I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Ro
bo
t
ics a
nd
Aut
o
m
a
t
io
n
(
I
J
RA)
Vo
l.
1
5
,
No
.
1
,
Ma
r
ch
20
2
6
,
p
p
.
1
07
~
1
21
I
SS
N:
2722
-
2
5
8
6
,
DOI
:
1
0
.
1
1
5
9
1
/i
jr
a
.
v
1
5
i
1
.
pp
1
07
-
1
21
107
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
r
a
.
ia
esco
r
e.
co
m
Autono
mo
us
reco
nstruction
of str
i
p
-
shredde
d docu
ments v
ia
self
-
supe
rv
ised de
ep learning
and g
lo
ba
l optimiza
tion
Yi
-
Cha
ng
Wu,
P
ei
-
Sh
a
n Ch
i
a
ng
,
Ya
o
-
Cheng
L
iu
D
e
p
a
r
t
me
n
t
F
o
r
e
n
s
i
c
S
c
i
e
n
c
e
D
i
v
i
si
o
n
,
I
n
v
e
st
i
g
a
t
i
o
n
B
u
r
e
a
u
,
M
i
n
i
s
t
r
y
o
f
J
u
st
i
c
e
,
T
a
i
p
e
i
,
Ta
i
w
a
n
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
l 7
,
2
0
2
5
R
ev
is
ed
Feb
1
3
,
2
0
2
6
Acc
ep
ted
Feb
2
1
,
2
0
2
6
Au
to
n
o
m
o
u
s
re
c
o
n
stru
c
ti
o
n
o
f
m
e
c
h
a
n
ica
ll
y
sh
re
d
d
e
d
d
o
c
u
m
e
n
ts
is
a
lab
o
r
-
in
ten
siv
e
c
h
a
l
len
g
e
i
n
fo
re
n
sic
a
n
d
a
rc
h
iv
a
l
w
o
rk
fl
o
ws
,
p
a
rti
c
u
larly
fo
r
sc
rip
ts
with
c
o
m
p
lex
stru
c
t
u
r
e
s
su
c
h
a
s
S
imp
li
fied
Ch
i
n
e
se
.
Wh
il
e
trad
it
io
n
a
l
m
a
n
u
a
l
re
a
ss
e
m
b
ly
is
ted
io
u
s,
e
x
isti
n
g
d
i
g
it
a
l
to
o
ls
t
y
p
i
c
a
ll
y
re
ly
o
n
e
x
ten
si
v
e
h
u
m
a
n
i
n
terv
e
n
ti
o
n
.
Th
is
p
a
p
e
r
p
re
se
n
ts
a
n
a
u
to
m
a
ted
re
a
ss
e
m
b
ly
fra
m
e
wo
rk
t
h
a
t
i
n
teg
ra
tes
a
li
g
h
twe
ig
h
t
c
o
n
v
o
lu
t
io
n
a
l
fe
a
tu
re
e
x
trac
to
r
with
g
lo
b
a
l
c
o
m
b
in
a
to
rial
o
p
t
imiz
a
ti
o
n
.
B
y
a
d
a
p
ti
n
g
t
h
e
e
sta
b
li
sh
e
d
S
q
u
e
e
z
e
Ne
t
v
1
.
1
b
a
c
k
b
o
n
e
,
we
e
m
p
lo
y
a
tas
k
-
sp
e
c
ifi
c
se
lf
-
su
p
e
rv
ise
d
lea
rn
in
g
stra
teg
y
tra
in
e
d
o
n
sy
n
t
h
e
ti
c
a
ll
y
sh
re
d
d
e
d
sa
m
p
les
,
e
n
a
b
li
n
g
t
h
e
a
d
a
p
ted
m
o
d
e
l
t
o
c
a
p
tu
re
lo
c
a
l
stro
k
e
c
o
n
ti
n
u
it
y
a
n
d
e
d
g
e
-
g
e
o
m
e
try
c
u
e
s
with
o
u
t
m
a
n
u
a
l
a
n
n
o
tatio
n
.
Th
e
fra
m
e
wo
rk
i
n
fe
rs
p
a
irwise
re
latio
n
sh
i
p
s
fro
m
c
a
li
b
ra
ted
e
d
g
e
-
re
g
io
n
in
p
u
ts,
o
rg
a
n
izin
g
c
o
m
p
a
ti
b
il
it
y
sc
o
re
s
in
to
a
n
a
sy
m
m
e
tri
c
trav
e
li
n
g
sa
les
m
a
n
p
ro
b
lem
(ATS
P
)
f
o
r
m
u
latio
n
.
Th
e
o
p
ti
m
a
l
fra
g
m
e
n
t
se
q
u
e
n
c
e
is
so
l
v
e
d
d
e
term
in
isti
c
a
ll
y
u
sin
g
t
h
e
Co
n
c
o
r
d
e
TS
P
so
lv
e
r,
y
ield
i
n
g
a
g
lo
b
a
ll
y
c
o
n
siste
n
t
re
c
o
n
stru
c
ti
o
n
.
Ex
p
e
rime
n
tal
re
su
l
ts
o
n
p
h
y
si
c
a
ll
y
sh
re
d
d
e
d
d
o
c
u
m
e
n
ts
d
e
m
o
n
stra
te
re
c
o
n
stru
c
ti
o
n
a
c
c
u
ra
c
ies
o
f
8
6
.
5
%
fo
r
S
imp
l
ifi
e
d
C
h
i
n
e
se
a
n
d
9
4
.
8
%
fo
r
Wes
tern
sc
rip
ts.
Th
e
se
re
su
lt
s
in
d
ica
te
th
a
t
th
e
p
ro
p
o
se
d
p
i
p
e
li
n
e
e
ffe
c
ti
v
e
ly
g
e
n
e
ra
li
z
e
s
fro
m
sy
n
t
h
e
ti
c
trai
n
in
g
d
a
ta
to
re
a
l
-
wo
rl
d
sc
e
n
a
rio
s
,
p
r
o
v
i
d
in
g
a
p
ra
c
ti
c
a
l,
h
ig
h
-
t
h
ro
u
g
h
p
u
t
f
o
u
n
d
a
ti
o
n
fo
r
a
u
to
m
a
ted
d
o
c
u
m
e
n
t
re
c
o
v
e
ry
u
n
d
e
r
c
o
m
p
u
tat
io
n
a
l
c
o
n
stra
i
n
ts t
y
p
ica
l
o
f
ro
b
o
t
ic o
r
e
m
b
e
d
d
e
d
s
y
ste
m
s.
K
ey
w
o
r
d
s
:
Au
to
n
o
m
o
u
s
r
ec
o
n
s
tr
u
ctio
n
C
h
in
ese
tex
t p
r
o
ce
s
s
in
g
Fo
r
en
s
ic
s
cien
ce
Fu
lly
co
n
v
o
l
u
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
Glo
b
al
o
p
tim
izatio
n
Self
-
s
u
p
er
v
is
ed
lear
n
in
g
Strip
-
s
h
r
ed
d
ed
d
o
cu
m
en
ts
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
:
Yi
C
h
an
g
W
u
Fo
r
en
s
ic
Scien
ce
Div
is
io
n
,
I
n
v
esti
g
atio
n
B
u
r
ea
u
,
Min
is
tr
y
o
f
J
u
s
tice
No
.
7
4
,
Z
h
o
n
g
h
u
a
R
d
.
,
Xin
d
ia
n
Dis
t.,
New
T
aip
ei
C
ity
2
3
1
,
T
aiwa
n
(
R
.
O.
C
.
)
E
m
ail: sh
in
ten
wu
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
r
ec
o
n
s
tr
u
ctio
n
o
f
s
h
r
ed
d
ed
p
ap
er
d
o
cu
m
en
ts
is
a
to
p
ic
o
f
s
ig
n
if
ican
t
r
elev
a
n
ce
to
f
o
r
en
s
ic
s
cien
ce
,
in
v
esti
g
ativ
e
d
is
cip
li
n
es,
an
d
ar
ch
ae
o
lo
g
y
,
an
d
h
a
s
g
ar
n
er
ed
in
cr
ea
s
in
g
atten
tio
n
in
r
ec
en
t
y
ea
r
s
.
Fo
llo
win
g
th
e
f
all
o
f
th
e
B
er
lin
W
all
in
1
9
8
9
,
th
e
Mi
n
is
tr
y
f
o
r
State
Secu
r
ity
o
f
E
ast
Ger
m
an
y
(
Stas
i)
attem
p
ted
to
d
estro
y
a
v
ast
a
m
o
u
n
t
o
f
in
tellig
en
ce
d
o
c
u
m
en
ts
,
r
esu
ltin
g
in
o
v
er
1
6
,
0
0
0
b
ag
s
o
f
s
h
r
e
d
d
ed
m
ater
ials
.
As
s
h
o
wn
in
Fig
u
r
e
1
[
1
]
,
an
em
p
lo
y
ee
is
h
o
l
d
in
g
f
r
ag
m
en
ts
o
f
th
e
s
h
r
e
d
d
ed
Stas
i
f
iles
.
T
h
e
Ger
m
an
g
o
v
e
r
n
m
en
t m
o
b
ilized
ap
p
r
o
x
im
ately
4
0
p
e
r
s
o
n
n
el,
an
d
af
ter
s
ix
y
ea
r
s
o
f
ef
f
o
r
t,
o
n
ly
3
0
0
d
o
cu
m
e
n
ts
wer
e
s
u
cc
ess
f
u
lly
r
ec
o
n
s
tr
u
ct
ed
[
2
]
.
T
h
e
r
ec
o
v
er
y
p
r
o
ce
s
s
r
em
ain
s
o
n
g
o
in
g
to
th
is
d
ay
.
Sch
au
er
et
a
l.
[
3
]
s
u
g
g
ested
th
at
s
h
r
ed
d
e
d
d
o
c
u
m
en
ts
ca
n
b
e
r
eg
ar
d
ed
as
a
v
ar
iatio
n
o
f
co
n
v
en
tio
n
al
jig
s
aw
p
u
zz
les.
Pu
zz
le
-
s
o
lv
in
g
tech
n
i
q
u
es
h
a
v
e
b
ee
n
ex
ten
s
iv
ely
ap
p
lied
in
d
iv
er
s
e
d
o
m
ain
s
,
i
n
clu
d
in
g
b
io
s
en
s
i
n
g
[
4
]
–
[
6
]
,
im
a
g
e
r
ec
o
n
s
tr
u
ctio
n
[
7
]
–
[
1
0
]
,
s
p
ee
ch
u
n
s
cr
am
b
lin
g
[
1
1
]
,
[
1
2
]
,
b
an
k
in
g
[
1
3
]
,
an
d
f
o
r
e
n
s
ic
an
aly
s
is
[
1
4
]
–
[
1
6
]
.
I
n
ar
ch
ae
o
lo
g
y
,
s
u
ch
tech
n
iq
u
es
ass
i
s
t
in
th
e
id
en
tific
atio
n
o
f
cu
ltu
r
al
h
er
itag
e
an
d
th
e
r
esto
r
atio
n
o
f
ar
tifa
cts
[
1
7
]
–
[
2
0
]
.
I
n
leg
al
in
v
esti
g
a
tio
n
s
,
th
ey
ar
e
em
p
lo
y
e
d
to
r
ec
o
v
er
d
elib
er
ately
d
am
ag
ed
d
o
cu
m
e
n
ts
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
2
5
8
6
I
AE
S
I
n
t
J
R
o
b
&
A
u
to
m
,
Vo
l
.
1
5
,
No
.
1
,
Ma
r
ch
20
2
6
:
1
07
-
1
21
108
p
h
o
to
g
r
ap
h
s
[
2
1
]
–
[
2
3
]
.
Fu
r
th
e
r
m
o
r
e,
in
m
ilit
ar
y
c
o
n
tex
ts
—
p
ar
ticu
lar
ly
in
co
m
b
at
z
o
n
es
—
th
ey
a
r
e
cr
u
cial
f
o
r
r
ec
o
g
n
izin
g
r
em
n
an
ts
o
f
d
estr
o
y
ed
d
o
cu
m
en
ts
.
I
n
s
u
m
m
ar
y
,
f
r
ag
m
e
n
t
r
ec
o
n
s
tr
u
ctio
n
tec
h
n
o
lo
g
ies
p
lay
a
v
ital
r
o
le
in
th
e
r
etr
iev
al
o
f
f
o
r
e
n
s
ic
ev
id
en
ce
,
p
r
eser
v
atio
n
o
f
c
u
ltu
r
al
ass
ets,
an
d
ac
q
u
is
itio
n
o
f
m
ilit
ar
y
in
tellig
en
ce
.
Du
e
to
th
e
i
r
r
eg
u
la
r
s
h
ap
es,
d
iv
er
s
e
s
izes,
an
d
la
r
g
e
n
u
m
b
er
o
f
r
em
ai
n
in
g
f
r
ag
m
e
n
t
s
,
m
an
u
al
r
ec
o
n
s
tr
u
ctio
n
is
h
ig
h
ly
in
ef
f
i
cien
t
an
d
o
f
ten
in
f
ea
s
ib
le.
C
o
n
s
eq
u
en
tly
,
t
h
e
co
m
p
lete
r
ec
o
n
s
tr
u
ctio
n
p
r
o
ce
s
s
,
en
co
m
p
ass
in
g
s
ca
n
n
in
g
,
an
al
y
s
is
,
an
d
r
ea
s
s
em
b
ly
,
m
u
s
t
b
e
am
en
ab
le
to
f
u
ll
au
to
m
atio
n
.
T
h
e
m
eth
o
d
o
lo
g
y
p
r
esen
ted
h
er
e
is
d
esig
n
ed
as
a
co
r
e
co
m
p
o
n
e
n
t
with
in
an
a
u
to
m
ated
p
i
p
elin
e.
T
h
e
in
teg
r
atio
n
b
eg
in
s
at
th
e
d
o
cu
m
e
n
t
d
ig
itizatio
n
s
tag
e,
wh
er
e
h
ig
h
-
r
eso
lu
tio
n
o
v
er
h
ea
d
s
ca
n
n
er
s
ca
p
tu
r
e
f
r
ag
m
en
t
im
ag
es
in
a
co
n
s
is
ten
t
an
d
r
ep
ea
tab
le
m
a
n
n
er
.
T
h
ese
d
ig
itized
f
r
ag
m
e
n
ts
ar
e
th
e
n
d
ir
ec
tly
s
tr
ea
m
ed
in
to
th
e
p
r
o
p
o
s
e
d
d
ee
p
lear
n
in
g
-
b
ased
r
ec
o
n
s
t
r
u
ctio
n
f
r
am
ewo
r
k
,
wh
ich
p
er
f
o
r
m
s
co
m
p
atib
ilit
y
esti
m
atio
n
an
d
g
lo
b
al
r
ea
s
s
em
b
ly
with
o
u
t
h
u
m
an
in
ter
v
en
tio
n
.
B
y
c
o
m
b
in
i
n
g
p
h
y
s
ical
d
o
cu
m
en
t
s
en
s
in
g
with
au
to
n
o
m
o
u
s
v
is
u
al
p
er
ce
p
tio
n
an
d
o
p
tim
izatio
n
-
b
ased
d
ec
is
io
n
m
ak
in
g
,
th
e
p
r
o
p
o
s
ed
s
y
s
tem
n
atu
r
ally
f
i
ts
in
to
r
o
b
o
tic
an
d
au
to
m
ated
d
o
cu
m
e
n
t p
r
o
c
ess
i
n
g
wo
r
k
f
lo
ws.
Fig
u
r
e
1
.
A
s
taf
f
m
e
m
b
er
d
is
p
lay
in
g
s
h
r
ed
d
ed
d
o
cu
m
e
n
ts
o
r
ig
in
ally
d
estro
y
e
d
b
y
th
e
Stas
i,
an
o
r
g
an
izatio
n
k
n
o
wn
f
o
r
its
ex
ten
s
iv
e
in
f
o
r
m
an
t n
etwo
r
k
u
s
ed
to
m
o
n
ito
r
E
ast Ge
r
m
an
citizen
s
[
1
]
T
h
e
ef
f
icien
c
y
o
f
m
an
u
al
r
ec
o
n
s
tr
u
ctio
n
o
f
s
h
r
ed
d
e
d
d
o
cu
m
en
ts
is
in
f
lu
en
ce
d
b
y
s
ev
er
al
f
ac
to
r
s
,
in
clu
d
in
g
t
h
e
co
m
p
lex
ity
o
f
th
e
d
o
cu
m
e
n
t,
th
e
n
atu
r
e
o
f
th
e
d
estru
ctio
n
p
r
o
ce
s
s
,
an
d
th
e
n
u
m
b
er
a
n
d
s
h
ap
e
o
f
th
e
f
r
a
g
m
en
ts
.
E
v
en
with
th
e
aid
o
f
ar
tific
ial
in
tellig
en
ce
s
y
s
tem
s
,
th
ese
f
ac
to
r
s
s
ig
n
if
ican
tly
im
p
ac
t
t
h
e
tim
e
r
eq
u
ir
ed
f
o
r
r
ec
o
n
s
tr
u
ctio
n
.
R
esear
ch
o
n
s
h
r
ed
d
ed
d
o
cu
m
e
n
t
r
ec
o
n
s
tr
u
ctio
n
en
c
o
m
p
ass
es
m
u
ltip
le
s
u
b
f
ield
s
,
wh
ich
ar
e
o
f
ten
d
i
f
f
er
en
tiated
b
y
s
h
r
ed
d
i
n
g
m
eth
o
d
s
(
e.
g
.
,
m
ec
h
an
ical
cu
ttin
g
o
r
m
an
u
al
tear
in
g
)
,
d
o
cu
m
en
t
ty
p
es
(
e.
g
.
,
b
lack
-
an
d
-
wh
ite
o
r
co
lo
r
,
tex
t
-
o
n
ly
,
im
ag
es,
o
r
m
ix
ed
co
n
ten
t
)
,
an
d
r
ec
o
n
s
tr
u
ctio
n
a
p
p
r
o
ac
h
es
(
e.
g
.
,
f
u
lly
au
t
o
m
ated
o
r
h
u
m
a
n
-
ass
is
ted
s
y
s
tem
s
)
.
C
u
r
r
en
tly
,
m
o
s
t
s
tu
d
ies
lev
er
ag
in
g
m
o
d
er
n
tech
n
o
lo
g
y
f
o
r
r
ec
o
n
s
tr
u
ctio
n
f
o
c
u
s
o
n
s
im
u
latin
g
s
h
r
ed
d
in
g
s
ce
n
ar
io
s
,
wh
er
e
id
en
tify
i
n
g
a
n
d
i
n
teg
r
atin
g
ad
jace
n
t
f
r
ag
m
e
n
ts
r
em
ain
s
th
e
co
r
e
ch
allen
g
e.
G
iv
en
th
at
m
o
s
t
d
o
cu
m
en
ts
ar
e
d
estro
y
ed
v
ia
m
ec
h
an
ical
s
h
r
ed
d
in
g
r
ath
er
th
an
m
an
u
al
tear
in
g
,
a
n
d
co
n
s
id
er
i
n
g
th
at
b
lack
-
an
d
-
wh
ite
tex
t
d
o
cu
m
e
n
ts
ar
e
th
e
m
o
s
t
co
m
m
o
n
ty
p
e,
th
is
s
tu
d
y
f
o
cu
s
es
o
n
d
e
v
elo
p
in
g
a
r
ec
o
n
s
tr
u
ctio
n
s
y
s
tem
f
o
r
m
ec
h
an
ically
s
h
r
ed
d
ed
b
lac
k
-
an
d
-
wh
ite
p
r
in
ted
d
o
cu
m
e
n
ts
.
Ad
d
itio
n
ally
,
we
ce
n
ter
o
u
r
wo
r
k
o
n
Simp
lif
ied
C
h
in
ese
—
a
wid
ely
u
s
ed
y
et
u
n
d
e
r
-
ex
p
lo
r
ed
lan
g
u
ag
e
i
n
th
e
liter
atu
r
e.
R
ec
o
n
s
tr
u
ctio
n
r
esear
ch
f
u
n
d
a
m
en
tally
r
elies
o
n
v
is
u
al
cu
es.
T
h
is
p
r
esen
ts
a
p
ar
ticu
la
r
ch
a
llen
g
e
f
o
r
tex
t
-
b
ased
d
o
c
u
m
en
ts
d
u
e
to
th
eir
lim
ited
co
lo
r
in
f
o
r
m
ati
o
n
(
i.e
.
,
b
in
ar
y
b
lack
-
an
d
-
wh
ite
ap
p
ea
r
an
ce
)
.
E
ar
l
y
ap
p
r
o
ac
h
es
to
tex
t
d
o
cu
m
e
n
t
r
ec
o
n
s
tr
u
ctio
n
p
r
im
ar
ily
r
eli
ed
o
n
p
u
zz
le
-
p
iece
s
h
ap
e
f
ea
tu
r
es
to
d
eter
m
in
e
f
r
ag
m
en
t
ad
jace
n
cy
[
2
4
]
–
[
2
6
]
.
Ho
wev
er
,
s
in
ce
m
ec
h
an
icall
y
s
h
r
ed
d
ed
f
r
ag
m
e
n
ts
ten
d
to
h
av
e
h
ig
h
l
y
s
im
ilar
s
h
ap
es,
co
n
v
en
tio
n
al
s
h
ap
e
-
b
ased
alg
o
r
ith
m
s
—
s
u
ch
as
th
o
s
e
u
s
in
g
p
o
ly
g
o
n
al
ap
p
r
o
x
i
m
atio
n
to
s
im
p
lify
cu
r
v
e
m
atch
i
n
g
—
ar
e
u
n
s
u
itab
le
f
o
r
o
u
r
tar
g
et
s
ce
n
ar
io
.
So
m
e
m
eth
o
d
s
[
2
7
]
–
[
3
1
]
h
av
e
f
o
cu
s
ed
o
n
co
lo
r
d
is
tr
ib
u
tio
n
,
u
tili
zin
g
co
lo
r
in
f
o
r
m
atio
n
to
ass
ess
th
e
co
m
p
atib
ilit
y
b
etwe
en
f
r
a
g
m
en
ts
.
W
h
ile
s
u
ch
ap
p
r
o
ac
h
es
lev
er
ag
e
th
e
r
ich
er
in
f
o
r
m
atio
n
p
r
esen
t
in
co
lo
r
im
ag
es
co
m
p
ar
ed
to
b
i
n
ar
y
(
b
lack
-
an
d
-
wh
ite)
d
ata,
th
ey
ty
p
ically
n
eg
lect
th
e
ed
g
e
d
eg
r
ad
atio
n
ca
u
s
ed
b
y
m
ec
h
an
ical
s
h
r
ed
d
i
n
g
.
As
a
r
esu
lt,
th
ey
ar
e
n
o
t
ap
p
licab
le
to
th
e
r
ec
o
n
s
tr
u
ctio
n
o
f
b
lack
-
an
d
-
w
h
ite
tex
tu
al
d
o
cu
m
en
ts
.
L
in
et
a
l.
[
3
2
]
in
tr
o
d
u
ce
d
a
n
a
p
p
r
o
ac
h
u
s
in
g
av
er
a
g
e
wo
r
d
len
g
th
t
o
r
ep
r
esen
t
E
n
g
lis
h
ch
ar
ac
te
r
s
an
d
a
p
p
lied
f
r
ag
m
en
t
e
n
c
o
d
in
g
to
d
escr
ib
e
d
o
cu
m
e
n
t
lay
o
u
ts
.
Pra
n
d
ts
tetter
an
d
R
aid
l
[
3
3
]
f
o
r
m
u
lated
t
h
e
r
ec
o
n
s
tr
u
ctio
n
o
f
s
tr
ip
-
s
h
r
e
d
d
ed
d
o
cu
m
en
ts
as
Evaluation Warning : The document was created with Spire.PDF for Python.
I
AE
S
I
n
t
J
R
o
b
&
A
u
to
m
I
SS
N:
2722
-
2
5
8
6
A
u
to
n
o
m
o
u
s
r
ec
o
n
s
tr
u
ctio
n
o
f str
ip
-
s
h
r
ed
d
ed
d
o
cu
men
ts
vi
a
s
elf
-
s
u
p
ervis
ed
d
ee
p
lea
r
n
in
g
…
(
Yi
-
C
h
a
n
g
Wu
)
109
a
v
ar
ian
t
o
f
th
e
class
ical
T
r
av
elin
g
Salesma
n
Pro
b
lem
(
T
SP
)
,
an
d
p
r
o
p
o
s
ed
a
v
ar
ia
b
le
n
ei
g
h
b
o
r
h
o
o
d
s
ea
r
ch
m
eth
o
d
to
o
p
tim
ize
th
e
r
ec
o
n
s
tr
u
ctio
n
p
r
o
ce
s
s
in
a
s
em
i
-
a
u
to
m
ated
f
r
am
ewo
r
k
.
B
alm
e
[
3
4
]
an
d
Mo
r
a
n
d
ell
[
3
5
]
em
p
lo
y
ed
b
in
a
r
y
im
ag
e
r
ep
r
esen
tatio
n
s
to
m
o
d
el
th
e
b
l
ac
k
-
an
d
-
wh
ite
ap
p
ea
r
an
ce
o
f
t
ex
tu
al
d
o
cu
m
e
n
ts
.
T
h
ey
a
d
d
r
ess
ed
th
e
is
s
u
e
o
f
v
er
tical
m
is
alig
n
m
en
t
b
etwe
e
n
ad
jace
n
t
f
r
a
g
m
en
ts
b
y
r
esp
ec
tiv
ely
co
m
p
u
tin
g
weig
h
ted
p
i
x
el
co
r
r
elatio
n
an
d
q
u
an
tify
in
g
th
e
d
e
g
r
ee
o
f
m
is
alig
n
m
en
t
b
etwe
en
b
lack
p
i
x
el
r
eg
i
o
n
s
.
L
i
et
a
l.
[
3
6
]
f
u
r
th
er
ad
v
a
n
ce
d
th
ese
r
u
le
-
b
ased
ap
p
r
o
ac
h
es
b
y
u
s
in
g
g
eo
m
etr
ic
tem
p
lates
an
d
b
la
n
k
-
ar
ea
d
etec
tio
n
to
r
ec
o
n
s
tr
u
ct
r
ec
tan
g
u
lar
E
n
g
lis
h
f
r
ag
m
e
n
ts
.
Ho
wev
er
,
s
u
ch
m
eth
o
d
s
d
ep
en
d
h
ea
v
ily
o
n
t
h
e
p
r
ed
ictab
ilit
y
o
f
W
ester
n
ch
ar
ac
ter
lay
o
u
ts
an
d
o
f
ten
s
tr
u
g
g
le
with
th
e
ir
r
e
g
u
lar
s
tr
o
k
e
d
en
s
ities
o
f
o
th
er
s
cr
ip
ts
.
So
m
e
s
tu
d
ies
h
av
e
ev
alu
ated
f
r
ag
m
e
n
t
co
m
p
atib
ilit
y
b
y
an
aly
zin
g
th
e
d
eg
r
ee
o
f
ch
ar
ac
ter
c
o
n
tin
u
ity
al
o
n
g
s
h
r
ed
d
e
d
ed
g
es.
Fo
r
in
s
tan
ce
,
Per
l
et
a
l.
[
3
7
]
i
n
v
esti
g
ated
th
e
u
s
e
o
f
OC
R
f
ea
tu
r
es
f
o
r
s
u
p
e
r
v
is
ed
c
h
ar
ac
t
er
r
ec
o
g
n
itio
n
an
d
alig
n
m
en
t,
p
r
o
p
o
s
in
g
a
n
E
n
g
l
is
h
OC
R
-
b
ased
m
eth
o
d
th
at
e
m
p
lo
y
s
ch
a
r
ac
ter
h
is
to
g
r
am
s
to
m
atch
f
r
ag
m
e
n
t
b
o
u
n
d
ar
ies.
T
h
eir
f
in
d
in
g
s
in
d
icate
d
th
at
r
ec
o
n
s
tr
u
cti
o
n
ac
cu
r
ac
y
d
ec
r
ea
s
es
n
o
n
-
m
o
n
o
to
n
ic
ally
as
th
e
n
u
m
b
er
o
f
tex
t
lin
es
d
im
in
is
h
es.
Paix
ao
et
a
l.
[
3
8
]
an
al
y
ze
d
th
e
s
h
ap
es
o
f
ch
ar
ac
ter
g
r
o
u
p
i
n
g
s
to
class
if
y
d
if
f
er
en
t
s
y
m
b
o
l c
o
m
b
in
atio
n
s
an
d
ca
lc
u
late
th
e
co
m
p
atib
ilit
y
o
f
f
r
ag
m
en
t p
air
s
.
Dee
p
lear
n
in
g
h
as
ac
h
iev
ed
s
tate
-
of
-
th
e
-
ar
t
p
er
f
o
r
m
an
ce
i
n
co
m
p
u
ter
v
is
io
n
task
s
s
u
ch
as
im
ag
e
class
if
icatio
n
,
o
b
ject
d
etec
ti
o
n
,
an
d
s
eg
m
e
n
tatio
n
.
Un
li
k
e
ea
r
lier
tem
p
late
-
m
atch
in
g
tech
n
iq
u
es
[
3
6
]
,
co
n
v
o
l
u
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
(
C
NNs)
ar
e
ca
p
ab
le
o
f
ca
p
t
u
r
in
g
f
in
e
-
g
r
ain
ed
s
tr
o
k
e
c
o
n
t
in
u
ity
an
d
lear
n
i
n
g
task
-
s
p
ec
if
ic
r
ep
r
esen
tatio
n
s
d
ir
ec
tly
f
r
o
m
r
aw
p
ix
els.
Sh
o
lo
m
o
n
et
a
l.
[
3
9
]
u
s
ed
n
eu
r
al
n
etwo
r
k
s
to
p
r
e
d
ict
wh
eth
er
two
p
u
zz
le
p
iece
ed
g
es
s
h
o
u
ld
b
e
ad
jace
n
t
b
y
f
ee
d
i
n
g
th
eir
ed
g
e
p
ix
el
i
n
f
o
r
m
atio
n
in
to
t
h
e
n
etwo
r
k
.
Ho
wev
er
,
th
ese
m
eth
o
d
s
wer
e
d
esig
n
ed
f
o
r
s
y
n
th
etica
lly
g
e
n
er
ated
f
r
ag
m
en
ts
an
d
n
o
t f
o
r
r
ea
l
-
wo
r
ld
s
h
r
ed
d
ed
d
o
cu
m
e
n
ts
.
Mo
s
t
p
r
io
r
r
esear
ch
f
o
cu
s
es
o
n
W
ester
n
lan
g
u
ag
es,
wh
ich
d
if
f
er
s
ig
n
if
ica
n
tly
f
r
o
m
C
h
in
ese
in
ter
m
s
o
f
ch
ar
ac
ter
s
tr
u
ctu
r
e.
C
h
in
ese
ch
ar
ac
ter
s
ar
e
s
q
u
ar
e
-
s
h
ap
ed
,
s
p
atially
u
n
if
o
r
m
,
an
d
in
d
ep
en
d
en
t
u
n
its
,
u
n
lik
e
W
ester
n
lan
g
u
ag
es
wh
ich
ar
e
co
m
p
o
s
ed
o
f
lin
ea
r
,
h
o
r
izo
n
tally
ar
r
an
g
e
d
letter
s
.
Stan
d
ar
d
p
r
i
n
ted
C
h
in
ese
ch
ar
ac
ter
s
ex
h
ib
it
a
1
:1
h
eig
h
t
-
to
-
wid
th
r
atio
.
W
h
e
n
C
h
in
ese
d
o
cu
m
e
n
ts
ar
e
s
h
r
e
d
d
ed
,
th
e
r
esu
ltin
g
f
r
ag
m
en
ts
m
a
y
c
o
n
tain
eith
e
r
d
am
ag
ed
ch
ar
ac
ter
s
al
o
n
g
th
e
ed
g
es
o
r
b
lan
k
r
e
g
io
n
s
co
r
r
es
p
o
n
d
in
g
to
in
ter
lin
e
s
p
ac
in
g
.
T
h
is
s
tu
d
y
attem
p
ts
t
o
ad
d
r
ess
th
e
r
ec
o
n
s
tr
u
ctio
n
o
f
s
h
r
ed
d
e
d
C
h
in
ese
d
o
cu
m
e
n
t
s
u
n
d
er
r
ea
l
-
wo
r
l
d
co
n
d
itio
n
s
b
y
lev
er
a
g
in
g
d
ee
p
lear
n
in
g
m
o
d
els
[
4
0
]
in
a
s
elf
-
s
u
p
er
v
is
ed
lear
n
in
g
f
r
am
ew
o
r
k
,
e
n
ab
lin
g
la
r
g
e
-
s
ca
le
s
am
p
le
ex
tr
ac
tio
n
an
d
lear
n
in
g
f
r
o
m
u
n
lab
eled
d
ata.
Ou
r
f
in
d
in
g
s
m
ay
also
o
f
f
er
v
alu
ab
le
in
s
ig
h
ts
f
o
r
d
o
cu
m
e
n
t
r
ec
o
n
s
tr
u
ctio
n
in
lan
g
u
ag
es
with
s
im
ilar
lo
g
o
g
r
ap
h
ic
wr
itin
g
s
y
s
tem
s
,
s
u
ch
as
J
ap
an
ese
an
d
Ko
r
ea
n
.
T
h
e
r
em
ain
d
er
o
f
th
is
p
a
p
er
is
o
r
g
an
ized
as
f
o
llo
ws.
Sectio
n
2
d
etails
th
e
p
r
o
p
o
s
ed
au
to
n
o
m
o
u
s
r
ec
o
n
s
tr
u
ctio
n
f
r
am
ewo
r
k
,
e
n
co
m
p
ass
in
g
d
o
c
u
m
en
t
d
ig
itiz
atio
n
,
its
in
teg
r
atio
n
with
au
to
m
ated
s
ca
n
n
in
g
s
y
s
tem
s
,
s
elf
-
s
u
p
er
v
is
ed
s
am
p
le
g
en
e
r
atio
n
,
m
o
d
el
tr
ain
in
g
with
th
e
Sq
u
ee
ze
Net
b
ac
k
b
o
n
e,
a
n
d
th
e
g
lo
b
a
l
o
p
tim
izatio
n
s
ea
r
ch
v
ia
AT
SP
.
Sectio
n
3
r
e
p
o
r
ts
th
e
e
x
p
er
im
en
tal
r
esu
lts
,
in
clu
d
in
g
a
c
o
m
p
r
e
h
en
s
iv
e
p
er
f
o
r
m
an
ce
b
en
c
h
m
ar
k
i
n
g
a
g
ain
s
t
ex
is
tin
g
m
eth
o
d
o
lo
g
ies
.
Sectio
n
4
co
n
clu
d
es
th
e
p
ap
er
b
y
s
u
m
m
ar
izin
g
o
u
r
f
in
d
in
g
s
an
d
d
is
cu
s
s
in
g
p
o
ten
tial f
u
tu
r
e
d
ir
ec
tio
n
s
f
o
r
r
ea
l
-
wo
r
ld
r
o
b
o
tic
ap
p
licatio
n
s
.
2.
M
E
T
H
O
D
T
h
is
s
tu
d
y
aim
s
to
d
ev
elo
p
a
m
o
d
el
th
at
q
u
an
tifie
s
th
e
co
m
p
atib
ilit
y
b
etwe
en
p
air
s
o
f
d
o
cu
m
e
n
t
f
r
ag
m
en
ts
.
Du
e
to
th
e
lab
o
r
-
i
n
ten
s
iv
e
n
atu
r
e
o
f
c
r
ea
tin
g
r
e
al
-
wo
r
ld
s
h
r
ed
d
e
d
d
atasets
an
d
th
e
lack
o
f
p
u
b
lic
d
atasets
f
o
r
s
tr
ip
-
s
h
r
ed
d
ed
d
o
cu
m
en
ts
,
we
ad
o
p
t
a
s
elf
-
s
u
p
er
v
is
ed
lear
n
i
n
g
ap
p
r
o
ac
h
th
at
au
to
m
atica
lly
d
eter
m
in
es
f
r
ag
m
en
t
ad
jace
n
c
y
d
u
r
in
g
th
e
s
am
p
lin
g
p
r
o
ce
s
s
.
Sp
ec
if
ically
,
we
s
im
u
late
d
o
cu
m
en
t
s
h
r
ed
d
in
g
d
ig
itally
an
d
ex
tr
ac
t
f
r
a
g
m
en
t p
air
s
as
tr
ain
in
g
s
am
p
les.
I
n
th
is
p
r
o
ce
s
s
,
ad
jace
n
t
f
r
ag
m
en
t p
air
s
ar
e
la
b
eled
a
s
p
o
s
itiv
e
s
am
p
les,
wh
ile
n
o
n
-
ad
jace
n
t
p
air
s
ar
e
lab
eled
as
n
eg
ativ
e
s
am
p
les.
A
f
u
lly
c
o
n
v
o
lu
ti
o
n
al
n
e
u
r
al
n
etwo
r
k
(
FC
NN)
is
tr
ain
ed
a
s
a
b
in
ar
y
class
if
ier
.
T
h
e
b
es
t
-
p
er
f
o
r
m
in
g
m
o
d
el
is
u
s
ed
t
o
ev
alu
ate
p
air
wis
e
co
m
p
atib
ilit
y
b
ased
o
n
lo
ca
l
v
is
u
al
co
n
ten
t
o
f
ea
c
h
f
r
a
g
m
e
n
t.
T
h
e
r
esu
ltin
g
m
atc
h
in
g
s
c
o
r
es
ar
e
s
to
r
ed
in
a
m
atr
ix
,
wh
ich
is
s
u
b
s
eq
u
en
tly
u
s
ed
as
in
p
u
t
f
o
r
a
g
r
ap
h
-
b
ased
o
p
tim
izatio
n
alg
o
r
ith
m
to
d
eter
m
in
e
th
e
o
p
tim
al
r
ea
s
s
em
b
ly
s
eq
u
en
ce
.
T
h
e
m
o
d
el
is
v
alid
ated
u
s
in
g
r
ea
l
s
h
r
ed
d
ed
d
o
cu
m
e
n
ts
.
T
h
e
f
o
llo
win
g
s
ec
tio
n
s
d
etail
th
e
k
ey
s
tep
s
o
f
th
e
p
r
o
p
o
s
ed
s
y
s
tem
,
in
clu
d
in
g
d
o
cu
m
e
n
t
d
ig
itizatio
n
,
tr
ai
n
in
g
s
am
p
le
g
en
e
r
atio
n
,
s
elf
-
s
u
p
er
v
is
ed
t
r
ain
in
g
,
p
air
wis
e
co
m
p
atib
ilit
y
s
co
r
in
g
,
an
d
o
p
tim
izatio
n
-
b
ased
r
e
ass
em
b
ly
.
T
h
e
o
v
er
all
wo
r
k
f
lo
w
is
illu
s
tr
ated
in
Fig
u
r
e
2
.
2
.
1
.
Do
cu
m
ent
dig
it
iza
t
io
n
C
o
m
m
er
cial
s
h
r
ed
d
er
s
ty
p
ical
ly
p
r
o
d
u
ce
eith
er
s
tr
ip
-
cu
t
(
s
p
ag
h
etti
-
lik
e)
o
r
c
r
o
s
s
-
cu
t
f
r
ag
m
en
ts
.
I
n
th
is
s
tu
d
y
,
we
f
o
cu
s
o
n
th
e
r
ec
o
n
s
tr
u
ctio
n
o
f
s
tr
ip
-
s
h
r
ed
d
e
d
d
o
cu
m
e
n
ts
,
wh
ich
r
ep
r
esen
t
th
e
m
o
s
t
co
m
m
o
n
s
h
r
ed
d
in
g
m
ec
h
an
is
m
u
s
ed
in
p
r
ac
tical
f
o
r
en
s
ic
an
d
a
r
ch
iv
al
s
ce
n
ar
io
s
.
T
h
e
d
ig
itizatio
n
p
r
o
ce
s
s
s
er
v
es
as
th
e
en
tr
y
p
o
in
t
o
f
th
e
p
r
o
p
o
s
ed
r
ec
o
n
s
tr
u
ctio
n
p
ip
elin
e
an
d
is
d
esig
n
ed
to
s
u
p
p
o
r
t
a
u
to
m
ate
d
,
h
ig
h
-
th
r
o
u
g
h
p
u
t
p
r
o
ce
s
s
in
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
2
5
8
6
I
AE
S
I
n
t
J
R
o
b
&
A
u
to
m
,
Vo
l
.
1
5
,
No
.
1
,
Ma
r
ch
20
2
6
:
1
07
-
1
21
110
Fig
u
r
e
2
.
Sy
s
tem
wo
r
k
f
lo
w
f
o
r
s
h
r
ed
d
e
d
d
o
c
u
m
en
t
r
ec
o
n
s
tr
u
ctio
n
Prin
ted
d
o
c
u
m
en
ts
ar
e
f
ir
s
t
m
ec
h
an
ically
s
h
r
ed
d
ed
,
an
d
v
is
u
ally
b
lan
k
f
r
a
g
m
en
ts
ar
e
d
is
ca
r
d
ed
.
T
h
e
r
em
ain
in
g
f
r
ag
m
e
n
ts
ar
e
m
o
u
n
ted
o
n
a
h
ig
h
-
s
atu
r
atio
n
,
n
o
n
-
g
r
ay
s
ca
le
b
ac
k
g
r
o
u
n
d
t
o
f
ac
ilit
ate
r
o
b
u
s
t
f
o
r
eg
r
o
u
n
d
–
b
ac
k
g
r
o
u
n
d
s
ep
ar
atio
n
.
T
h
is
d
esig
n
c
h
o
ice
e
n
ab
les
r
eliab
le
au
to
m
ate
d
p
r
o
ce
s
s
in
g
in
d
o
w
n
s
tr
ea
m
s
tag
es with
o
u
t r
eq
u
ir
in
g
m
an
u
al
an
n
o
tatio
n
o
r
in
ter
v
en
tio
n
.
Fra
g
m
en
t
ex
tr
ac
tio
n
is
p
er
f
o
r
m
ed
u
s
in
g
k
-
m
ea
n
s
clu
s
ter
in
g
in
th
e
R
GB
co
lo
r
s
p
ac
e,
wh
er
e
im
ag
e
p
ix
els
ar
e
g
r
o
u
p
ed
in
t
o
th
r
ee
class
es
co
r
r
esp
o
n
d
in
g
to
f
r
a
g
m
en
t
co
n
te
n
t,
p
ap
e
r
s
u
b
s
tr
ate,
an
d
b
ac
k
g
r
o
u
n
d
.
Af
ter
id
en
tify
in
g
th
e
b
ac
k
g
r
o
u
n
d
clu
s
ter
,
all
ass
o
ciate
d
p
ix
el
s
ar
e
r
em
o
v
ed
to
o
b
tain
clea
n
f
r
ag
m
en
t
c
o
n
to
u
r
s
.
E
ac
h
f
r
ag
m
en
t
is
th
en
is
o
lated
an
d
s
to
r
e
d
as
an
in
d
i
v
id
u
al
im
ag
e.
B
in
ar
izatio
n
is
s
u
b
s
eq
u
en
tly
ap
p
lie
d
u
s
in
g
th
e
Sau
v
o
la
m
eth
o
d
[
4
1
]
.
Sin
ce
f
r
a
g
m
en
ts
ar
e
p
lace
d
with
co
n
s
is
ten
t
o
r
ien
tatio
n
d
u
r
i
n
g
ac
q
u
is
itio
n
,
n
o
r
o
tatio
n
co
r
r
ec
tio
n
is
r
eq
u
ir
ed
,
th
er
eb
y
r
ed
u
cin
g
g
e
o
m
etr
ic
d
is
to
r
tio
n
an
d
p
r
eser
v
in
g
e
d
g
e
i
n
f
o
r
m
atio
n
cr
itical
f
o
r
c
o
m
p
atib
ilit
y
esti
m
atio
n
.
All
ex
tr
ac
ted
f
r
ag
m
e
n
ts
ar
e
s
tan
d
ar
d
ized
an
d
f
o
r
war
d
e
d
t
o
th
e
lea
r
n
in
g
-
b
ased
r
ec
o
n
s
tr
u
ctio
n
s
tag
e.
2
.
1
.
1
.
I
nte
g
ra
t
io
n wit
h
a
uto
m
a
t
ed
s
ca
nn
ing
s
y
s
t
em
s
T
h
e
d
o
cu
m
e
n
t
d
i
g
itizatio
n
p
r
o
ce
d
u
r
e
is
in
h
e
r
en
tly
co
m
p
ati
b
le
with
a
u
to
m
ated
s
ca
n
n
in
g
an
d
r
o
b
o
tic
d
o
cu
m
e
n
t
-
p
r
o
ce
s
s
in
g
p
ip
elin
e
s
.
Ov
er
h
ea
d
s
ca
n
n
er
s
,
s
u
ch
a
s
th
e
Scan
Sn
ap
SV6
0
0
em
p
l
o
y
ed
in
th
is
s
tu
d
y
,
p
r
o
v
id
e
n
o
n
-
c
o
n
tact
im
a
g
e
ac
q
u
is
itio
n
an
d
s
tab
le
im
a
g
in
g
g
eo
m
etr
y
,
m
ak
in
g
t
h
em
well
s
u
ited
f
o
r
in
teg
r
atio
n
in
to
au
to
m
ated
f
o
r
e
n
s
ic
o
r
ar
c
h
iv
al
s
y
s
tem
s
.
I
n
an
o
p
er
atio
n
al
s
ettin
g
,
s
h
r
ed
d
e
d
f
r
ag
m
e
n
ts
ca
n
b
e
s
eq
u
en
tially
s
ca
n
n
ed
with
o
u
t
m
an
u
al
alig
n
m
en
t,
af
te
r
wh
ich
all
s
u
b
s
eq
u
en
t
s
tag
es
—
in
clu
d
i
n
g
b
ac
k
g
r
o
u
n
d
r
em
o
v
al,
c
o
n
to
u
r
ex
tr
ac
tio
n
,
b
in
ar
izatio
n
,
f
r
ag
m
e
n
t
co
m
p
at
ib
ilit
y
esti
m
atio
n
,
an
d
o
p
tim
iz
atio
n
-
b
ased
r
ea
s
s
em
b
ly
—
ar
e
ex
ec
u
ted
e
n
tire
ly
in
s
o
f
twar
e
o
n
a
p
r
o
ce
s
s
in
g
u
n
it
d
ir
ec
tly
co
n
n
ec
ted
to
th
e
s
ca
n
n
er
.
T
h
is
en
d
-
to
-
en
d
au
t
o
m
atio
n
estab
lis
h
es
a
s
ea
m
less
d
ata
f
lo
w
f
r
o
m
p
h
y
s
ical
f
r
ag
m
en
t
ac
q
u
is
itio
n
to
d
ig
ital
r
ec
o
n
s
tr
u
ctio
n
,
p
o
s
itio
n
in
g
th
e
p
r
o
p
o
s
e
d
f
r
am
ewo
r
k
as
a
co
r
e
co
m
p
u
tatio
n
al
co
m
p
o
n
en
t
with
i
n
au
to
m
ated
d
o
c
u
m
en
t
r
ec
o
v
er
y
an
d
r
o
b
o
tic
in
f
o
r
m
atio
n
-
p
r
o
ce
s
s
in
g
s
y
s
tem
s
.
2
.
2
.
T
ra
ini
ng
s
a
m
ple pre
pa
ra
t
io
n
T
o
co
n
s
tr
u
ct
th
e
tr
ain
in
g
d
ataset,
we
p
r
ep
ar
ed
an
in
ter
n
ally
g
en
er
ated
co
r
p
u
s
co
n
s
is
tin
g
o
f
3
0
0
p
a
g
es
o
f
Simp
lifie
d
C
h
in
ese
d
o
c
u
m
e
n
ts
an
d
1
0
0
p
a
g
es
o
f
E
n
g
lis
h
d
o
cu
m
e
n
ts
,
all
d
ig
itally
cr
ea
te
d
in
s
tan
d
ar
d
o
f
f
ice
f
o
r
m
ats.
All
d
o
cu
m
e
n
ts
wer
e
co
n
v
er
ted
i
n
to
3
0
0
-
d
p
i
g
r
ay
s
ca
le
im
ag
es
p
r
io
r
to
p
r
e
p
r
o
ce
s
s
in
g
.
T
o
g
en
er
ate
s
u
f
f
icien
t
tr
ain
in
g
d
ata
f
o
r
th
e
b
o
u
n
d
ar
y
class
if
ier
,
ea
ch
d
o
cu
m
en
t
was
d
ig
itally
s
h
r
ed
d
ed
,
an
d
5
0
%
o
f
th
e
d
o
cu
m
e
n
ts
wer
e
allo
ca
ted
to
tr
ain
in
g
wh
ile
th
e
r
em
ain
i
n
g
5
0
%
wer
e
r
eser
v
ed
f
o
r
v
alid
atio
n
,
en
s
u
r
in
g
th
at
n
o
d
o
cu
m
e
n
t c
o
n
tr
ib
u
ted
s
am
p
les
to
b
o
th
s
ets.
E
ac
h
d
o
cu
m
e
n
t
im
ag
e
was
f
ir
s
t
b
in
ar
ized
u
s
in
g
th
e
Sau
v
o
l
a
m
eth
o
d
[
4
1
]
an
d
th
e
n
p
ar
titi
o
n
ed
in
to
3
0
v
er
tical
s
tr
ip
s
o
f
eq
u
al
w
id
th
,
ea
ch
p
r
eser
v
in
g
th
e
f
u
ll
h
eig
h
t
o
f
th
e
o
r
ig
in
al
im
a
g
e.
T
o
s
im
u
late
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
AE
S
I
n
t
J
R
o
b
&
A
u
to
m
I
SS
N:
2722
-
2
5
8
6
A
u
to
n
o
m
o
u
s
r
ec
o
n
s
tr
u
ctio
n
o
f str
ip
-
s
h
r
ed
d
ed
d
o
cu
men
ts
vi
a
s
elf
-
s
u
p
ervis
ed
d
ee
p
lea
r
n
in
g
…
(
Yi
-
C
h
a
n
g
Wu
)
111
ir
r
eg
u
lar
ities
co
m
m
o
n
ly
o
b
s
er
v
ed
in
p
h
y
s
ical
s
h
r
ed
d
e
d
ed
g
es
an
d
av
o
id
o
v
er
ly
s
m
o
o
th
s
tr
ip
b
o
u
n
d
ar
ies,
th
e
o
u
ter
m
o
s
t
two
-
p
ix
el
c
o
lu
m
n
s
o
n
b
o
th
s
id
es
o
f
ea
c
h
s
tr
ip
we
r
e
r
ep
lace
d
with
p
s
eu
d
o
-
r
an
d
o
m
b
lack
-
a
n
d
-
wh
ite
p
atter
n
s
d
r
awn
f
r
o
m
a
u
n
if
o
r
m
b
in
ar
y
d
is
tr
ib
u
tio
n
U(
0
,
1
)
.
B
ef
o
r
e
s
am
p
le
ex
tr
ac
tio
n
,
th
e
s
im
u
lated
s
tr
ip
s
f
r
o
m
ea
ch
d
o
cu
m
en
t
wer
e
r
an
d
o
m
ly
s
h
u
f
f
led
to
m
in
im
ize
s
am
p
lin
g
b
ias
an
d
in
cr
ea
s
e
co
v
er
ag
e
ac
r
o
s
s
d
if
f
er
en
t
tex
tu
al
r
eg
io
n
s
.
T
r
ain
in
g
s
am
p
les
wer
e
ex
tr
ac
ted
alo
n
g
t
h
e
b
o
u
n
d
ar
y
b
etwe
en
ea
ch
p
air
o
f
a
d
jace
n
t
s
tr
ip
s
.
Sp
ec
if
ically
,
a
3
2
×3
2
p
atch
was
cr
o
p
p
ed
ev
er
y
two
p
ix
els,
co
n
s
is
tin
g
o
f
1
6
p
ix
els
f
r
o
m
ea
c
h
s
id
e
o
f
th
e
b
o
u
n
d
ar
y
.
Patch
es
g
en
er
ated
f
r
o
m
co
r
r
ec
tly
m
atch
ed
s
tr
ip
p
air
s
wer
e
lab
eled
as
p
o
s
itiv
e,
wh
er
ea
s
th
o
s
e
f
r
o
m
m
is
m
atch
ed
p
air
s
wer
e
l
ab
eled
as
n
eg
ativ
e.
An
eq
u
al
n
u
m
b
er
o
f
p
o
s
itiv
e
a
n
d
n
eg
ativ
e
s
am
p
les we
r
e
co
ll
ec
ted
p
er
d
o
cu
m
e
n
t.
Acr
o
s
s
th
e
4
0
0
-
p
ag
e
co
r
p
u
s
,
t
h
is
p
r
o
ce
s
s
p
r
o
d
u
ce
d
ap
p
r
o
x
im
ately
1
4
m
illi
o
n
ca
n
d
id
ate
p
atch
es.
T
o
en
s
u
r
e
th
at
o
n
l
y
in
f
o
r
m
ativ
e
b
o
u
n
d
a
r
y
r
e
g
io
n
s
wer
e
r
etain
e
d
,
s
am
p
les with
a
f
o
r
eg
r
o
u
n
d
-
p
i
x
el
r
atio
b
elo
w
0
.
1
wer
e
d
is
ca
r
d
ed
,
as
s
u
ch
p
atc
h
es
ty
p
ically
c
o
r
r
esp
o
n
d
to
b
lan
k
m
ar
g
in
s
o
r
s
ca
n
n
in
g
n
o
is
e.
T
h
e
r
esu
ltin
g
d
ataset
—
s
to
r
ed
in
b
in
ar
y
im
a
g
e
f
o
r
m
at
a
n
d
ass
u
m
in
g
b
lack
tex
t o
n
a
wh
ite
b
ac
k
g
r
o
u
n
d
—
co
n
tain
ed
r
o
u
g
h
ly
6
to
7
m
illi
o
n
s
am
p
les ea
ch
f
o
r
t
h
e
tr
ain
in
g
a
n
d
v
alid
atio
n
s
ets.
2
.
3
.
F
e
a
t
ure
ex
t
r
a
ct
io
n ba
c
kb
o
ne
a
nd
co
m
pu
t
a
t
io
na
l r
a
t
i
o
na
le
W
e
s
elec
ted
Sq
u
ee
ze
Net
v
1
.
1
[
4
2
]
,
p
r
etr
ain
e
d
o
n
I
m
ag
eNe
t,
as
th
e
b
ac
k
b
o
n
e
f
ea
tu
r
e
ex
t
r
ac
to
r
f
o
r
f
r
ag
m
en
t
co
m
p
atib
ilit
y
p
r
ed
i
ctio
n
.
Sq
u
ee
ze
Net
v
1
.
1
is
a
f
u
lly
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
e
two
r
k
d
esig
n
e
d
to
ac
h
iev
e
co
m
p
etitiv
e
r
ec
o
g
n
iti
o
n
p
er
f
o
r
m
an
ce
u
n
d
e
r
a
r
estricte
d
p
ar
am
eter
an
d
m
em
o
r
y
b
u
d
g
et.
A
p
r
im
a
r
y
d
esig
n
co
n
s
id
er
atio
n
in
o
u
r
f
r
am
ewo
r
k
is
th
e
b
alan
ce
b
etwe
en
r
ep
r
esen
tatio
n
al
ca
p
ac
ity
an
d
o
p
er
atio
n
al
th
r
o
u
g
h
p
u
t.
W
h
ile
h
ig
h
-
ca
p
a
city
ar
ch
itectu
r
es
s
u
c
h
as
R
e
s
Net
o
r
E
f
f
icien
tNet
o
f
f
e
r
in
cr
ea
s
ed
d
ep
t
h
an
d
im
p
r
o
v
e
d
p
er
f
o
r
m
a
n
ce
o
n
g
lo
b
al
s
em
an
tic
b
en
ch
m
ar
k
s
,
th
ei
r
co
m
p
u
tatio
n
al
o
v
er
h
ea
d
is
o
f
ten
p
r
o
h
ib
itiv
e
f
o
r
h
ig
h
-
th
r
o
u
g
h
p
u
t
r
o
b
o
tic
r
ea
s
s
em
b
ly
p
ip
elin
es.
I
n
th
e
p
r
esen
t
task
,
f
r
ag
m
e
n
t
co
m
p
atib
ilit
y
is
d
eter
m
in
ed
p
r
im
ar
ily
b
y
l
o
ca
l
s
tr
o
k
e
co
n
tin
u
ity
an
d
ed
g
e
-
g
e
o
m
etr
y
alig
n
m
en
t
r
at
h
er
th
an
h
ig
h
-
lev
el
s
em
an
tic
ab
s
tr
ac
tio
n
.
C
o
n
s
eq
u
e
n
tly
,
t
h
e
ef
f
ec
tiv
e
f
ea
tu
r
e
s
p
ac
e
is
s
ig
n
if
ican
tly
less
co
m
p
lex
th
a
n
t
h
at
r
e
q
u
ir
ed
f
o
r
g
en
er
al
o
b
ject
r
ec
o
g
n
itio
n
.
Dee
p
er
ar
c
h
itectu
r
es
th
er
ef
o
r
e
p
r
o
v
id
e
d
im
in
is
h
in
g
r
etu
r
n
s
f
o
r
b
o
u
n
d
ar
y
m
atc
h
in
g
wh
ile
in
tr
o
d
u
cin
g
s
u
b
s
tan
tial
laten
cy
an
d
m
em
o
r
y
o
v
er
h
ea
d
.
Fu
r
th
er
m
o
r
e,
p
r
ac
tical
d
o
cu
m
e
n
t
r
ea
s
s
em
b
ly
r
eq
u
ir
es
ev
alu
atin
g
th
o
u
s
an
d
s
o
f
ca
n
d
id
ate
f
r
ag
m
en
t
p
air
i
n
g
s
,
r
esu
ltin
g
in
an
in
h
er
en
t
O(
n
2
)
co
m
p
u
tatio
n
al
co
m
p
lex
ity
.
Un
d
e
r
th
is
co
n
s
tr
ain
t,
a
lig
h
tweig
h
t b
ac
k
b
o
n
e
is
ess
en
tial to
p
r
ev
en
t t
h
e
v
is
io
n
m
o
d
u
l
e
f
r
o
m
d
o
m
in
atin
g
s
y
s
tem
laten
cy
.
Acc
o
r
d
in
g
ly
,
Sq
u
ee
ze
Net
v
1
.
1
is
s
p
ec
if
ically
tailo
r
ed
to
th
ese
r
eq
u
ir
em
e
n
ts
,
o
f
f
er
i
n
g
r
ep
r
esen
tatio
n
al
ca
p
ab
ilit
y
c
o
m
p
ar
ab
le
to
lar
g
er
d
ee
p
n
etw
o
r
k
s
wh
ile
u
tili
zin
g
a
p
p
r
o
x
im
ately
5
0
tim
es
f
ewe
r
p
ar
am
eter
s
.
T
h
is
r
ed
u
ctio
n
in
m
o
d
el
co
m
p
lex
it
y
is
in
s
tr
u
m
en
tal
f
o
r
d
ep
lo
y
m
en
t
in
r
eso
u
r
ce
-
co
n
s
tr
ain
ed
o
r
laten
cy
-
s
en
s
itiv
e
en
v
ir
o
n
m
e
n
ts
,
en
s
u
r
in
g
th
at
th
e
r
ec
o
n
s
tr
u
ctio
n
p
ip
elin
e
r
em
ain
s
v
i
ab
le
f
o
r
r
ea
l
-
tim
e
au
to
m
ated
f
o
r
e
n
s
ic
wo
r
k
f
lo
w
s
.
As
illu
s
tr
ated
in
Fig
u
r
e
3
,
th
e
n
etwo
r
k
b
eg
in
s
with
an
i
n
itial
co
n
v
o
lu
tio
n
al
lay
er
(
co
n
v
1
)
,
f
o
llo
wed
b
y
ei
g
h
t
Fire
m
o
d
u
les
(
Fire
2
–
Fire
9
)
in
ter
leav
ed
with
th
r
ee
m
a
x
-
p
o
o
lin
g
lay
er
s
,
an
d
ter
m
in
ates
with
a
g
lo
b
al
a
v
e
r
ag
e
p
o
o
lin
g
lay
er
.
T
h
e
in
ter
n
al
s
tr
u
ctu
r
e
o
f
ea
ch
Fire
m
o
d
u
le
is
s
h
o
w
n
in
Fig
u
r
e
4
an
d
co
n
s
is
ts
o
f
a
s
q
u
ee
ze
lay
er
with
1
×1
f
ilter
s
an
d
an
ex
p
a
n
d
lay
e
r
co
m
b
in
in
g
p
ar
allel
1
×1
an
d
3
×3
f
ilter
s
.
T
h
is
ar
ch
itectu
r
al
d
esig
n
m
in
im
izes
p
ar
am
eter
c
o
u
n
t w
h
ile
p
r
eser
v
in
g
th
e
f
in
e
-
g
r
ai
n
ed
tex
tu
al
f
ea
t
u
r
es
r
eq
u
ir
ed
f
o
r
ac
cu
r
ate
f
r
a
g
m
e
n
t
alig
n
m
en
t.
I
n
th
is
s
tu
d
y
,
th
e
o
r
ig
in
al
Sq
u
ee
ze
Net
v
1
.
1
co
n
f
ig
u
r
atio
n
is
r
etain
ed
with
o
u
t
s
tr
u
ctu
r
al
m
o
d
if
icatio
n
to
en
s
u
r
e
r
ep
r
o
d
u
ci
b
ilit
y
an
d
c
o
m
p
atib
ilit
y
with
r
ea
l
-
tim
e
au
to
m
ated
d
o
cu
m
e
n
t r
ec
o
n
s
tr
u
ctio
n
wo
r
k
f
lo
ws.
Fig
u
r
e
3
.
T
h
e
ar
ch
itectu
r
e
o
f
S
q
u
ee
ze
Net
1
.
1
2
.
4
.
Self
-
s
up
er
v
is
ed
t
ra
ini
ng
a
nd
pa
irwise c
o
m
pa
t
ibi
lity
s
co
ring
T
o
a
d
a
p
t
th
e
I
m
a
g
eN
et
-
p
r
e
tr
ain
e
d
b
ac
k
b
o
n
e
t
o
th
e
d
o
c
u
m
e
n
t
r
e
c
o
n
s
tr
u
c
ti
o
n
tas
k
,
e
a
ch
b
i
n
ar
y
f
r
a
g
m
en
t
-
p
a
ir
i
m
a
g
e
is
r
ep
lic
ate
d
a
cr
o
s
s
t
h
r
ee
ch
an
n
els
t
o
f
o
r
m
a
f
ix
ed
-
s
i
ze
2
2
7
×2
2
7
×
3
i
n
p
u
t
.
T
h
e
f
i
n
a
l
co
n
v
o
lu
ti
o
n
al
l
ay
e
r
is
r
ep
lac
e
d
wit
h
a
t
wo
-
f
il
te
r
o
u
t
p
u
t
c
o
r
r
esp
o
n
d
i
n
g
to
b
i
n
a
r
y
cl
ass
i
f
i
ca
ti
o
n
(
c
o
m
p
a
ti
b
le
v
e
r
s
u
s
in
c
o
m
p
a
ti
b
le
)
,
wit
h
wei
g
h
ts
i
n
iti
ali
ze
d
f
r
o
m
a
z
er
o
-
m
ea
n
Ga
u
s
s
i
an
d
is
t
r
i
b
u
ti
o
n
wit
h
a
s
t
an
d
ar
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
2
5
8
6
I
AE
S
I
n
t
J
R
o
b
&
A
u
to
m
,
Vo
l
.
1
5
,
No
.
1
,
Ma
r
ch
20
2
6
:
1
07
-
1
21
112
d
e
v
i
ati
o
n
o
f
0
.
0
1
.
T
r
ai
n
i
n
g
is
p
e
r
f
o
r
m
e
d
f
o
r
1
0
e
p
o
c
h
s
u
s
in
g
th
e
Ad
am
o
p
tim
iz
er
[
4
3
]
an
d
ca
t
e
g
o
r
i
ca
l
c
r
o
s
s
-
en
t
r
o
p
y
lo
s
s
,
w
it
h
a
m
i
n
i
-
b
at
c
h
s
i
ze
o
f
2
5
6
.
M
o
d
el
p
e
r
f
o
r
m
an
ce
is
e
v
al
u
ate
d
o
n
a
v
a
li
d
at
i
o
n
s
et
at
t
h
e
e
n
d
o
f
ea
c
h
e
p
o
c
h
,
a
n
d
th
e
c
h
ec
k
p
o
i
n
t a
ch
ie
v
i
n
g
t
h
e
h
i
g
h
est
v
al
id
ati
o
n
ac
c
u
r
a
cy
is
s
el
ec
te
d
f
o
r
s
u
b
s
eq
u
e
n
t
i
n
f
e
r
e
n
c
e.
Fig
u
r
e
4
.
T
h
e
ar
ch
itectu
r
e
o
f
f
ir
e
m
o
d
u
le
Du
r
in
g
in
f
er
e
n
ce
,
th
e
tr
ain
ed
n
etwo
r
k
ev
alu
ates
th
e
p
air
wis
e
co
m
p
atib
ilit
y
o
f
n
o
n
-
b
la
n
k
f
r
ag
m
en
ts
F
=
{
f
1
,
f
2
,
…,
f
n
}
.
Fo
r
ea
ch
o
r
d
er
ed
p
air
(
f
p
,
f
q
)
,
a
lik
elih
o
o
d
s
co
r
e
M
pq
is
co
m
p
u
ted
to
esti
m
ate
th
e
p
r
o
b
ab
ilit
y
th
at
f
q
is
th
e
im
m
ed
iate
r
ig
h
t
n
eig
h
b
o
r
o
f
f
p
.
E
ac
h
ev
alu
atio
n
u
s
es
a
ca
lib
r
ated
im
ag
e
o
f
s
iz
e
H
×3
2
,
c
o
m
p
o
s
ed
o
f
th
e
r
ig
h
tm
o
s
t
1
6
p
ix
els
o
f
f
p
an
d
t
h
e
lef
tm
o
s
t
1
6
p
i
x
els
o
f
f
q
.
T
o
co
m
p
e
n
s
ate
f
o
r
v
er
tical
m
is
alig
n
m
en
t
co
m
m
o
n
l
y
in
tr
o
d
u
ce
d
d
u
r
in
g
m
ec
h
an
ical
s
h
r
ed
d
i
n
g
,
a
v
er
ti
ca
l
o
f
f
s
et
p
ar
am
eter
m
=1
0
is
ap
p
lied
,
r
esu
ltin
g
i
n
2
1
ca
n
d
id
ate
ev
al
u
atio
n
s
p
er
f
r
ag
m
en
t
p
air
.
T
h
e
m
ax
im
u
m
p
r
o
b
a
b
ilit
y
am
o
n
g
th
ese
ca
n
d
i
d
ates
is
r
etain
ed
as
th
e
f
in
al
co
m
p
atib
ilit
y
s
co
r
e
in
m
atr
ix
M
.
2
.
5
.
G
l
o
ba
l
o
ptim
iza
t
io
n
v
ia
AT
SP
f
o
rm
ula
t
io
n
T
h
e
o
p
tim
al
f
r
ag
m
e
n
t
s
eq
u
en
ce
is
o
b
tain
ed
b
y
f
o
r
m
u
latin
g
th
e
r
ea
s
s
em
b
ly
task
as
A
s
y
m
m
etr
ic
T
r
av
elin
g
Salesma
n
Pr
o
b
lem
(
AT
SP
)
.
A
d
is
tan
ce
m
atr
ix
N
is
d
er
iv
ed
f
r
o
m
th
e
co
m
p
atib
ilit
y
m
atr
ix
M
,
wh
er
e
N
pq
=
m
ax
(
M
)
−
M
pq
f
o
r
p
≠
q
.
,
an
d
d
iag
o
n
al
elem
en
ts
ar
e
s
et
to
in
f
in
ity
.
T
h
is
f
o
r
m
u
latio
n
d
ef
in
es
a
d
ir
ec
ted
weig
h
ted
g
r
a
p
h
in
w
h
ich
ea
ch
v
er
tex
co
r
r
esp
o
n
d
s
to
a
f
r
a
g
m
en
t.
T
o
f
in
d
th
e
o
p
tim
al
g
lo
b
al
s
eq
u
en
ce
,
th
e
p
r
o
b
lem
is
tr
ea
ted
as
f
in
d
in
g
th
e
s
h
o
r
test
p
ath
th
at
v
is
its
ea
ch
n
o
d
e
ex
ac
tl
y
o
n
ce
.
T
h
is
is
ac
h
iev
ed
b
y
in
tr
o
d
u
ci
n
g
a
v
ir
tu
al
n
o
d
e
c
o
n
n
ec
ted
to
all
f
r
ag
m
en
ts
v
ia
ze
r
o
-
weig
h
t
ed
g
es,
ef
f
ec
tiv
ely
tr
a
n
s
f
o
r
m
in
g
th
e
f
r
a
g
m
en
t
o
r
d
e
r
in
g
task
i
n
to
a
s
tan
d
ar
d
AT
S
P.
T
o
le
v
er
ag
e
th
e
in
d
u
s
tr
y
-
s
tan
d
ar
d
C
o
n
co
r
d
e
T
SP
s
o
lv
er
[
4
4
]
,
th
e
AT
SP
is
f
u
r
th
er
c
o
n
v
er
te
d
in
to
a
s
y
m
m
etr
ic
T
SP
u
s
in
g
th
e
two
-
n
o
d
e
tr
a
n
s
f
o
r
m
atio
n
m
et
h
o
d
[
4
5
]
a
n
d
s
o
lv
e
d
ex
ac
tly
with
th
e
QSOp
t3
lib
r
ar
y
.
T
h
e
r
esu
ltin
g
f
r
ag
m
en
t
o
r
d
er
in
g
is
d
ete
r
m
in
is
tic
an
d
g
lo
b
ally
o
p
tim
ized
,
p
r
o
v
id
i
n
g
a
r
eliab
le
h
ig
h
-
lev
el
ex
ec
u
tio
n
r
e
f
er
en
ce
f
o
r
au
to
m
ated
o
r
r
o
b
o
tic
d
o
cu
m
en
t r
ea
s
s
em
b
ly
s
y
s
tem
s
.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
3
.
1
.
E
x
perim
ent
a
l
da
t
a
s
et
s
a
nd
prepro
ce
s
s
ing
T
o
ev
alu
ate
t
h
e
f
ea
s
ib
ilit
y
an
d
alig
n
m
en
t
ac
c
u
r
ac
y
o
f
th
e
p
r
o
p
o
s
ed
s
h
r
e
d
d
ed
d
o
cu
m
e
n
t
r
ec
o
n
s
tr
u
ctio
n
m
eth
o
d
o
n
Sim
p
lifie
d
C
h
in
ese
tex
ts
a
n
d
to
e
x
p
lo
r
e
its
p
er
f
o
r
m
a
n
ce
o
n
W
ester
n
lan
g
u
a
g
es,
we
co
n
d
u
cte
d
ex
p
er
im
e
n
ts
o
n
tw
o
r
ea
l
-
wo
r
ld
s
h
r
e
d
d
ed
d
atasets
:
th
e
D2
-
m
ec
d
ataset
[
4
6
]
an
d
th
e
C
s
im
d
ataset.
T
h
e
D2
-
m
ec
d
ataset
c
o
n
s
is
ts
o
f
f
r
ag
m
en
ts
f
r
o
m
2
0
E
n
g
lis
h
p
lain
-
tex
t
d
o
c
u
m
en
ts
s
o
u
r
ce
d
f
r
o
m
th
e
I
SR
I
-
T
k
Evaluation Warning : The document was created with Spire.PDF for Python.
I
AE
S
I
n
t
J
R
o
b
&
A
u
to
m
I
SS
N:
2722
-
2
5
8
6
A
u
to
n
o
m
o
u
s
r
ec
o
n
s
tr
u
ctio
n
o
f str
ip
-
s
h
r
ed
d
ed
d
o
cu
men
ts
vi
a
s
elf
-
s
u
p
ervis
ed
d
ee
p
lea
r
n
in
g
…
(
Yi
-
C
h
a
n
g
Wu
)
113
OC
R
d
atab
ase
[
4
7
]
,
wh
ich
wer
e
s
h
r
ed
d
ed
u
s
in
g
a
L
ea
d
e
r
s
h
ip
m
o
d
el
7
3
4
8
s
h
r
e
d
d
er
.
T
h
e
f
r
ag
m
en
ts
wer
e
m
an
u
ally
r
ea
s
s
em
b
led
,
s
ca
n
n
e
d
at
a
r
eso
lu
tio
n
o
f
3
0
0
d
p
i,
a
n
d
d
ig
itized
in
to
im
ag
e
f
o
r
m
at.
T
h
e
C
s
im
d
ataset
was
s
p
ec
if
ically
co
n
s
tr
u
cted
to
ad
d
r
ess
th
e
lack
o
f
p
u
b
licly
av
aila
b
le
d
atasets
in
v
o
lv
in
g
th
e
d
o
cu
m
e
n
t
ty
p
e
s
tar
g
eted
in
th
is
s
tu
d
y
.
An
ex
am
p
le
is
s
h
o
wn
in
Fig
u
r
e
5
,
d
o
c
u
m
en
ts
wer
e
g
en
er
ated
in
Mic
r
o
s
o
f
t
W
o
r
d
u
s
in
g
A4
p
ap
er
,
SimSu
n
f
o
n
t,
lef
t
-
to
-
r
ig
h
t
h
o
r
iz
o
n
tal
tex
t
lay
o
u
t,
s
in
g
le
-
lin
e
s
p
ac
in
g
,
an
d
a
f
o
n
t
s
ize
o
f
1
2
p
t.
Af
ter
s
h
r
ed
d
in
g
with
an
I
DE
AL
2
2
6
0
s
tr
ip
-
cu
t
s
h
r
ed
d
e
r
(
4
m
m
wid
th
)
,
n
o
n
-
in
f
o
r
m
ativ
e
b
lan
k
f
r
ag
m
e
n
ts
wer
e
r
em
o
v
ed
.
T
h
e
f
r
a
g
m
en
t
ed
g
es
we
r
e
ir
r
eg
u
lar
an
d
o
c
ca
s
io
n
ally
ex
h
i
b
ited
m
in
o
r
i
n
f
o
r
m
atio
n
l
o
s
s
,
r
ef
lec
tin
g
th
e
p
h
y
s
ical
ch
ar
ac
ter
is
tics
o
f
r
ea
l
s
h
r
ed
d
ed
d
o
cu
m
en
t
s
,
in
co
n
tr
ast
to
th
e
u
n
if
o
r
m
ity
s
ee
n
in
d
ig
itally
s
im
u
lated
f
r
ag
m
en
ts
.
T
h
e
r
em
ain
in
g
f
r
ag
m
e
n
ts
—
ap
p
r
o
x
im
ately
3
8
–
4
1
p
er
p
ag
e
—
wer
e
af
f
ix
ed
to
u
n
if
o
r
m
ly
s
atu
r
ated
y
ello
w
A4
s
h
ee
ts
an
d
d
ig
itized
u
s
in
g
a
Scan
Sn
ap
SV6
0
0
o
v
er
h
ea
d
s
ca
n
n
er
at
3
0
0
d
p
i.
Fra
g
m
en
ts
wer
e
th
en
ex
tr
ac
te
d
f
r
o
m
th
e
s
ca
n
n
ed
im
a
g
e
an
d
s
av
ed
as in
d
iv
id
u
al
f
iles
.
Fig
u
r
e
5
.
Sam
p
le
d
o
cu
m
en
t u
s
ed
f
o
r
r
ec
o
n
s
tr
u
ctio
n
Fig
u
r
e
6
illu
s
tr
ates
th
is
p
r
e
p
r
o
ce
s
s
in
g
p
ip
elin
e.
Fig
u
r
e
6
(
a
)
s
h
o
ws
th
e
o
r
ig
in
al
f
r
a
g
m
en
ts
p
asted
o
n
to
y
ello
w
b
ac
k
g
r
o
u
n
d
p
ap
e
r
.
F
ig
u
r
e
6
(
b
)
p
r
esen
ts
1
0
s
am
p
le
f
r
ag
m
e
n
ts
ex
tr
ac
ted
f
r
o
m
6
(
a
)
,
wh
ich
wer
e
b
in
ar
ized
s
u
ch
th
at
th
e
f
r
ag
m
en
ts
ap
p
ea
r
wh
ite,
th
e
b
ac
k
g
r
o
u
n
d
b
lac
k
,
an
d
th
e
ch
ar
ac
te
r
s
tr
o
k
es
in
b
lack
,
en
ab
lin
g
ef
f
ec
tiv
e
f
o
r
eg
r
o
u
n
d
-
b
ac
k
g
r
o
u
n
d
s
ep
ar
atio
n
.
T
ab
le
1
s
u
m
m
ar
izes
th
e
to
tal
n
u
m
b
er
o
f
f
r
a
g
m
en
ts
an
d
th
e
n
u
m
b
er
o
f
co
r
r
ec
tly
ass
em
b
le
d
f
r
a
g
m
en
ts
in
th
e
test
d
o
cu
m
en
ts
f
r
o
m
t
h
e
D2
-
m
ec
an
d
C
s
im
d
atasets
.
Fo
r
th
e
2
0
d
o
cu
m
e
n
ts
in
t
h
e
D2
-
m
ec
d
ataset,
ea
c
h
d
o
cu
m
e
n
t
co
n
tain
s
b
etwe
en
2
0
an
d
2
8
f
r
a
g
m
en
ts
,
with
0
t
o
2
m
is
ass
em
b
led
f
r
ag
m
e
n
ts
p
er
ca
s
e.
T
h
e
C
s
im
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
2
5
8
6
I
AE
S
I
n
t
J
R
o
b
&
A
u
to
m
,
Vo
l
.
1
5
,
No
.
1
,
Ma
r
ch
20
2
6
:
1
07
-
1
21
114
d
ataset
co
n
tain
s
s
u
b
s
tan
tially
m
o
r
e
f
r
ag
m
e
n
ts
(
3
8
–
4
1
p
er
p
ag
e)
,
with
3
–
7
ass
em
b
ly
er
r
o
r
s
p
er
d
o
cu
m
e
n
t.
Fig
u
r
e
7
p
r
esen
ts
ex
am
p
les o
f
r
ec
o
n
s
tr
u
ctio
n
r
esu
lts
,
wh
er
e
Fig
u
r
es 7
(
a
)
an
d
7
(
b
)
ar
e
d
r
aw
n
f
r
o
m
th
e
D2
-
m
ec
an
d
C
s
im
d
atasets
,
r
esp
ec
tiv
e
ly
.
An
e
x
am
in
atio
n
o
f
th
e
r
e
co
n
s
tr
u
cted
o
u
tp
u
ts
r
ev
ea
led
th
at
th
e
f
r
ag
m
en
ts
wer
e
p
lace
d
s
id
e
-
by
-
s
id
e
wit
h
o
u
t
r
o
tatio
n
,
an
d
th
e
v
er
tic
al
alig
n
m
en
t
d
if
f
er
e
n
ce
b
etw
ee
n
f
r
ag
m
e
n
ts
wa
s
m
in
im
ized
.
(
a)
(
b
)
Fig
u
r
e
6
.
Pre
p
r
o
ce
s
s
in
g
r
esu
lt
s
o
f
s
h
r
ed
d
e
d
p
ap
e
r
f
r
a
g
m
en
ts
(
a)
f
r
a
g
m
en
ts
p
asted
o
n
y
ello
w
b
ac
k
g
r
o
u
n
d
p
ap
er
an
d
(
b
)
ten
e
x
tr
ac
ted
an
d
b
in
ar
ized
f
r
ag
m
en
t im
ag
es
T
ab
le
1
.
T
o
tal
n
u
m
b
er
o
f
f
r
a
g
m
en
ts
an
d
n
u
m
b
er
o
f
c
o
r
r
ec
tly
m
atch
ed
p
air
s
in
th
e
D2
-
m
ec
an
d
C
s
im
d
atasets
D
a
t
a
s
e
t
D2
-
mec
C
si
m
N
o
.
t
o
t
a
l
c
o
r
r
e
c
t
t
o
t
a
l
c
o
r
r
e
c
t
1
26
25
40
35
2
28
27
40
34
3
28
27
40
33
4
26
25
40
35
5
24
23
40
34
6
27
26
40
36
7
27
26
41
35
8
25
23
41
37
9
27
26
40
35
10
26
24
40
36
11
25
23
41
35
12
25
24
39
33
13
20
18
41
34
14
25
23
40
34
15
26
25
39
32
16
24
24
39
35
17
25
23
38
34
18
26
25
38
32
19
24
23
40
33
20
23
21
41
35
21
40
36
22
41
38
23
41
35
24
38
34
25
40
33
Evaluation Warning : The document was created with Spire.PDF for Python.
I
AE
S
I
n
t
J
R
o
b
&
A
u
to
m
I
SS
N:
2722
-
2
5
8
6
A
u
to
n
o
m
o
u
s
r
ec
o
n
s
tr
u
ctio
n
o
f str
ip
-
s
h
r
ed
d
ed
d
o
cu
men
ts
vi
a
s
elf
-
s
u
p
ervis
ed
d
ee
p
lea
r
n
in
g
…
(
Yi
-
C
h
a
n
g
Wu
)
115
(
a)
(
b
)
Fig
u
r
e
7
.
R
ec
o
n
s
tr
u
ctio
n
r
esu
l
ts
(
a)
s
am
p
le
f
r
o
m
th
e
D2
-
m
ec
d
ataset
with
9
6
.
4
% a
cc
u
r
ac
y
；
(
b
)
Sam
p
le
f
r
o
m
th
e
C
s
im
d
ataset
with
9
2
.
7
% a
cc
u
r
ac
y
3
.
2
.
Rec
o
ns
t
ruct
io
n a
cc
ura
c
y
a
nd
qu
a
ntit
a
t
iv
e
ev
a
lua
t
io
n
3
.
2
.
1
.
Acc
ura
cy
def
ini
t
io
n
T
o
q
u
a
n
tify
r
ec
o
n
s
tr
u
ctio
n
q
u
ality
,
we
d
ef
in
e
r
ec
o
n
s
tr
u
ctio
n
ac
cu
r
ac
y
as:
A
c
c
ura
c
y
=
1
−
N
umb
e
r
of
in
c
or
r
e
c
tl
y
position
e
d
fr
a
gme
n
ts
To
ta
l
n
umb
e
r
of
fr
a
gme
n
ts
3
.
2
.
2
.
P
rima
ry
re
s
ult
s
Fig
u
r
e
8
illu
s
tr
ates
th
e
r
ec
o
n
s
tr
u
ctio
n
ac
c
u
r
ac
y
ac
h
iev
e
d
o
n
in
d
iv
id
u
al
test
d
o
cu
m
en
ts
.
T
h
e
r
esu
lts
in
d
icate
th
at
th
e
p
er
-
d
o
cu
m
en
t
r
ec
o
n
s
tr
u
ctio
n
ac
c
u
r
ac
y
e
x
ce
ed
s
9
0
.
0
%
f
o
r
E
n
g
lis
h
d
o
cu
m
en
ts
an
d
8
2
.
1
%
f
o
r
C
h
in
ese
d
o
cu
m
en
ts
.
W
h
en
a
v
er
ag
ed
o
n
a
d
o
cu
m
en
t
-
wis
e
b
asis
,
th
e
p
r
o
p
o
s
ed
m
eth
o
d
ac
h
iev
es
an
av
e
r
ag
e
r
ec
o
n
s
tr
u
ctio
n
ac
c
u
r
ac
y
o
f
9
4
.
8
% o
n
th
e
D2
-
m
ec
d
ataset
an
d
8
6
.
5
% o
n
th
e
C
s
im
d
ataset,
r
e
s
p
ec
tiv
ely
.
Fig
u
r
e
8
.
Acc
u
r
ac
y
o
f
r
ec
o
n
s
tr
u
ctio
n
r
esu
lts
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
2
5
8
6
I
AE
S
I
n
t
J
R
o
b
&
A
u
to
m
,
Vo
l
.
1
5
,
No
.
1
,
Ma
r
ch
20
2
6
:
1
07
-
1
21
116
3
.
2
.
3
.
Relia
bil
it
y
a
na
ly
s
is
I
n
ad
d
itio
n
to
r
ec
o
n
s
tr
u
ctio
n
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
wer
e
a
d
o
p
ted
to
f
u
r
t
h
er
ch
ar
ac
ter
ize
th
e
r
eliab
ilit
y
o
f
f
r
ag
m
e
n
t
p
lace
m
e
n
t.
T
h
ese
m
etr
ics
p
r
o
v
id
e
co
m
p
lem
en
ta
r
y
in
s
ig
h
ts
in
to
th
e
co
r
r
ec
tn
ess
an
d
c
o
m
p
leten
ess
o
f
th
e
r
ec
o
n
s
tr
u
ctio
n
p
r
o
ce
s
s
.
I
n
th
e
p
r
o
p
o
s
ed
s
ettin
g
,
ea
ch
f
r
ag
m
en
t
is
ass
ig
n
ed
to
a
u
n
iq
u
e
p
o
s
itio
n
with
o
u
t
th
e
g
en
er
atio
n
o
f
s
p
u
r
io
u
s
o
r
d
u
p
licate
f
r
a
g
m
e
n
ts
.
C
o
n
s
eq
u
en
tly
,
r
ec
o
n
s
tr
u
c
tio
n
er
r
o
r
s
ar
is
e
ex
clu
s
iv
ely
f
r
o
m
m
is
p
lace
m
en
ts
(
s
wap
p
in
g
)
r
ath
er
th
an
in
co
r
r
ec
t
f
r
ag
m
e
n
t
in
s
er
tio
n
s
.
Un
d
e
r
th
is
co
n
s
tr
ain
t,
th
e
s
y
s
tem
a
ch
iev
ed
a
p
r
e
cisi
o
n
o
f
1
.
0
f
o
r
b
o
t
h
th
e
D2
-
m
ec
an
d
C
s
im
d
atasets
,
co
n
f
ir
m
in
g
t
h
at
ev
er
y
p
lace
d
f
r
ag
m
e
n
t
co
r
r
esp
o
n
d
s
to
a
v
alid
d
o
cu
m
e
n
t c
o
m
p
o
n
en
t.
T
ab
le
2
r
e
p
o
r
ts
th
e
av
er
ag
e
d
p
r
ec
is
io
n
,
r
ec
all,
a
n
d
F1
-
s
co
r
e
ac
r
o
s
s
all
test
d
o
cu
m
en
ts
.
T
h
e
r
esu
lts
d
em
o
n
s
tr
ate
h
ig
h
r
ec
o
n
s
tr
u
ct
io
n
r
eliab
ilit
y
,
with
th
e
F1
-
s
co
r
e
r
em
ain
i
n
g
r
o
b
u
s
t
ac
r
o
s
s
b
o
th
E
n
g
lis
h
an
d
C
h
in
ese
s
cr
ip
ts
.
W
h
ile
th
e
r
ec
all
—
wh
ich
is
eq
u
iv
alen
t
to
d
o
cu
m
e
n
t
-
lev
el
r
ec
o
n
s
tr
u
ctio
n
ac
cu
r
ac
y
—
v
ar
ies
b
etwe
en
th
e
two
d
atasets
,
th
e
co
n
s
is
ten
tly
h
ig
h
p
r
ec
is
io
n
a
n
d
F1
-
s
co
r
es
u
n
d
er
s
co
r
e
th
e
s
tab
ilit
y
o
f
th
e
s
elf
-
s
u
p
er
v
is
ed
FC
NN
in
ex
tr
ac
tin
g
d
is
cr
im
in
ativ
e
f
ea
tu
r
es f
r
o
m
p
h
y
s
ical
f
r
ag
m
en
ts
.
T
ab
le
2
.
E
v
alu
atio
n
m
et
r
ics b
ased
o
n
p
er
-
d
o
c
u
m
en
t a
v
er
ag
in
g
D
a
t
a
s
e
t
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
sc
o
r
e
D2
-
mec
1
0
.
9
4
8
0
.
9
7
3
C
si
m
1
0
.
8
6
5
0
.
9
2
8
3
.
3
.
Ana
ly
s
is
o
f
influencing
f
a
ct
o
rs:
s
cr
ipt
s
t
ruct
ure
a
nd
s
t
ro
k
e
co
m
plex
it
y
I
n
m
o
s
t
ca
s
es,
th
e
r
ec
o
n
s
tr
u
ct
io
n
ac
cu
r
ac
y
f
o
r
E
n
g
lis
h
tex
t
was
h
ig
h
er
t
h
an
t
h
at
f
o
r
C
h
in
ese
tex
t.
T
h
is
d
is
cr
ep
an
cy
ca
n
b
e
attr
ib
u
ted
to
s
ev
er
al
f
ac
to
r
s
,
o
n
e
o
f
wh
ich
is
th
e
s
tr
u
ctu
r
al
ch
ar
ac
t
er
is
tics
o
f
C
h
in
ese
ch
ar
ac
ter
s
.
C
h
in
ese
ch
ar
ac
ter
s
ar
e
ty
p
ically
s
q
u
ar
e
-
s
h
ap
ed
,
u
n
if
o
r
m
ly
s
ized
,
a
n
d
ev
e
n
ly
ar
r
an
g
ed
,
wh
ic
h
o
f
ten
in
tr
o
d
u
ce
s
v
e
r
tical
wh
ites
p
ac
e
co
lu
m
n
s
b
etwe
en
ch
ar
ac
te
r
b
lo
ck
s
.
W
h
en
th
e
s
h
r
e
d
d
e
r
cu
ts
alo
n
g
th
ese
wh
ites
p
ac
e
co
lu
m
n
s
—
r
esu
ltin
g
in
ed
g
e
f
r
a
g
m
en
ts
co
n
tain
i
n
g
m
i
n
im
al
in
f
o
r
m
atio
n
—
th
e
s
e
n
o
n
-
i
n
f
o
r
m
ativ
e
ed
g
e
f
r
ag
m
en
ts
m
a
y
b
e
m
is
tak
en
ly
r
ea
s
s
em
b
led
to
g
et
h
er
d
u
r
in
g
r
ec
o
n
s
tr
u
ctio
n
.
I
n
co
n
tr
ast,
ch
ar
ac
ter
s
i
n
E
n
g
lis
h
tex
t
ar
e
g
e
n
er
ally
m
o
r
e
ir
r
eg
u
la
r
ly
ar
r
a
n
g
ed
,
m
ak
in
g
it
less
lik
ely
f
o
r
lar
g
e
b
lan
k
co
lu
m
n
s
to
f
o
r
m
,
an
d
th
u
s
r
e
d
u
cin
g
th
e
ch
a
n
ce
s
o
f
s
u
ch
r
ea
s
s
em
b
ly
e
r
r
o
r
s
.
T
h
e
n
u
m
b
er
o
f
f
r
ag
m
en
ts
m
a
y
also
af
f
ec
t
r
ec
o
n
s
tr
u
ctio
n
ac
cu
r
ac
y
.
W
ith
f
ewe
r
f
r
ag
m
e
n
ts
,
th
er
e
ar
e
f
ewe
r
p
air
in
g
ca
n
d
i
d
ates,
wh
ich
m
ay
r
ed
u
ce
th
e
p
r
o
b
ab
il
ity
o
f
m
is
m
atch
es.
I
n
th
e
D2
-
m
ec
d
ataset,
th
e
n
u
m
b
er
o
f
f
r
ag
m
e
n
ts
p
er
d
o
c
u
m
en
t
was
r
elativ
ely
s
m
all,
wh
er
ea
s
th
e
C
s
im
d
ataset
co
n
tain
ed
s
ig
n
if
ican
tly
m
o
r
e
f
r
ag
m
e
n
ts
p
er
p
a
g
e.
T
o
v
alid
ate
th
is
h
y
p
o
th
esis
th
at
f
ewe
r
f
r
ag
m
en
ts
wo
u
l
d
in
c
r
ea
s
e
r
ec
o
n
s
tr
u
ctio
n
ac
cu
r
ac
y
,
we
c
o
n
d
u
cted
a
n
a
d
d
itio
n
al
ex
p
er
im
en
t
in
wh
ich
s
ev
er
al
C
s
im
d
o
cu
m
e
n
ts
wer
e
m
an
u
all
y
to
r
n
in
to
eig
h
t
ir
r
eg
u
lar
s
tr
ip
-
lik
e
f
r
ag
m
en
ts
.
T
h
e
r
esu
lts
s
h
o
wed
t
h
at
th
e
r
ec
o
n
s
tr
u
ctio
n
ac
c
u
r
a
cy
r
ea
ch
ed
1
0
0
%.
Fig
u
r
e
9
s
h
o
ws
an
e
x
am
p
le
o
f
o
n
e
o
f
th
ese
r
ec
o
n
s
tr
u
ctio
n
s
.
T
h
is
s
u
g
g
ests
th
at
r
ec
o
n
s
tr
u
ctio
n
ac
c
u
r
ac
y
is
in
d
ee
d
in
f
l
u
en
ce
d
b
y
th
e
n
u
m
b
er
o
f
f
r
ag
m
en
ts
.
W
e
also
in
v
esti
g
ated
th
e
in
f
l
u
en
ce
o
f
s
tr
o
k
e
d
e
n
s
ity
an
d
c
h
ar
ac
ter
co
m
p
lex
ity
.
C
h
in
ese
ch
ar
ac
ter
s
ten
d
to
h
av
e
m
o
r
e
c
o
m
p
lex
s
tr
o
k
e
s
tr
u
ct
u
r
es.
W
h
en
s
h
r
ed
d
ed
,
C
h
in
ese
ch
a
r
ac
ter
s
p
r
o
d
u
c
e
a
lar
g
e
n
u
m
b
e
r
o
f
d
is
jo
in
ted
s
tr
o
k
es,
r
esu
ltin
g
in
a
g
r
ea
ter
n
u
m
b
er
o
f
ca
n
d
id
ate
m
atch
in
g
p
o
in
ts
d
u
r
in
g
r
ec
o
n
s
tr
u
ctio
n
.
I
n
co
n
tr
ast,
W
ester
n
ch
ar
ac
ter
s
ar
e
s
tr
u
ctu
r
ally
s
im
p
ler
,
with
f
ewe
r
s
tr
o
k
e
d
is
co
n
tin
u
ities
.
W
e
h
y
p
o
th
esize
th
at
ch
ar
ac
ter
s
ets
with
h
ig
h
er
s
tr
o
k
e
co
m
p
le
x
ity
o
r
s
m
aller
f
o
n
t sizes
m
ay
lead
to
in
cr
ea
s
ed
r
e
co
n
s
tr
u
ctio
n
er
r
o
r
s
,
as
th
e
d
e
n
s
ity
o
f
s
tr
o
k
e
b
r
ea
k
p
o
in
ts
p
er
u
n
it
ar
ea
b
ec
o
m
es
h
ig
h
er
.
T
o
v
alid
ate
th
is
h
y
p
o
t
h
esis
,
we
in
cr
ea
s
ed
th
e
f
o
n
t
s
ize
o
f
C
h
in
ese
ch
ar
ac
ter
s
in
th
e
test
d
o
cu
m
en
ts
an
d
co
n
d
u
cte
d
r
ec
o
n
s
tr
u
ctio
n
ex
p
er
im
en
ts
u
s
in
g
4
mm
-
wid
e
m
ac
h
in
e
-
s
h
r
e
d
d
ed
f
r
ag
m
en
ts
.
T
h
e
r
esu
lts
co
n
f
ir
m
ed
th
at
en
lar
g
in
g
th
e
f
o
n
t size
r
ed
u
ce
d
th
e
d
e
n
s
ity
o
f
s
tr
o
k
e
d
is
co
n
tin
u
ities
p
er
u
n
it
ar
ea
a
n
d
s
ig
n
if
ica
n
tly
i
m
p
r
o
v
e
d
r
ec
o
n
s
tr
u
ctio
n
ac
cu
r
ac
y
.
As
s
h
o
wn
i
n
Fig
u
r
e
1
0
,
wh
en
th
e
f
o
n
t
s
ize
in
th
e
test
ex
am
p
le
(
Fig
u
r
e
5
)
in
c
r
ea
s
ed
f
r
o
m
1
2
to
2
8
,
th
e
r
ec
o
n
s
tr
u
ctio
n
ac
cu
r
ac
y
r
ea
ch
e
d
1
0
0
%.
T
h
ese
f
in
d
i
n
g
s
s
u
g
g
est
th
at
r
ec
o
n
s
t
r
u
ctio
n
ac
cu
r
ac
y
is
also
in
f
lu
e
n
ce
d
b
y
th
e
s
tr
o
k
e
co
m
p
lex
ity
a
n
d
f
o
n
t size
o
f
th
e
s
cr
ip
t.
C
o
m
p
ar
ed
with
s
y
n
th
etic
f
r
ag
m
en
ts
u
s
ed
in
s
im
u
lated
ex
p
er
im
en
ts
,
r
ea
l
s
h
r
ed
d
ed
d
o
cu
m
e
n
ts
in
tr
o
d
u
ce
s
ev
e
r
al
f
ac
to
r
s
th
at
ca
n
d
eg
r
a
d
e
r
ec
o
n
s
tr
u
ctio
n
p
e
r
f
o
r
m
a
n
ce
.
T
h
ese
in
clu
d
e
ed
g
e
d
am
ag
e
f
r
o
m
th
e
s
h
r
ed
d
in
g
p
r
o
ce
s
s
,
an
g
u
lar
d
is
to
r
tio
n
s
b
etwe
en
f
r
ag
m
en
ts
an
d
th
e
o
r
ig
in
al
lay
o
u
t
ca
u
s
ed
b
y
m
ec
h
an
ical
cu
ttin
g
an
d
d
i
g
itizatio
n
,
s
ca
n
n
er
r
eso
lu
tio
n
lim
itatio
n
s
,
im
ag
e
n
o
is
e
in
t
r
o
d
u
ce
d
d
u
r
in
g
s
ca
n
n
in
g
,
v
er
tical
m
is
alig
n
m
en
t
b
etwe
en
ad
jac
en
t
f
r
a
g
m
en
ts
,
a
n
d
co
n
to
u
r
ex
tr
ac
tio
n
n
o
is
e.
I
n
ad
d
itio
n
,
t
h
e
n
u
m
b
e
r
o
f
f
r
ag
m
en
ts
,
th
e
n
atu
r
e
o
f
th
e
s
cr
ip
t,
an
d
f
o
n
t
s
ize
also
p
lay
cr
itical
r
o
les.
Desp
ite
th
ese
ch
allen
g
es,
o
u
r
ex
p
er
im
en
ts
d
em
o
n
s
tr
ate
th
e
f
ea
s
ib
ilit
y
o
f
th
e
p
r
o
p
o
s
ed
r
ec
o
n
s
tr
u
ctio
n
m
eth
o
d
o
n
r
ea
l
s
h
r
ed
d
ed
C
h
in
ese
d
o
cu
m
e
n
ts
.
Evaluation Warning : The document was created with Spire.PDF for Python.