TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 13, No. 2, Februa
ry 20
15, pp. 300 ~ 304
DOI: 10.115
9
1
/telkomni
ka.
v
13i2.687
3
300
Re
cei
v
ed O
c
t
ober 1
9
, 201
4; Revi
se
d Decem
b
e
r
18, 2014; Accept
ed Ja
nua
ry 4,
2015
A Novel Method for Sensing Obscene Videos using
Scene Change Detection
Rash
ed Mus
t
afa*
1,2,3
, Dingju Zhu*
1,2,4,5
1
Labor
ator
y
for
Smart Comput
ing a
nd Inform
ation Sci
enc
es, Shenzh
en Inst
itutes of Advan
c
ed T
e
chnol
og
y,
Chin
ese Aca
d
e
m
y
of Sci
enc
es
2
Universit
y
of Chin
ese Aca
d
e
m
y
of Scie
nces
, Beijin
g, Chin
a
3
Departme
n
t of Computer Sci
ence a
nd En
gi
neer
i
ng, Un
iver
sit
y
of Chitta
go
ng, Bang
la
des
h
4
Shenzh
en Pu
blic Pl
atform for
T
r
iple-pl
a
y
Vi
deo T
r
anscodi
ng Ce
nter
5
School of Co
mputer Scie
nc
e, South Ch
i
n
a
normal U
n
iver
sit
y
Gua
ngzh
o
u
, Chin
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: ras
hed.m@si
at.ac.cn, dj.zh@siat.ac.cn
A
b
st
r
a
ct
Vide
o scen
e
c
han
ge
detectio
n
has
great i
m
portanc
e of
ma
nag
ing
an
d a
n
a
ly
z
i
n
g
l
a
rge
a
m
o
unt o
f
vide
os. T
r
aditi
ona
lly this tec
hni
que
use
d
for ind
e
xi
ng, s
e
g
m
e
n
ting
an
d categ
o
ri
z
i
n
g
different type
s of
vide
os. Very few
w
o
rks addressed to c
l
ass
i
fy obsce
ne us
ing sc
ene c
h
a
nge
detecti
on
meth
od. In th
i
s
researc
h
w
e
p
r
opos
ed
a si
mple
ap
pro
a
ch f
o
r se
nsin
g o
b
j
e
ction
abl
e vi
de
os by
obs
ervin
g
sce
ne c
h
a
n
g
e
s
into
different v
i
deo
ge
nres. Vi
deo
scen
e
s ar
e gro
u
p
ed
into set of key fram
es. A
fte
r
a
nal
y
z
in
g du
ra
ti
on o
f
each sce
ne an
d counti
ng the
nu
mb
er of
key frames of des
i
gnate
d
scen
e
, it has bee
n sh
ow
n that obsce
ne
vide
os h
a
ve i
n
f
r
equ
ent sce
ne
chan
gin
g
n
a
tur
e
. W
h
ile
in s
p
o
r
ts, dramas, music a
nd
actio
n
films
hav
e l
a
rg
e
nu
mb
er of sc
e
ne ch
an
ges. W
e
use
d
six
type
s of vid
eo
ge
nr
es an
d th
e d
e
c
i
sion
h
a
s b
een
made
by s
e
tting
a thresho
l
d b
a
s
ed on extr
acted key fra
m
es.
Experi
m
e
n
ta
l result
show
ed that
t
he accura
cy is 83.33%
a
n
d
false pos
itive r
a
te is 16.6
7
%.
Ke
y
w
ords
:
scene ch
an
ge d
e
t
ection, key frames, co
ntent b
a
sed vi
de
o retrieval, o
b
sce
nity detection
Copy
right
©
2015 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
Video scen
e
cha
nge d
e
tection is very u
s
eful
for vide
o indexing, a
nalysi
s
and
content-
based
retri
e
val of visual
in
formation
[1,
2]. It is a vit
a
l op
eratio
n i
n
all
multime
d
ia a
ppli
c
atio
ns
su
ch a
s
vid
eo on d
e
ma
nd (VO
D
), d
i
gital arch
ive
s
, ne
ws m
e
dia etc [3]. Low-level vid
eo
segm
entation
is the fi
rst
step in all vid
eo
sce
ne
ch
ange
dete
c
tion [3]. A sce
ne comp
ose
d
of
several shots, which
den
o
t
ed as
contin
uou
s fr
ame
s
of one a
c
tio
n
[1], [3-4]. Existing wo
rks
demon
strated
luminan
ce
o
r
color
histo
g
r
am diffe
ren
c
es of
con
s
e
c
utive frame
s
[2]. As we
kn
ow
that luminan
ce is
su
scepti
b
le to sm
all
cha
nge
s a
n
d
hen
ce
can
n
ot give app
ro
priate
re
sults for
scene
ch
ang
e dete
c
tion [
5
, 6]. Scene
cha
nge
dete
c
tion al
so
pe
rforme
d in
co
mpre
ssed vid
eo
data (MPEG
-1) [7, 8]. In that app
roa
c
h
key fram
e
s
can b
e
re
pre
s
ente
d
as bi
nary edg
e m
aps.
After calculati
ng co
rrel
a
tion
between the
edge ma
ps, two con
s
ecutive frames ca
n be com
pared.
Earlier
wo
rks empha
si
zed
low level st
ructures
on video
s and ve
ry few works demon
strate
d
scene
seg
m
entation on
obje
c
tionabl
e
videos [12,
13]. Some p
apers indi
cat
ed obje
c
tion
able
video processing in
pa
rall
el and di
stri
b
u
ted fash
ion
but failed to
address o
b
scenity detectio
n
approp
riately [26, 27]. Online obje
c
tion
a
b
le video
s an
d image
s a
r
e
now e
a
sily a
c
ce
ssi
ble du
e
to
availability of high-speed I
n
ter
net and
rapid growth of
multim
edi
a technology.
A report
shows
that a large n
u
mbe
r
of teens an
d child
ren se
arch
po
rnog
ra
phi
c co
ntents everyd
ay [28]. This is a
threat for the
society and
con
c
e
r
n
s
of Internet sa
fety. Taking ca
re of this issu
e, scienti
s
ts
are
workin
g ha
rd
and initiated
different filter tech
niqu
es
to scree
n
ma
liciou
s
conten
ts. In this pa
per
we p
r
e
s
ente
d
a metho
d
to find ob
scen
e
videos
usi
ng
scene
ch
ang
e dete
c
tion. V
i
deo
scene
s
are
grou
ped
into
set of
key f
r
ame
s
. O
b
se
rving
cha
nge
of scen
es,
we
analyzed
the p
r
e
s
en
ce of
obsce
nity among a large set of obscene
and beni
gn video
s.
The rest of th
is pap
er
ca
n be organi
ze
d
acco
rding to
the followin
g
stru
ctures:
section
2
briefly
d
e
scri
bes
different scene chan
g
e
detec
tion m
e
thod
s and th
eir appli
c
atio
ns, our p
r
op
o
s
ed
method
i
s
d
e
scrib
ed
i
n
section
3,
q
u
antitativ
e re
sults a
r
e
anal
yzed i
n
se
ction 4
and
fin
a
lly
se
ction 5 con
t
ains con
c
lu
si
on and future
work.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Rob
u
st SINS/GNSS Integration Method
for
High
Dyna
m
i
c Applicati
ons (Falin
Wu)
301
2. Scene Ch
ange Dete
cti
on Appro
a
c
h
es
A digital video can
be form
ed by frame
s
whi
c
h a
r
e p
r
e
s
ente
d
as
co
nse
c
utive ma
nner fo
r
viewer’
s
p
e
rception [23]. Key frame d
e
n
o
ted a
s
re
pre
s
entative fra
m
e whi
c
h
co
ntain si
gnificant
conte
n
t of a
shot. Base
d on the conte
n
t complexity
of shots, on
e or more ke
y frames can
be
extracted
fro
m
a
sin
g
le
sh
ot [24]. Shot
denote
d
a
s
continuo
us
fra
m
es
taken by
sin
g
le ca
mera a
s
contin
uou
s a
c
tion of time and sp
ace. Cut or ha
rd
cut are ab
ru
pt transition
s
from one sh
ot to
anothe
r. Soft
transitio
ns a
r
e kno
w
n
as wipes, fa
de
s
a
nd
di
ssolve
s. In
this effect one sh
ot
can be
repla
c
e
d
by
anothe
r. It also
calle
d a
s
grad
ual tran
sitions. Fa
de
are
of two types, fad
e
out
and
fade in. The
first on
e is a g
r
adu
al tra
n
siti
on
bet
wee
n
a
scene
and
a
con
s
tant ima
ge an
d fade i
n
is between a
con
s
tant ima
ge and a
sce
ne [25].
2.1. SAD (Su
m
of Absolu
te
Differen
ce
s) (So
ft cu
t)
It is a simple algorithm where two seq
uent
ial frame
s
are
comp
ared usin
g add
ition of
absolute valu
es of
ea
ch pi
xel. After that subtra
ctio
n
occurs f
r
om
corre
s
p
ondin
g
pixels [9-1
1],
[18]. The re
sult is a po
sitive numb
e
r
whi
c
h is fu
rthe
r u
s
ed a
s
score.
SAD is su
sceptible to min
o
r
scene
chang
es. Th
e fal
s
e
hits o
c
curs
when fa
st
cam
e
ra
moveme
nt or
su
dde
n
light on i
n
a
d
a
rk
scene. It hardly react
s
to soft cuts [1
9]. Yet,
SAD
is use
d
often to produ
ce
a basi
c
set
of
"possible hit
s
" as it detects all visi
ble hard cuts
with utmost proba
bil
i
ty.
2.2. Histogr
a
m
Differ
e
nce
s
(HD) (Ha
r
d cut)
It is similar t
o
Sum of ab
solute diffe
re
nce
s
. It com
putes hi
stog
ram differen
c
e of two
seq
uential vi
deo fra
m
e
s
. Histo
g
ra
m tel
l
s qu
antitative
distri
bution
of colo
rs i
n
a
frame [20]. HD is
less suscepti
b
le to mino
r cha
nge
s of
scene
s
an
d
hen
ce fewer false hits.
HD i
s
compl
e
tely
depe
nd
s
with
histo
g
ra
m
calcul
ation
whi
c
h i
s
it
s m
a
jo
r d
r
a
w
ba
ck. It is b
e
lieved
that two
fram
es
can h
a
ve the
same hi
stog
ram
s
. For ex
ample,
de
sse
r
t and be
ach
pictures
can
have the sa
me
histog
ram th
ough the co
ntents are n
o
t the same
.
For hard cu
t detection this method i
s
no
t
suitabl
e [21].
2.3. Edge Ch
ange Ratio (ECR) (Wipe or dissolv
e)
Edge cha
nge
ratio (ECR)
also comp
ares
co
ntents
of video fra
m
es. It can
have the
cap
ability of tran
sformi
ng f
r
ame
s
into e
dge pi
cture
s
.
Usin
g an im
age processi
ng tool (dilati
on),
ECR compute a probabilit
y finding that following fram
e contai
ns t
he sam
e
objects [13, 22]. It can
detec
t hard
cuts
as
well as
different s
o
ft c
u
ts
. Ho
we
ver, it cann
ot detect
wipe
s as it con
s
ide
r
s
the fading in
object
s
as
regul
ar movi
ng obje
c
ts t
h
rou
gh the scen
e. De
spit
e, ECR can
be
extended ma
nually to reco
gnize sp
eci
a
l forms of soft cuts [23].
2.4. Shot Ch
ange Dete
cti
on based o
n
Sliding Windo
w
Method
(SCDSW)
In video se
g
m
entation, traditional
slidi
ng win
d
o
w
(CSW) ha
s b
een u
s
ed
by many
resea
r
chers f
o
r ad
aptive th
resholdi
ng [1
2], [14-15
]. CSW can dete
c
t hard cut by taking th
e rat
i
os
of present feature value a
nd its local n
e
ighb
or
h
ood.
Howeve
r, it
has a si
gnificant numbe
r of
false al
arm
s
and mi
ssed
cuts. It is
sh
own i
n
[1
6
-
1
7
] that, this
method
can
be imp
r
oved
by
combi
n
ing
wi
th colo
r histo
g
ram diffe
ren
c
e
s
. The imp
r
oved
sliding
windo
w met
hod ha
s thre
e
step
s proce
ssi
ng such as p
r
e
filtering, slidin
g wind
ow
filtering and
scene a
c
tiven
e
ss
investigatio
n of frame by fram
e di
sconti
nuity values. Camera/obj
ect motion
s are mo
re ro
b
u
st
using cut detection
whi
c
h i
s
based
on possibility va
lues [24]. One
of the purposes i
s
to relax
the
threshold
or
para
m
eter se
lection
proble
m
that is to
make
the int
e
rme
d
iate p
a
r
amete
r
s to
be
valid for a va
st ra
nge
of video p
r
o
g
ram
s
an
d to
redu
ce the
influen
ce of th
e final
thre
shol
d on
th
e
whol
e dete
c
tion accu
ra
cy [25].
3. Scene Ch
ange Dete
cti
on for Ob
sc
ene Videos (propos
ed method
)
In
se
ction
2 we de
scribe
d
so
me com
m
on
vide
o sce
ne cha
nge
de
tection metho
d
s (SAD,
HD, ECR a
n
d
SCDS
W). All
method
s dev
oted to det
e
c
t
sud
den
or g
r
adual tran
sitions
of scen
e
s
.
In those wo
rks
we didn’t
find any indication
that how often th
e scene
s are chan
ging.
Our
prop
osed
me
thod i
s
b
a
se
d on
a
gro
u
nd truth
of
freque
nt an
d i
n
frequ
ent n
a
ture
of chan
g
i
ng
scene
s. The
experim
ent carri
ed out on
13 unstructu
red video
s wi
th arbitra
r
y length co
ntaini
ng
obje
c
tionabl
e
and
be
nign
scen
es. At fi
rst
all Key f
r
ame
s
a
r
e
extra
c
te
d u
s
ing
an
o
p
en
sou
r
ce to
ol
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 13, No. 2, Februa
ry 2015 : 300 – 304
302
ffmpeg [17]. Then summ
arize the re
sult for inst
an
ce Ta
ble 1 d
e
mon
s
trate
d
the extracte
d
key
frame
s
, its types a
nd hit or misse
s
statu
s
of different
video gen
re
s.
Table 1. Extracted
key fra
m
es of differe
nt videos
SL/No
Extracte
d
Ke
y
F
r
ames
T
y
pe*
Remark*
1
24 O
H
2
101 O
M
4
10 O
H
5
17 O
H
6
39 O
H
7
48 O
H
8
46 B
H
9
76 B
H
10
65 B
H
11
1 B
M
12
29 B
H
13
20 B
M
* O
Ob
scen
e, B
Benign, H
Hit, M
Miss
It is shown that the numb
e
r of key fra
m
es
for o
b
scene video
s is significantly smalle
r
than beni
gn
videos. Th
e spe
c
ific g
e
n
r
es of be
nign
videos a
r
e d
r
ama, ne
ws, sports, jo
ke
s
and
musi
c video.
4. Results
The p
e
rfo
r
m
ance of ou
r
method h
a
s
been
elu
c
idat
ed in
T
abl
e 2
.
Highe
r tru
e
positive
rate and lo
we
r false po
sitive rate sig
n
ifie
s the strength
of our app
ro
ach.
T
abl
e 2.
Accu
racy
cha
r
t
T
r
ue Positive Rate (
T
PR)
83.33%
False Positive R
a
te (FPR
)
16.67%
False Negative R
a
te (F
NR)
33.33%
True N
egative R
a
te (T
NR)
67%
The follo
wing
figure
(Figu
r
e 1)
sho
w
e
d
scene
ch
an
ge sce
nari
o
of different types of
video gen
re
s. He
re, drama, movie
traile
r, TV
sho
w
and
obscen
e
video
s have
been
demon
strated
for simpli
cit
y
. It has bee
n ob
serve
d
from the
ran
d
o
m sh
ots th
at, most of the
videos chan
ge
scene
s
a
fter few or j
u
st a
se
con
d
du
ration. But
scen
e changi
ng natu
r
e of
obje
c
tionabl
e
videos is
qui
et long. Fo
r i
n
stan
ce
sc
en
e ch
ang
es of
the giv
en vid
eo ge
nres
are 5,
1, 9 and 34 se
con
d
s resp
ectively. It mean
s obs
ce
n
e
videos hav
e longe
st sce
ne duration a
n
d
the shorte
st
scen
e
cha
nge
s in
movie t
r
a
iler. It is
to
b
e
note
d
that
TV sh
ows
ha
ve long
er
sce
ne
duratio
n than
drama
s
. The
reason for th
is is tha
t, sho
t
taking time in TV shows
are lo
wer tha
n
dram
as [25].
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Rob
u
st SINS/GNSS Integration Method
for
High
Dyna
m
i
c Applicati
ons (Falin
Wu)
303
Figure 1. Sce
ne ch
ang
e on
different types of video ge
nre
s
5. Conclusio
n
There is a vast amount of
obje
c
tionabl
e
videos
available in online due to rapid g
r
owth of
Information a
nd co
mmuni
cation tech
nol
ogy
. It is very
dif
f
icult to filter all mali
cio
u
s content
s from
Internet. Researche
r
s are
advocating t
heir ef
fo
rt
s t
o
do
so. In this resea
r
ch
we p
r
op
osed
a
simple m
e
th
od of identifying obsce
n
e
videos u
s
i
ng scen
e ch
ange d
e
tecti
on. It has been
observed tha
t
obsce
ne video
s infreq
u
ently chan
g
e
scen
e
s
whe
t
her in other types of vid
eos
su
ch as acti
on films, dra
m
as, news, movie tr
ailer and
TV sho
w
s have sig
n
i
ficant numbe
r of
scene ch
ang
es [T
able 1].
There is a controvers
ial be
havior on live and edited musi
c videos. Live
musi
c vide
os don’t h
a
ve
enou
gh
scen
e ch
ang
es
which co
ntra
di
ct
with ob
scene
vide
os, but
edited musi
c videos have
significa
nt scen
e
ch
a
nge
s [T
able 1]. Usi
ng the ground truth, we
identified
m
o
re
than 80% videos co
ntai
ning
o
b
sce
n
ity
.
Skin colo
r
and e
r
oto
gen
etic bo
dy pa
rts
detectio
n
ca
n
further ap
plicable for bette
r accu
ra
cy
.
Ackn
o
w
l
e
dg
ements
This
re
sea
r
ch wa
s supp
orted in
part
by
Shenzh
en Te
chni
cal
Proje
c
t (grant no.
HLE20
110
42
2008
2A) an
d
National
Nat
u
ral Sci
e
n
c
e
Found
ation o
f
China (gra
n
t
no. 611051
33)
and Shen
zh
e
n
Public Te
ch
nical Platform
(gra
nt no. CXC201
005
26
0003A
).
Referen
ces
[1]
SC Che
n
, ML Sh
yu, C Z
han
g, RL Ka
sh
yap. Vid
e
o
scene cha
n
ge detecti
on
method us
ing
unsu
pervis
ed segme
n
tatio
n
and ob
ject
tra
cking.
IEEE International
Conferenc
e on M
u
ltim
edia
an
d
Expo (ICME)
. W
a
seda U
n
ive
r
sit
y
. T
o
kyo, Ja
pan. 20
01; 57-
60.
[2]
W
Z
hu, C
T
o
klu, SP Liou. A
u
tomatic ne
w
s
video se
gmen
tation an
d cate
goriz
ation b
a
s
ed on cl
ose
d
captio
ne
d te
xt.
IEEE International Conference on M
u
ltim
edia & Ex
po
. T
o
k
y
o, Jap
an.
200
1; 10
36-
103
9.
[3]
X W
a
n
g
, Z
W
eng. Scene
abrupt ch
an
ge detecti
on.
Cana
dia
n
C
onfere
n
ce o
n
Electrical a
n
d
Co
mp
uting, En
gin
eeri
ng.
20
0
0
; 880–
88
3.
[4]
CO F
ilipe d
e
, S E
w
ert
on, Michae
l Eckman
n, W
a
lt
er J. Scheir
e
r, Ander
son
Roc
ha. Open set sourc
e
camera attrib
ution a
nd d
e
vice
linki
ng.
Pattern Recognition Letters
. 2014; 3
9
: 92–1
01.
[5]
SC Chen, S
Sista, ML Shyu,
RL K
a
sh
ya
p. An In
de
xi
n
g
an
d S
earch
i
ng Structur
e f
o
r Multim
edi
a
Database S
y
st
ems.
IS&T
/SPIE confere
n
ce
o
n
Storag
e
a
nd
Retriev
a
l for M
edi
a Dat
abas
e
s
. 2000;
262-
270.
[6]
H Soon
gi, C B
eobk
eu
n, C Yo
onsik. Ad
aptiv
e T
h
reshold
i
ng
for Scene C
h
ang
e Detecti
o
n
.
IEEE Third
Internatio
na
l C
onfere
n
ce o
n
Cons
u
m
er Ele
c
tronics
- Berli
n
(ICCE-Berli
n)
, 2013.
[7]
N Nav
i
d, VKB
Paul
o, MR Jo
natha
n, VS M
and
ya
m. On the Us
e of Opt
i
cal F
l
o
w
for S
c
ene
Ch
ang
e
Detectio
n an
d Descripti
on.
Jo
urna
l of Intelli
g
ent & Robotic
Systems.
20
13
: 1-30.
[8]
F
Del, Manfre
d, B Laszlo. Stat
e-of-the-art a
nd future ch
all
eng
es
in vi
de
o
scene d
e
tecti
on: a surve
y
,
Multi
m
ed
ia sys
tems.
20
13; 19
(5): 427-4
54.
[9]
MT Dalton, T Amit, KhS
Mang
lem, R S
udi
pta. A Surve
y
on Vi
d
eo Segm
entat
ion.
Intelligent
Co
mp
uting, Ne
tw
orking, and Informatics
, Spr
i
ng
er India. 2
0
14; 903-
91
2.
[10]
Dieg
o
, F
e
rran,
P Danie
l
, S Joan, ML An
toni
o. Vide
o
alig
nment fo
r chang
e det
ection.
Image
Processing, IEEE Transactions on.
20
11; 2
0
(7): 185
8-1
8
6
9
.
[11]
W
ang, Ji
anch
a
o
, L B
i
ng,
H
W
e
iming, W
Ou. Horro
r v
i
d
eo sc
ene
rec
o
gniti
on v
i
a m
u
l
t
iple
insta
n
c
e
lear
nin
g
, In
Ac
oustics, Speec
h and S
i
gnal
Processing (ICASSP). IEEE In
ternational Conference
on
.
201
1; 132
5-13
28.
[12]
Kim, Y C
han
g, OJ K
w
o
n
, C
Seokrim. A
pr
actical
s
y
stem
for d
e
tectin
g
obsce
ne v
i
d
e
o
s
.
C
o
n
s
um
er
Electronics, IEEE Transactions on
. 20
11; 5
7
(2): 646-
65
0.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 13, No. 2, Februa
ry 2015 : 300 – 304
304
[13]
Ulges, A
d
ria
n
, S Christia
n, B
Damia
n
, S Ar
min, Porn
o
g
r
ap
hy
de
te
cti
o
n
in
vi
de
o
be
ne
fi
ts (a
l
o
t) fro
m
a
multi-modal approac
h.
Procee
din
g
s of the 20
12 ACM int
e
rn
at
ion
a
l w
o
rksh
op on A
udi
o an
d multi
m
e
d
i
a
meth
ods for lar
ge-sca
le vid
eo
ana
lysis
. 20
12
21-2
6
.
[14]
da Si
lva El
eut
erio, M Pe
dro
,
Mateus de
Castro
Po
lastr
o
, Brazili
an F
eder
al Po
lice.
An ad
aptiv
e
sampli
ng strat
e
g
y
for autom
atic detec
ti
on of
child porn
o
g
r
aph
ic
vide
os.
Procee
din
g
s of
the Sevent
h
Internatio
na
l C
onfere
n
ce o
n
F
o
rens
ic
Co
mp
u
t
er Science, Br
asili
a, DF
, Bra
z
il
. 201
2.
[15]
Ochoa, Victor
M T
o
rres, YY
Sule, AC F
a
ou
zi. A
dult Vi
de
o
Conte
n
t Detect
ion
Usin
g Mac
h
in
e L
earn
i
n
g
T
e
chniques.
Sign
al, Ima
g
e
Technol
ogy
and Internet
Based Systems (SITIS), IEEE Eighth
Internatio
na
l C
onfere
n
ce.
2
012
.
[16]
El-Hafe
e
z, A T
a
rek. A
Ne
w
E
ffective
Sy
stem fo
r Fi
l
t
e
r
in
g
Po
rn
og
ra
phy
Vi
d
e
o
s
.
Intern
ati
ona
l Jo
urn
a
l
on Co
mputer S
c
ienc
e & Engi
n
eeri
n
g
. 20
10.
[17]
Open sourc
e
multimedia frame
w
ork, http://www
.ffmpeg.or
g. Accessed 12
th
March 2014
[18]
Hamza
h
, A R
o
stam, AR Ro
sman, MN Z
a
rina. Su
m of Absol
u
te
Diffe
rences alg
o
rit
h
m
in
ster
e
o
corresp
ond
enc
e pr
obl
em for
stereo m
a
tchin
g
i
n
com
puter
vision
a
ppl
icati
on.
C
o
mput
er Scienc
e
a
n
d
Information Technology (ICCSIT), 3r
d IEEE
Internationa
l Conference on
. 201
0.
[19]
Adhik
a
ri, Pri
y
a
darsh
ine
e
, G
Neeta, D
J
y
ot
h
i
, BG Hog
a
d
e
, Abrupt sc
ene
chan
ge
detecti
on.
Intelligent
Links
. 20
08: 35
.
[20]
Hua
ng, C
hun
g
-
Lin, L
Bin
g
-Y
ao. A ro
bust s
c
ene-
c
han
ge
d
e
tection
metho
d
for vi
deo
se
gmentati
o
n
,
Circuits and Sy
stem
s for Video Te
chnology. IEEE Transactions on.
2
001; 11(1
2
):
128
1-1
288.
[21]
Patel, Nil
esh V
,
Ish
w
ar K Seth
i. Video sh
ot
d
e
tection a
nd ch
aracteriz
a
tion f
o
r vide
o data
b
a
ses.
Pattern
Reco
gniti
on.
1
997; 30(
4): 583
-592.
[22]
Lie
nhart, R
a
i
n
er W
.
Com
p
a
r
ison
of
aut
o
m
atic sh
ot b
oun
dar
y
detec
tion
alg
o
rithm
s
.
Electron
i
c
Imag
in
g'
99,
Internati
ona
l Soci
et
y
for Optics a
nd Photo
n
ics,
199
8.
[23]
Jacobs A, et a
l
. Automatic
s
hot bo
un
dar
y
detectio
n
com
b
ini
ng c
o
lor, e
dge, a
nd moti
on featur
es o
f
adj
acent frame
s
.
T
R
ECVID 2004 W
o
rksh
op
Noteb
ook Pa
p
e
rs
. 2004.
[24]
Z
hang, Do
ng, W
e
i
Qi,
H
o
n
g
Jian
g
Z
h
ang.
A
ne
w
s
hot b
oun
dar
y detect
i
on
al
gor
ithm.
Advanc
es in
Multi
m
ed
ia Info
rmati
on Proc
es
sing-PCM
200
1
. Spring
er Ber
lin He
id
elb
e
rg.
200
1; 63-7
0
.
[25]
Hua
ng, Ch
un
g
-
Lin, Bi
ng-Ya
o
Lia
o
. A rob
u
st
scene-c
han
ge
detectio
n
meth
od for vi
de
o se
gmentati
o
n
,
Circuits and Sy
stem
s for Video
Technology, IEEE Transactions on.
2
001;
11(1
2
): 128
1-1
288.
[26]
Mustafa R, Z
hu D. A
n
Inv
e
stigati
on
into
Conte
n
t Bas
ed Vi
de
o pr
oc
essin
g
i
n
Cl
o
ud C
o
mp
utin
g
Paradigm,
Int'
l Conf.
Imag
e
Pr
ocessi
ng and
Co
mp
uter
Visi
on,
IPCV’
13. ISBN #: 1-
60
13
2-25
3-4. 2
0
1
3
;
II: 933-938.
[27]
Mustafa R, Z
h
u D. Object
i
o
n
abl
e ima
ge
det
ection
in c
l
ou
d
computi
ng
par
adi
gm-a rev
i
e
w
.
J
ourn
a
l of
Co
mp
uter Scie
nce
. 201
3; 9(1
2
): 1715-
17
21.
[28] http://family
saf
e
medi
a.com/pornography
_st
a
tist
ics.html#import
ant_countries; Accessed at 20
th
March
201
4.
Evaluation Warning : The document was created with Spire.PDF for Python.