
u
e
to
n
u
m
er
o
u
s
co
n
v
o
l
u
tio
n
al
lay
er
s
.
I
t
h
elp
s
cr
ea
te
f
in
e
an
d
lay
er
e
d
v
is
io
n
ch
ar
ac
ter
is
tics
.
T
h
e
J
ac
ca
r
d
i
s
u
s
ed
to
co
m
p
ar
e
th
e
s
im
ilar
ity
an
d
im
ag
e
d
iv
e
r
g
en
ce
o
f
v
ar
io
u
s
s
am
p
les
b
e
ca
u
s
e
o
f
its
ef
f
ec
tiv
en
ess
.
M
o
r
eo
v
e
r
,
it
e
n
ab
les
b
etter
d
is
cr
im
in
at
iv
e
im
ag
es
th
at
h
av
e
s
em
an
tic
co
n
te
n
t
at
a
h
ig
h
er
lev
el.
T
h
e
r
aw
d
ataset
im
ag
es
ar
e
r
esized
in
to
p
a
r
ticu
lar
d
im
en
s
io
n
s
to
en
s
u
r
e
d
ataset
s
im
ilar
ity
.
T
h
e
n
,
th
e
n
o
r
m
aliza
tio
n
is
u
s
ed
t
o
u
n
if
o
r
m
th
e
v
alu
es
f
r
o
m
d
if
f
er
en
t
q
u
er
y
r
esu
lts
wh
ich
g
en
er
ates
v
alu
es
i
n
0
to
1
r
an
g
es.
T
h
e
VGG1
9
-
J
ac
ca
r
d
ac
h
iev
es
an
av
er
ag
e
p
r
ec
is
io
n
o
f
9
9
.
0
a
n
d
9
9
.
8
%
f
o
r
C
altec
h
2
5
6
an
d
C
o
r
el
1
K
d
atasets
.
I
n
th
e
f
u
tu
r
e,
o
th
er
n
eu
r
al
n
etwo
r
k
s
will b
e
d
is
co
v
er
e
d
to
f
u
r
th
er
en
h
a
n
ce
th
e
C
B
I
R
p
er
f
o
r
m
an
ce
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
R
ajath
Ar
ak
er
e
Nar
ay
an
aswam
y
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Vid
y
alak
s
h
m
i
Kr
is
h
n
e
Go
wd
a
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
th
at
s
u
p
p
o
r
t
th
e
f
i
n
d
in
g
s
o
f
th
is
s
tu
d
y
a
r
e
o
p
e
n
ly
av
ailab
le
in
[
Kag
g
le]
at
“Caltech
2
5
6
d
ataset.
”
h
ttp
s
://www.
k
ag
g
le.
co
m
/d
atasets
/jes
s
icali9
5
3
0
/caltec
h
2
5
6
.
a
n
d
“Co
r
el
1
K
d
ataset.
”
h
ttp
s
://www.
k
ag
g
le.
co
m
/d
atasets
/elk
am
el/co
r
el
-
im
ag
es.
RE
F
E
R
E
NC
E
S
[
1
]
P
.
S
h
a
m
n
a
,
V
.
K
.
G
o
v
i
n
d
a
n
,
a
n
d
K
.
A
.
A
b
d
u
l
N
a
z
e
e
r
,
“
C
o
n
t
e
n
t
-
b
a
s
e
d
med
i
c
a
l
i
ma
g
e
r
e
t
r
i
e
v
a
l
b
y
s
p
a
t
i
a
l
m
a
t
c
h
i
n
g
o
f
v
i
s
u
a
l
w
o
r
d
s,
”
J
o
u
r
n
a
l
o
f
K
i
n
g
S
a
u
d
U
n
i
v
e
rsi
t
y
-
C
o
m
p
u
t
e
r
a
n
d
I
n
f
o
rm
a
t
i
o
n
S
c
i
e
n
c
e
s
,
v
o
l
.
3
4
,
n
o
.
2
,
p
p
.
5
8
–
7
1
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
k
s
u
c
i
.
2
0
1
8
.
1
0
.
0
0
2
.
[
2
]
G
.
K
.
R
a
j
u
,
P
.
P
a
d
ma
n
a
b
h
a
m,
a
n
d
A
.
G
o
v
a
r
d
h
a
n
,
“
E
n
h
a
n
c
e
d
c
o
n
t
e
n
t
-
b
a
se
d
i
m
a
g
e
r
e
t
r
i
e
v
a
l
w
i
t
h
t
r
i
o
-
d
e
e
p
f
e
a
t
u
r
e
e
x
t
r
a
c
t
o
r
s
w
i
t
h
mu
l
t
i
-
s
i
mi
l
a
r
i
t
y
f
u
n
c
t
i
o
n
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
I
n
t
e
l
l
i
g
e
n
t
E
n
g
i
n
e
e
r
i
n
g
a
n
d
S
y
s
t
e
m
s
,
v
o
l
.
1
5
,
n
o
.
6
,
p
p
.
5
1
1
–
5
2
5
,
2
0
2
2
,
d
o
i
:
1
0
.
2
2
2
6
6
/
i
j
i
e
s
2
0
2
2
.
1
2
3
1
.
4
6
.
[
3
]
M
.
M
.
M
o
n
o
w
a
r
,
M
.
A
.
H
a
m
i
d
,
A
.
Q
.
O
h
i
,
M
.
O
.
A
l
a
ssafi
,
a
n
d
M
.
F
.
M
r
i
d
h
a
,
“
A
u
t
o
R
e
t
:
a
s
e
l
f
-
su
p
e
r
v
i
se
d
sp
a
t
i
a
l
r
e
c
u
r
r
e
n
t
n
e
t
w
o
r
k
f
o
r
c
o
n
t
e
n
t
-
b
a
s
e
d
i
m
a
g
e
r
e
t
r
i
e
v
a
l
,
”
S
e
n
so
rs
,
v
o
l
.
2
2
,
n
o
.
6
,
2
0
2
2
,
d
o
i
:
1
0
.
3
3
9
0
/
s2
2
0
6
2
1
8
8
.
[
4
]
M
.
R
a
s
h
a
d
,
I
.
A
f
i
f
i
,
a
n
d
M
.
A
b
d
e
l
f
a
t
a
h
,
“
R
b
Q
E
:
a
n
e
f
f
i
c
i
e
n
t
m
e
t
h
o
d
f
o
r
c
o
n
t
e
n
t
-
b
a
se
d
me
d
i
c
a
l
i
ma
g
e
r
e
t
r
i
e
v
a
l
b
a
se
d
o
n
q
u
e
r
y
e
x
p
a
n
si
o
n
,
”
J
o
u
rn
a
l
o
f
D
i
g
i
t
a
l
I
m
a
g
i
n
g
,
v
o
l
.
3
6
,
n
o
.
3
,
p
p
.
1
2
4
8
–
1
2
6
1
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
0
2
7
8
-
0
2
2
-
0
0
7
6
9
-
7.
[
5
]
A
.
R
a
h
m
a
n
,
E.
W
i
n
a
r
k
o
,
a
n
d
K
.
M
u
s
t
o
f
a
,
“
C
o
n
t
e
n
t
-
b
a
s
e
d
p
r
o
d
u
c
t
i
m
a
g
e
r
e
t
r
i
e
v
a
l
u
s
i
n
g
sq
u
a
r
e
d
-
h
i
n
g
e
l
o
ss
t
r
a
i
n
e
d
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
E
l
e
c
t
ri
c
a
l
a
n
d
C
o
m
p
u
t
e
r
En
g
i
n
e
e
ri
n
g
,
v
o
l
.
1
3
,
n
o
.
5
,
p
p
.
5
8
0
4
–
5
8
1
2
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
e
c
e
.
v
1
3
i
5
.
p
p
5
8
0
4
-
5
8
1
2
.
[
6
]
F
.
A
.
A
l
g
h
a
md
i
,
“
A
n
e
f
f
e
c
t
i
v
e
h
y
b
r
i
d
f
r
a
mew
o
r
k
b
a
s
e
d
o
n
c
o
m
b
i
n
a
t
i
o
n
o
f
c
o
l
o
r
a
n
d
t
e
x
t
u
r
e
f
e
a
t
u
r
e
s
f
o
r
c
o
n
t
e
n
t
-
b
a
se
d
i
ma
g
e
r
e
t
r
i
e
v
a
l
,
”
Ar
a
b
i
a
n
J
o
u
rn
a
l
f
o
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
ri
n
g
,
v
o
l
.
4
9
,
n
o
.
3
,
p
p
.
3
5
7
5
–
3
5
9
1
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
3
3
6
9
-
0
2
3
-
0
8
0
8
7
-
y.
[
7
]
S
.
F
a
d
a
e
i
,
“
N
e
w
d
o
mi
n
a
n
t
c
o
l
o
r
d
e
s
c
r
i
p
t
o
r
f
e
a
t
u
r
e
s
b
a
s
e
d
o
n
w
e
i
g
h
t
i
n
g
o
f
m
o
r
e
i
n
f
o
r
ma
t
i
v
e
p
i
x
e
l
s
u
s
i
n
g
s
u
i
t
a
b
l
e
mas
k
s
f
o
r
c
o
n
t
e
n
t
-
b
a
se
d
i
m
a
g
e
r
e
t
r
i
e
v
a
l
,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
E
n
g
i
n
e
e
ri
n
g
,
T
ra
n
s
a
c
t
i
o
n
s
B:
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
3
5
,
n
o
.
8
,
2
0
2
2
,
d
o
i
:
1
0
.
5
8
2
9
/
I
JE.
2
0
2
2
.
3
5
.
0
8
B
.
0
1
.
[
8
]
S
.
Te
n
a
,
R
.
H
a
r
t
a
n
t
o
,
a
n
d
I
.
A
r
d
i
y
a
n
t
o
,
“
C
o
n
t
e
n
t
-
b
a
se
d
i
m
a
g
e
r
e
t
r
i
e
v
a
l
f
o
r
t
r
a
d
i
t
i
o
n
a
l
I
n
d
o
n
e
si
a
n
w
o
v
e
n
f
a
b
r
i
c
i
ma
g
e
s
u
si
n
g
a
mo
d
i
f
i
e
d
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
m
e
t
h
o
d
,
”
J
o
u
r
n
a
l
o
f
I
m
a
g
i
n
g
,
v
o
l
.
9
,
n
o
.
8
,
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
j
i
ma
g
i
n
g
9
0
8
0
1
6
5
.
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.
14
,
No
.
2
,
J
u
n
e
2
0
2
5
:
4
3
0
-
438
438
[
9
]
S
.
A
g
r
a
w
a
l
,
A
.
C
h
o
w
d
h
a
r
y
,
S
.
A
g
a
r
w
a
l
a
,
V
.
M
a
y
y
a
,
a
n
d
S
.
K
a
ma
t
h
,
“
C
o
n
t
e
n
t
-
b
a
s
e
d
m
e
d
i
c
a
l
i
ma
g
e
r
e
t
r
i
e
v
a
l
sy
st
e
m
f
o
r
l
u
n
g
d
i
s
e
a
se
s
u
s
i
n
g
d
e
e
p
C
N
N
s,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
(
S
i
n
g
a
p
o
re)
,
v
o
l
.
1
4
,
n
o
.
7
,
p
p
.
3
6
1
9
–
3
6
2
7
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
0
7
/
s
4
1
8
7
0
-
022
-
0
1
0
0
7
-
7.
[
1
0
]
J.
A
.
G
.
S
w
e
e
t
a
a
n
d
B
.
S
i
v
a
g
a
m
i
,
“
S
O
M
b
a
s
e
d
mu
l
t
i
-
st
a
g
e
C
B
I
R
me
t
h
o
d
u
s
i
n
g
t
h
e
f
e
a
t
u
r
e
s
o
f
C
P
V
TH
,
P
O
P
M
V
a
n
d
M
D
L
B
P
i
n
t
e
g
r
a
t
e
d
w
i
t
h
mu
l
t
i
-
q
u
e
r
y
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
I
n
t
e
l
l
i
g
e
n
t
E
n
g
i
n
e
e
ri
n
g
a
n
d
S
y
st
e
m
s
,
v
o
l
.
1
7
,
n
o
.
4
,
p
p
.
6
2
2
–
6
3
3
,
2
0
2
4
,
d
o
i
:
1
0
.
2
2
2
6
6
/
I
JI
ES2
0
2
4
.
0
8
3
1
.
4
7
.
[
1
1
]
F
.
A
h
ma
d
,
“
D
e
e
p
i
m
a
g
e
r
e
t
r
i
e
v
a
l
u
s
i
n
g
a
r
t
i
f
i
c
i
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
i
n
t
e
r
p
o
l
a
t
i
o
n
a
n
d
i
n
d
e
x
i
n
g
b
a
se
d
o
n
si
mi
l
a
r
i
t
y
me
a
su
r
e
me
n
t
,
”
C
AAI Tr
a
n
sa
c
t
i
o
n
s
o
n
I
n
t
e
l
l
i
g
e
n
c
e
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
7
,
n
o
.
2
,
p
p
.
2
0
0
–
2
1
8
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
4
9
/
c
i
t
2
.
1
2
0
8
3
.
[
1
2
]
V
.
S
.
M
a
h
a
l
l
e
,
N
.
M
.
K
a
n
d
o
i
,
a
n
d
S
.
B
.
P
a
t
i
l
,
“
A
p
o
w
e
r
f
u
l
m
e
t
h
o
d
f
o
r
i
n
t
e
r
a
c
t
i
v
e
c
o
n
t
e
n
t
-
b
a
s
e
d
i
m
a
g
e
r
e
t
r
i
e
v
a
l
b
y
v
a
r
i
a
b
l
e
c
o
m
p
r
e
s
s
e
d
c
o
n
v
o
l
u
t
i
o
n
a
l
i
n
f
o
n
e
u
r
a
l
n
e
t
w
o
r
k
s
,
”
V
i
s
u
a
l
C
o
m
p
u
t
e
r
,
v
o
l
.
4
0
,
n
o
.
8
,
p
p
.
5
2
5
9
–
5
2
8
5
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
0
0
3
7
1
-
023
-
03104
-
5.
[
1
3
]
M
.
C
.
C
h
i
u
,
Y
.
H
.
L
e
e
,
a
n
d
T.
M
.
C
h
e
n
,
“
I
n
t
e
g
r
a
t
i
n
g
c
o
n
t
e
n
t
-
b
a
se
d
i
ma
g
e
r
e
t
r
i
e
v
a
l
a
n
d
d
e
e
p
l
e
a
r
n
i
n
g
t
o
i
m
p
r
o
v
e
w
a
f
e
r
b
i
n
ma
p
d
e
f
e
c
t
p
a
t
t
e
r
n
s
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
J
o
u
r
n
a
l
o
f
I
n
d
u
s
t
ri
a
l
a
n
d
Pr
o
d
u
c
t
i
o
n
En
g
i
n
e
e
ri
n
g
,
v
o
l
.
3
9
,
n
o
.
8
,
p
p
.
6
1
4
–
6
2
8
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
8
0
/
2
1
6
8
1
0
1
5
.
2
0
2
2
.
2
0
7
4
1
5
5
.
[
1
4
]
N
.
K
e
i
s
h
a
m
a
n
d
A
.
N
e
e
l
i
ma
,
“
Ef
f
i
c
i
e
n
t
c
o
n
t
e
n
t
-
b
a
s
e
d
i
ma
g
e
r
e
t
r
i
e
v
a
l
u
si
n
g
d
e
e
p
s
e
a
r
c
h
a
n
d
r
e
sc
u
e
a
l
g
o
r
i
t
h
m,
”
S
o
f
t
C
o
m
p
u
t
i
n
g
,
v
o
l
.
2
6
,
n
o
.
4
,
p
p
.
1
5
9
7
–
1
6
1
6
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
0
7
/
s0
0
5
0
0
-
0
2
1
-
0
6
6
6
0
-
x.
[
1
5
]
E.
B
r
e
z
n
i
k
,
E.
W
e
t
z
e
r
,
J.
L
i
n
d
b
l
a
d
,
a
n
d
N
.
S
l
a
d
o
j
e
,
“
C
r
o
ss
-
m
o
d
a
l
i
t
y
s
u
b
-
i
mag
e
r
e
t
r
i
e
v
a
l
u
s
i
n
g
c
o
n
t
r
a
s
t
i
v
e
m
u
l
t
i
m
o
d
a
l
i
ma
g
e
r
e
p
r
e
s
e
n
t
a
t
i
o
n
s
,
”
S
c
i
e
n
t
i
f
i
c
Re
p
o
rt
s
,
v
o
l
.
1
4
,
n
o
.
1
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
5
9
8
-
0
2
4
-
6
8
8
0
0
-
1.
[
1
6
]
R
.
A
.
N
a
r
a
y
a
n
a
sw
a
my
,
V
.
K
.
G
o
w
d
a
,
a
n
d
G
.
N
.
K
.
M
u
r
t
h
y
,
“
C
o
n
t
e
n
t
-
b
a
se
d
i
m
a
g
e
r
e
t
r
i
e
v
a
l
u
si
n
g
t
h
e
2
-
d
i
me
n
si
o
n
a
l
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
I
n
t
e
l
l
i
g
e
n
t
En
g
i
n
e
e
ri
n
g
a
n
d
S
y
st
e
m
s
,
v
o
l
.
1
7
,
n
o
.
4
,
p
p
.
6
7
–
7
6
,
2
0
2
4
,
d
o
i
:
1
0
.
2
2
2
6
6
/
I
JI
ES2
0
2
4
.
0
8
3
1
.
0
6
.
[
1
7
]
F
.
Ta
h
e
r
i
,
K
.
R
a
h
b
a
r
,
a
n
d
P
.
S
a
l
i
mi
,
“
Ef
f
e
c
t
i
v
e
f
e
a
t
u
r
e
s
i
n
c
o
n
t
e
n
t
-
b
a
se
d
i
m
a
g
e
r
e
t
r
i
e
v
a
l
f
r
o
m
a
c
o
mb
i
n
a
t
i
o
n
o
f
l
o
w
-
l
e
v
e
l
f
e
a
t
u
r
e
s
a
n
d
d
e
e
p
B
o
l
t
z
m
a
n
n
ma
c
h
i
n
e
,
”
Mu
l
t
i
m
e
d
i
a
T
o
o
l
s
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
8
2
,
n
o
.
2
4
,
p
p
.
3
7
9
5
9
–
3
7
9
8
2
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
4
2
-
022
-
1
3
6
7
0
-
w.
[
1
8
]
S
.
F
a
d
a
e
i
,
A
.
D
e
h
g
h
a
n
i
,
a
n
d
B
.
R
a
v
a
e
i
,
“
C
o
n
t
e
n
t
-
b
a
se
d
i
ma
g
e
r
e
t
r
i
e
v
a
l
u
s
i
n
g
m
u
l
t
i
-
sc
a
l
e
a
v
e
r
a
g
i
n
g
l
o
c
a
l
b
i
n
a
r
y
p
a
t
t
e
r
n
s
,
”
D
i
g
i
t
a
l
S
i
g
n
a
l
Pr
o
c
e
ssi
n
g
:
A
Re
v
i
e
w
J
o
u
r
n
a
l
,
v
o
l
.
1
4
6
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
d
sp
.
2
0
2
4
.
1
0
4
3
9
1
.
[
1
9
]
W
.
W
a
n
g
,
P
.
J
i
a
o
,
H
.
L
i
u
,
X
.
M
a
,
a
n
d
Z.
S
h
a
n
g
,
“
Tw
o
-
st
a
g
e
c
o
n
t
e
n
t
b
a
s
e
d
i
ma
g
e
r
e
t
r
i
e
v
a
l
u
s
i
n
g
s
p
a
r
s
e
r
e
p
r
e
se
n
t
a
t
i
o
n
a
n
d
f
e
a
t
u
r
e
f
u
si
o
n
,
”
M
u
l
t
i
m
e
d
i
a
T
o
o
l
s
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
8
1
,
n
o
.
1
2
,
p
p
.
1
6
6
2
1
–
1
6
6
4
4
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
1
0
4
2
-
0
2
2
-
1
2
3
4
8
-
7.
[
2
0
]
S
.
B
.
G
e
e
t
h
a
,
R
.
M
u
t
h
u
k
k
u
m
a
r
,
a
n
d
V
.
S
e
e
n
i
v
a
s
a
g
a
m,
“
E
n
h
a
n
c
i
n
g
sc
a
l
a
b
i
l
i
t
y
o
f
i
ma
g
e
r
e
t
r
i
e
v
a
l
u
s
i
n
g
v
i
su
a
l
f
u
si
o
n
o
f
f
e
a
t
u
r
e
d
e
s
c
r
i
p
t
o
r
s,”
I
n
t
e
l
l
i
g
e
n
t
A
u
t
o
m
a
t
i
o
n
a
n
d
S
o
f
t
C
o
m
p
u
t
i
n
g
,
v
o
l
.
3
1
,
n
o
.
3
,
p
p
.
1
7
3
7
–
1
7
5
2
,
2
0
2
2
,
d
o
i
:
1
0
.
3
2
6
0
4
/
I
A
S
C
.
2
0
2
2
.
0
1
8
8
2
2
.
[
2
1
]
K
.
R
a
h
b
a
r
a
n
d
F
.
Ta
h
e
r
i
,
“
E
n
h
a
n
c
i
n
g
i
ma
g
e
r
e
t
r
i
e
v
a
l
t
h
r
o
u
g
h
e
n
t
r
o
p
y
-
b
a
s
e
d
d
e
e
p
me
t
r
i
c
l
e
a
r
n
i
n
g
,
”
M
u
l
t
i
m
e
d
i
a
T
o
o
l
s
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
4
2
-
0
2
4
-
1
9
2
9
6
-
4.
[
2
2
]
J.
Li
,
“
C
a
l
t
e
c
h
2
5
6
d
a
t
a
set
,
”
K
a
g
g
l
e
,
2
0
1
8
,
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
w
w
w
.
k
a
g
g
l
e
.
c
o
m/
d
a
t
a
s
e
t
s
/
j
e
ssi
c
a
l
i
9
5
3
0
/
c
a
l
t
e
c
h
2
5
6
.
[
2
3
]
A
.
El
k
a
m
e
l
,
“
C
o
r
e
l
1
K
d
a
t
a
set
,
”
K
a
g
g
l
e
,
2
0
2
0
,
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s:
/
/
w
w
w
.
k
a
g
g
l
e
.
c
o
m/
d
a
t
a
s
e
t
s
/
e
l
k
a
me
l
/
c
o
r
e
l
-
i
m
a
g
e
s.
[
2
4
]
R
.
H
i
d
a
y
a
t
,
A
.
H
a
r
j
o
k
o
,
a
n
d
A
.
M
u
s
d
h
o
l
i
f
a
h
,
“
A
r
o
b
u
s
t
i
m
a
g
e
r
e
t
r
i
e
v
a
l
m
e
t
h
o
d
u
s
i
n
g
mu
l
t
i
-
h
i
e
r
a
r
c
h
i
c
a
l
a
g
g
l
o
m
e
r
a
t
i
v
e
c
l
u
s
t
e
r
i
n
g
a
n
d
D
a
v
i
s
-
B
o
u
l
d
i
n
i
n
d
e
x
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
I
n
t
e
l
l
i
g
e
n
t
E
n
g
i
n
e
e
r
i
n
g
a
n
d
S
y
st
e
m
s
,
v
o
l
.
1
5
,
n
o
.
2
,
p
p
.
4
4
1
–
4
5
3
,
2
0
2
2
,
d
o
i
:
1
0
.
2
2
2
6
6
/
i
j
i
e
s
2
0
2
2
.
0
4
3
0
.
4
0
.
[
2
5
]
S
.
M
o
h
se
n
,
A
.
M
.
A
l
i
,
E.
S
.
M
.
E
l
-
R
a
b
a
i
e
,
A
.
El
k
a
see
r
,
S
.
G
.
S
c
h
o
l
z
,
a
n
d
A
.
M
.
A
.
H
a
ssa
n
,
“
B
r
a
i
n
t
u
mo
r
c
l
a
ssi
f
i
c
a
t
i
o
n
u
s
i
n
g
h
y
b
r
i
d
si
n
g
l
e
i
m
a
g
e
s
u
p
e
r
-
r
e
s
o
l
u
t
i
o
n
t
e
c
h
n
i
q
u
e
w
i
t
h
R
e
sN
e
x
t
1
0
1
_
3
2
×
8
d
a
n
d
V
G
G
1
9
p
r
e
-
t
r
a
i
n
e
d
mo
d
e
l
s,
”
I
EE
E
Ac
c
e
ss
,
v
o
l
.
1
1
,
p
p
.
5
5
5
8
2
–
5
5
5
9
5
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
3
.
3
2
8
1
5
2
9
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Ra
ja
th
Ar
a
k
e
r
e
Na
r
a
y
a
n
a
sw
a
m
y
is
a
n
As
sista
n
t
P
ro
fe
ss
o
r
o
f
Co
m
p
u
te
r
S
c
ien
c
e
a
n
d
En
g
in
e
e
rin
g
a
t
G
S
S
S
I
n
stit
u
te
o
f
En
g
i
n
e
e
rin
g
a
n
d
Tec
h
n
o
lo
g
y
fo
r
W
o
m
e
n
,
M
y
su
r
u
a
n
d
h
a
s
a
tea
c
h
in
g
e
x
p
e
rien
c
e
o
f
1
3
y
e
a
rs.
He
p
u
rsu
i
n
g
is
a
P
h
.
D.
in
In
f
o
rm
a
ti
o
n
Re
tri
e
v
a
l,
a
n
d
C
o
m
p
u
ter
Visi
o
n
a
t
S
ri
S
id
d
h
a
rth
a
In
sti
tu
t
e
o
f
Tec
h
n
o
l
o
g
y
,
S
S
AH
E
Un
i
v
e
rsity
,
Tu
m
a
k
u
r
u
,
sp
e
c
ializin
g
in
c
o
m
p
u
ter
v
isi
o
n
,
in
f
o
rm
a
ti
o
n
re
tri
e
v
a
l,
m
a
c
h
in
e
lea
rn
in
g
,
d
e
e
p
lea
rn
in
g
,
a
n
d
ima
g
e
p
r
o
c
e
ss
in
g
.
He
is
p
a
ss
io
n
a
te
a
b
o
u
t
tea
c
h
i
n
g
a
n
d
is
d
e
d
ica
ted
to
m
e
n
t
o
rin
g
u
n
d
e
r
g
ra
d
u
a
te
a
n
d
g
ra
d
u
a
te
st
u
d
e
n
ts
i
n
th
e
ir
re
se
a
rc
h
e
n
d
e
a
v
o
r
s.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
ra
jath
1
9
1
9
8
5
@g
m
a
il
.
c
o
m
.
Vid
y
a
l
a
k
sh
m
i
K
r
ish
n
e
G
o
wda
re
c
e
iv
e
d
a
B.
E
.
d
e
g
re
e
in
2
0
0
5
a
t
S
ri
S
id
d
h
a
rt
h
a
In
stit
u
te
o
f
Tec
h
n
o
l
o
g
y
,
Tu
m
a
k
u
ru
,
Ka
rn
a
tak
a
,
a
n
M
.
Tec
h
d
e
g
r
e
e
in
2
0
1
1
a
t
t
h
e
Na
ti
o
n
a
l
In
stit
u
te
o
f
E
n
g
i
n
e
e
rin
g
,
M
y
s
u
r
u
,
Ka
rn
a
tak
a
,
a
n
d
a
P
h
.
D
in
2
0
2
0
fo
rm
VTU
Be
lag
a
v
i,
Ka
rn
a
tak
a
.
S
h
e
is
c
u
rre
n
tl
y
wo
rk
i
n
g
a
s
a
n
As
sista
n
t
P
ro
fe
ss
o
r
i
n
t
h
e
D
e
p
a
rtme
n
t
o
f
In
fo
rm
a
ti
o
n
S
c
ien
c
e
a
n
d
En
g
in
e
e
rin
g
a
n
d
h
a
s
a
tea
c
h
in
g
e
x
p
e
rien
c
e
o
f
1
2
y
e
a
rs.
He
r
re
se
a
rc
h
in
tere
st
in
c
lu
d
e
s
Co
m
p
u
ter
Ne
tw
o
rk
s,
wire
les
s
c
o
m
m
u
n
ica
ti
o
n
s
,
a
n
d
Im
a
g
e
p
r
o
c
e
ss
in
g
.
S
h
e
h
a
s
4
p
a
p
e
rs
a
n
d
6
c
o
n
fe
re
n
c
e
s
p
u
b
li
s
h
e
d
in
p
e
e
r
-
re
v
iew
e
d
I
n
tern
a
ti
o
n
a
l
Jo
u
rn
a
ls
in
h
e
r
c
re
d
it
a
n
d
h
a
s
re
c
e
iv
e
d
th
e
b
e
st
p
a
p
e
r
a
wa
rd
(IE
EE
-
2
0
1
9
)
a
re
s
e
a
rc
h
p
a
p
e
r
h
a
s
b
e
e
n
p
u
b
li
s
h
e
d
a
s
a
c
h
a
p
ter
in
a
b
o
o
k
c
a
ll
e
d
HCMC
.
As
a
n
e
x
ten
si
o
n
o
f
h
e
r
wo
rk
a
n
d
k
n
o
wle
d
g
e
,
sh
e
a
lso
is
g
u
i
d
in
g
P
h
.
D
.
st
u
d
e
n
ts
i
n
th
e
sa
m
e
field
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
v
id
y
a
lak
sh
m
i@ss
it
.
e
d
u
.
i
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.