Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
m
put
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
9
, No
.
3
,
J
un
e
201
9
, pp.
1750~1
756
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v9
i
3
.
pp.17
50
-
1756
1750
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
RTL
im
plement
atio
n
of i
mage c
om
pressio
n techn
iqu
es
in
WSN
S. Arun
a
Dee
pt
hi
1
,
E.
Sree
niva
s
a
R
ao
2
,
M.
N. Gir
i
P
r
as
ad
3
1,3
Depa
rtment
of
Elec
tron
ic
s
and Com
m
unic
at
ion Engi
ne
eri
ng,
Ja
waha
rlal
Nehru
Te
chno
logi
c
al Unive
rsit
y
,
Indi
a
2
Depa
rt
m
ent
of Electronics a
n
d
comm
unic
at
ion Engi
ne
eri
ng,
Va
savi
co
ll
eg
e
of
E
ngine
er
ing, India
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
ul
10
, 2
01
8
Re
vised
Dec
17
, 2
01
8
Accepte
d
Dec
2
7
, 201
8
The
W
ireless
se
nsor
net
works
have
l
imitations
r
ega
rding
data
r
e
dundancy
,
power
a
nd
r
equ
ire
high
bandwi
dth
when
used
for
m
ult
imedia
dat
a
.
Im
age
compress
ion
me
thods
over
co
m
e
the
se
prob
le
m
s.
Non
-
neg
at
iv
e
Matrix
Fact
ori
za
t
ion
(NM
F)
m
et
hod
is
useful
in
appr
o
ximati
ng
high
dimensional
dat
a
wher
e
th
e
dat
a
h
as
non
-
ne
gat
iv
e
comp
one
nts.
Another
m
e
thod
of
the
NMF
called
(PN
MF
)
Projec
ti
v
e
Nonnega
ti
v
e
Ma
tri
x
Fa
ct
ori
za
t
io
n
is
used
for
le
arn
ing
spat
ia
l
l
y
loc
a
lized
v
isual
pa
tterns.
Si
m
ula
ti
on
r
esult
s
show
the
compari
son
betw
ee
n
SV
D
,
NMF,
PN
M
F
compr
ession
sche
m
es.
Com
pre
ss
ed
images
are
tr
ansm
it
te
d
from
base
stat
ion
to
cl
us
te
r
he
ad
node
a
nd
rec
e
ive
d
from
ordina
r
y
nodes.
The
sta
t
ion
ta
kes
on
th
e
image
restor
a
ti
on.
Im
age
qual
ity
,
compres
sion
rat
io
,
signal
to
noise
ra
ti
o
a
nd
ene
rg
y
consu
m
pti
on
are
the
essential
m
et
ric
s
m
ea
sured
f
or
compress
ion
per
form
anc
e
.
I
n
thi
s
pape
r
,
the
compress
ion
m
et
hods
are
designe
d
using
Matl
ab
.
The
p
ar
amete
rs
li
k
e
PS
NR,
the
tot
a
l
node
ene
rg
y
con
sum
pti
on
are
cal
cul
a
te
d.
R
TL
sc
hemati
c
of
NM
F SVD,
PN
MF
m
et
hods i
s ge
ner
a
te
d
b
y
usin
g
Veri
log
HD
L.
Ke
yw
or
d
s
:
NMF
PN
MF
RTL
SVD
WSN
Copyri
ght
©
201
9
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
S. Ar
un
a
Dee
pt
hi,
Dep
a
rtm
ent o
f El
ect
ro
nics
and C
omm
un
ic
ation
En
gin
ee
rin
g,
Vasa
vi C
ollege
of E
ng
i
neer
i
ng
,
Hyde
rab
a
d, I
ndia
.
Em
a
il
: sadeepthi@staf
f.vce.a
c.in
1.
INTROD
U
CTION
A
sen
sor
ne
tw
ork
[
1]
,
[
2
]
co
ns
ist
s
of
m
eas
ur
i
ng,
com
pu
ti
ng,
an
d
c
omm
un
ic
at
io
n
el
em
ents
that
gi
ve
an
a
dm
inist
rat
or
t
he
a
bili
ty
to
obse
rv
e
an
d
r
eact
to
e
ve
nts.
Wh
il
e
se
ver
al
se
nsor
s
can
be
c
onnec
te
d
t
o
con
t
ro
ll
ers
a
nd
processi
ng
sta
ti
on
s
directl
y
(e.g
.
,
us
i
ng
loca
l
area
netw
ork
s),
m
any
sensors
sen
d
the
gat
her
e
d
data
wi
relessl
y
to
a
centrali
z
ed
processi
ng
sta
ti
on
.
T
his
is
nee
ded
since
m
any
netw
ork
ap
plica
ti
on
s
r
equ
i
re
hundre
ds
or
th
ou
s
an
ds
of
sen
so
r
nodes,
of
te
n
set
up
in
re
m
ote
and
unre
ac
hab
le
areas
.
Figure
1
s
how
s
two
sens
or
fiel
ds
m
on
it
or
ing
tw
o
differe
nt
ge
ogra
ph
ic
re
gions
a
nd
c
onnec
ti
ng
t
o
the
I
nt
ern
et
us
i
ng
th
ei
r
ba
se
sta
ti
on
s
[
3
].
The
se
ns
in
g
a
nd
co
ntr
ol
te
chnolo
gy
c
om
pr
ise
el
ect
ric
and
m
agn
et
ic
fiel
d
sens
ors;
r
adio
-
wa
ve
fr
e
qu
e
ncy
s
e
nsors;
opti
cal
,
el
ect
ro
-
opti
c
-
,
a
nd
inf
rare
d
se
ns
ors;
rad
a
rs;
la
sers;
locat
io
n/n
a
vig
at
io
n
s
ens
or
s;
sei
s
m
ic
and
pressu
re
-
wa
ve
sens
or
s;
e
nvir
on
m
ental
par
a
m
et
er
sens
or
s
(e.
g.
,
wi
nd,
hu
m
idit
y,
heat);
an
d
bio
c
hem
ic
al
national
secur
it
y
-
or
ie
nted
se
nsors
.
To
day’s
sens
or
s
ca
n
be
descr
ibe
d
as
‘‘
sm
art’’
inex
pensi
ve
dev
ic
es
e
qu
i
pped
with
m
ult
iple
onbo
a
r
d
sensing
el
em
en
ts;
they
are
lo
w
-
c
os
t
low
-
po
wer
m
ulti
fu
nctiona
l
nodes
that are
l
og
ic
al
ly
hom
e
d
to
a ce
ntral si
nk no
de [
4
]
.
W
i
reless
M
ul
tim
edia
Sens
or
Netw
orks
(
WMSN
s)
h
as
broa
d
a
ppli
cat
ion
s
[
5]
in
in
dustria
l
pro
du
ct
io
n,
en
vir
on
m
ental
m
on
it
or
ing
[
6
]
.
The
s
pace
-
tim
e
relat
ivit
y
-
base
d
da
ta
co
m
pr
ession
al
gorithm
m
ai
nly
include
s
pr
e
dicti
on
co
ding
an
d
li
near
fitt
ing
m
e
thod
for
tim
e
seri
es.
A
predict
io
n
co
ding
m
e
tho
d
i
s
pro
po
se
d
by
[
7
]
.
It
eval
uates
the
s
ource
data
base
d
on
t
he
ti
m
e
relat
ivit
y
of
the
s
ource
da
ta
.
The
WMS
N'
s
are
diff
e
re
nt
from
tradit
iona
l
WSN
s
[8
]
,
[
9]
in
case
of
data
processi
ng,
a
nd
ene
rg
y
c
on
su
m
ption
of
w
irel
ess
transceive
r.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
RTL
im
ple
men
t
ation of i
mage c
ompressi
on te
chn
i
qu
e
s in
W
SN
(
S. Ar
una Deept
hi
)
1751
Figure
1
.
W
irel
ess senso
r net
works
2.
PROP
OSE
D
METHO
D
In
a
WSN,
the
data
t
ran
sm
issi
on
pr
ocess
[
10
]
ca
n
be
di
vi
ded
into
data
com
pr
essio
n
e
ncodin
g
an
d
data
dec
od
i
ng.
The
sen
ding
a
nd
receivin
g
of
i
m
ages
in
W
S
Ns
can
be
sho
wn
in
Fi
gure
2
.
SVD,
NMF
,
PN
M
F
i
m
age co
m
pr
es
sio
n m
et
ho
ds
a
re
us
ed
.
Figure
2
.
The
im
ag
e sen
ding
and recei
ving
process
First,
the
giv
e
n
inp
ut
i
m
age
is
resized
to
512x51
2.
The
re
siz
ed
i
m
age
is
pr
ep
ro
ce
ssed
a
nd
co
nv
e
rted
into
c
orrespo
ndin
g
pix
el
val
ue
s
us
in
g
a
Ma
t
la
b
pro
gram
a
nd
store
d
i
n
th
e
m
e
m
or
y
file
.
The
c
onve
rted
pix
el
values
are
in
put
to
im
age
com
pr
essi
on
m
et
ho
d
wh
ic
h
is
de
ve
lop
e
d
by
us
in
g
Ver
il
og
H
D
L
w
hich
is
bas
ed
on
hard
war
e
w
hich
m
akes th
e
process
faster
.
H
ardware/s
of
t
w
are c
o
-
sim
ulatio
n
sho
wn in
Fi
gure
3.
Figu
re
3
.
Ha
rdwar
e/
s
of
t
war
e
c
o
-
sim
ulati
on
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
3
,
June
2019
:
1750
-
1756
1752
Figure
4
s
how
s
the
i
m
age
values
w
hich
are
i
n
file
s
are
store
d
in
the
m
e
mo
ry
of
the
Mode
lSim
us
ing
the
com
m
and
"$r
ea
dm
e
m
h.
”
fo
r
he
xad
eci
m
al
values
an
d
“$
read
m
e
m
b”
for
bin
a
ry
va
lues.
T
he
file
s
con
ta
in
hex
a
decim
al
v
al
ues.
H
DL
D
AEMO
N
com
m
and
co
ntr
ols
the
serv
e
r
th
at
su
pport
s
interact
ion
s
with
HD
L.
Figure
5
s
how
HD
L
D
AEMO
N betwee
n
M
ATL
AB a
nd
Mod
el
Sim
.
Figure
4
.
Pixel
v
al
ue
s in
m
e
mo
ry l
ocati
on
Figure
5
.
HDL
DA
EM
O
N between
MAT
LA
B an
d
Mo
delSi
m
2.1
.
Im
ag
e
com
pressio
n u
si
ng
N
MF
The
NMF
al
gorithm
[
11
],
[
1
2
]
exec
utes
m
or
e
it
erati
on
s
to
get
bette
r
i
m
age
qu
al
it
y.
Im
ages
are
div
ide
d
i
nto
se
ver
al
blo
c
ks
,
a
nd
ada
ptive
im
age
c
om
pr
ession
al
gorithm
is
ta
k
en
base
d
on
the
NMF
[1
3
],
[
1
4
]
to
process
eac
h
i
m
age
blo
ck
.
The
com
pr
ess
ion
rati
o
of
the
NMF
[1
5
],
[
16
]
m
et
ho
d
de
pe
nds
on
the
siz
e
of
the
rank
of
the
ba
se
m
at
rix.
The
refor
e
,
the
co
m
pr
ession
rati
o
can
be
m
od
ifie
d
with
the
dem
and
s
in
the
energy
consum
ption
.
Lo
wer
qu
al
it
y
is
sel
ect
ed
to
im
pr
ov
e
the
c
om
pr
ession
rati
o
to
sa
ve
ene
rgy
.
Suppose
a
n
or
i
gin
al
i
m
age
with
the
siz
e
of
m
x
n,
is
div
ide
d
int
o
sev
eral
bl
oc
ks
the
dim
ensi
on
s
of
w
hich
a
re
p
x
q.
T
he
im
age
co
m
pr
essio
n
ra
ti
o
is gi
ven b
y
(1)
.
=
=
(1)
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
RTL
im
ple
men
t
ation of i
mage c
ompressi
on te
chn
i
qu
e
s in
W
SN
(
S. Ar
una Deept
hi
)
1753
2.2
.
Im
ag
e
com
pressio
n u
si
ng
SVD
Singular
V
al
ue
D
ec
om
po
sit
ion
(S
V
D
)
is
represente
d by the
(
2)
.
(2)
wh
e
re
G
is a
m
× n m
at
rix
Im
age co
m
pr
ession
[
17
]
i
nvol
ves red
ucin
g
t
he
r
e
dundancies
in
a
n
im
age which
a
re i
n
the
form
o
f
1.
Psycho
-
visu
al
redu
nd
a
ncy,
w
hich
is
due
to
the
lim
it
a
ti
on
s
[1
8
]
of
the
hum
an
visu
al
s
yst
e
m
to
interpr
et
fine det
ai
ls i
n
a
n
im
age.
(i.e.
, visuall
y n
on
-
es
sentia
l feat
ures
)
2.
In
te
r
pix
el
re
dund
a
ncy,
3.
Cod
i
ng r
e
dund
ancy
Rem
ov
in
g
the
re
dundancies
is
[
1
9
]
,[
20
]
r
edu
ci
ng
the
num
ber
of
bits
in
a
n
im
age
with
out
th
e
decr
ease
in
the
i
m
age
qual
it
y.
H
ow
e
ve
r,
m
erely
app
ly
in
g
S
VD
[
21
],
[
22
]
on
an
im
age
do
no
t
com
pr
ess
it
an
d
a
few
sin
gula
r
values
wer
e
re
ta
ined
w
hile
ot
her
un
i
qu
e
val
ues
we
re
ig
nor
ed.
T
he
sin
gu
l
ar
v
al
ue
s
we
re
ta
ken
in d
escen
ding
order
on
the d
i
agonal o
f D,
and
that first sin
gu
la
r
value
co
ntains the m
os
t
sign
ific
ant am
ount of
inf
or
m
at
ion
, a
nd foll
ow
i
ng si
ngular val
ues
c
on
ta
in
d
ec
reasi
ng am
ou
nts
of
i
m
age in
f
or
m
ation
.
Thus
, th
e
low
e
r
singular
valu
es
con
ta
inin
g
le
s
s info
rm
ation
c
an be ig
nore
d wit
hout
distor
ti
on in
t
he
im
age.
Fr
om
the
property
1
of
SVD,
it
can
be
ref
e
r
red
that
“t
he
r
ank
of
G
is
ta
ken
as
t
he
nu
m
ber
of
non
-
zero
sin
gula
r
values
"
.
But
if
the
lower
ord
er
singular
values
after
the
r
ank
of
the
m
a
trix
ha
ve
ne
gligible
values
, th
ey
a
r
e co
ns
ide
red as
noise
.
In
(
2) ab
ove ca
n
al
s
o be
wr
it
te
n
as
………+
(3)
wh
e
re
r
is t
he
r
ank of
G.
Fr
om
pr
ope
rty
1
of
S
VD
(r
e
f
er
sect
ion
3
a
bove
),
it
fo
ll
ow
s
that
trun
cat
in
g
(
3)
ti
ll
r
values
do
es
no
t
m
ake
any
signi
ficant
change
in
the
i
m
age.
But
co
m
pr
ess
ion
will
be
ve
ry
le
ss
wh
il
e
t
he
i
m
age
qu
al
it
y
is
sam
e.Fo
r
go
od am
ou
nt of c
om
pr
ession
t
o b
e achiev
ed
, onl
y t
he
fir
st k
v
al
ues of
(
4)
(4)
wh
e
re
k
<
r
The
im
age r
econ
st
ru
ct
e
d
will
r
ed
u
ce
[
2]
the
stora
ge
sp
a
ce r
equ
i
rem
ent to k
*(m
+n+1
) byt
es as ag
ai
ns
t
the
stora
ge
s
pa
ce
requirem
e
nt
of
m
*n
byte
s
of
th
e
ori
gi
nal
un
c
om
pr
e
ssed
im
age.
N
ow,
com
pr
essi
on
is
achieve
d
if
the
storag
e
s
pac
e
req
ui
red
by
the
com
pr
esse
d
i
m
age
is
le
s
s
than
that
require
d
b
y
the
ori
gin
al
i
m
age.
m
*n
> k*(
m
+n+1
)
(5)
2.3 Im
age c
om
pressio
n u
si
ng
PN
MF
In
NMF
,
W
an
d
H
ha
s
r×
(m
+
n)
fr
ee
pa
ram
eter
s
[
20]
.
Co
ns
i
der
V
a
nd
W
a
re
col
um
n
vect
or
s
an
d
H
is
a
scal
ar:
obvi
ously
,
there
a
re
m
any
so
luti
ons
W=
1/
H
V
with
H
a
r
bitrary.
Ba
sed
on
t
his,
a
no
vel
m
et
ho
d
cal
le
d
Pro
j
ect
ive
Non
-
ne
gative Ma
trix
Fact
or
iz
at
io
n
(
PNM
F)
is ta
ken a
s the soluti
on t
o
the
pr
ob
le
m
(6)
wh
e
re || •
||
is a
m
a
trix nor
m
.
The
m
os
t
us
ef
ul
norm
is
the
Eucli
dea
n
dis
ta
nce
betwee
n
two
m
at
rices
A
an
d
B,
or
th
e
F
robe
nius
m
at
rix
norm
o
f
their
dif
fer
e
nc
e:
(7)
a d
ive
r
gen
ce
of m
at
rix
A fr
om
B d
efined
as
(8)
Both
Eucli
dea
n
distance
a
nd
d
ive
rg
e
nce
are low
e
r
boun
ded
by
zero
,
a
nd
vanish
if
an
d
only
if
A
=
B.
The
PN
MF
m
et
hod
seem
s
to
offer
s
om
e
adv
a
ntage
s
as
com
par
ed
to
NMF.
The
fi
r
st
one
has
i
nc
reased
or
t
hogonalit
y
of
the
basis
ve
ct
or
s.
T
his
is
du
e
to
the
sim
il
arit
y
of
the
c
rite
rion
[
23
]
to
SVD.
Re
m
ov
ing
the
po
sit
ivit
y
const
raint
bu
t
ke
epin
g
the
ra
nk
co
ns
trai
nt,
an
ort
hogo
nal
e
igenv
e
ct
or
b
asi
s
is
a
so
luti
on.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
3
,
June
2019
:
1750
-
1756
1754
Fo
r
po
sit
ive
ba
ses,
or
t
hogona
li
ty
is
relat
ed
to
s
pa
rsen
e
ss.
Co
ns
ide
r
the
case
in
wh
ic
h
the
V
m
at
ri
x
is
a
trai
ning
set
,
an
d
the
goal
is
to
find
the
re
pr
es
entat
ion
not
on
ly
fo
r
the
colu
m
ns
of
V
bu
t
f
or
ne
w
vect
or
s
,
too
.
Fo
r
P
NMF,
t
he
repr
ese
ntati
on
for
a
ny
col
umn
of
V,
say
v,
is
sim
ply
WW
T
v
a
nd
that
ca
n
be
easi
ly
com
pu
te
d
for
a
ne
w
vector,
to
o.
I
n
NM
F,
the
re
is
no
s
uch
natu
ral
re
presentat
io
n
be
c
ause
bo
t
h
W
a
nd
H
a
re
nee
de
d,
a
nd
m
at
rix
H
has
only
n
c
ol
um
ns
.
The
ext
ra
c
olum
n
in
H
woul
d
ha
ve
t
o
be
re
c
om
pu
te
d
from
the
crit
eri
on.
T
hir
d,
as pointe
d ou
t
by
[
24
].
NMF
has
a
close
rela
ti
on
to
cluste
ring.
3.
RESU
LT
S
3.1
.
Rel
at
i
ve
pa
r
amet
er
s in
simul
ati
on ex
peri
ment
In
t
his
pa
pe
r
Ma
tl
ab
is
us
e
d
f
or
com
pr
es
sing
im
ages
usi
ng
NMF,
S
VD
a
nd
Ver
il
og
is
us
e
d
for
gen
e
rati
ng
RT
L
schem
at
ic
of
S
VD
an
d
N
MF
al
gorithm
s
in
WSN.
Figure
6
s
hows
input
im
age
of
siz
e
512×
512×
8
-
bit
gr
ay
scal
e
im
a
ges
is
c
ollec
te
d
by
cam
era
nodes
pe
rio
dical
ly
.
Her
e
15
ca
m
eras
co
ver
each
detect
ion
ra
nge
of
a
10
0
x1
00
m
rectangula
r
area,
an
d
the
ir
sensi
ng
ra
dius
is
11m
.
Ele
ve
n
ordina
ry
node
s
are
dep
l
oyed
i
n
the
c
omm
un
ic
a
ti
on
ra
nge
of
each
cam
era
node.
The
sim
ula
te
d
netw
ork
str
uctu
re
f
or
t
hre
e
m
et
ho
ds i
s s
hown in Fi
gure
7.
Figure
6
.
I
nput
i
m
age 512x
512
(a)
(
b)
(c)
Figure
7
(a)
.
N
et
work str
ucture o
f NMF c
ompressi
on m
echan
ism
(b
) Net
work str
uctu
re
of PN
M
F
com
pr
essio
n
m
echan
ism
(c)
N
et
work str
ucture o
f
S
VD c
ompressi
on m
echan
ism
3.
2
.
A
na
l
ys
is
of ener
gy cons
umpt
i
on
Figure.
8
s
how
s
NMF
ene
rg
y
distrib
utio
n
figur
e
when
sen
ding
a
512
x5
12x8
-
bit
im
ag
e
to
the
ba
s
e
sta
ti
on
.
The
si
m
ula
ti
on
resu
lt
ind
ic
at
es
that
the
us
e
of
the
three
c
om
pr
ession
m
et
ho
ds
de
scribe
d
in
this
pap
e
r
reduces
t
he
a
ve
rag
e
e
ne
rg
y
c
on
s
um
ption
of
the
cam
era
nodes
by
nea
rly
an
order
of
m
agn
it
ud
e
lo
we
r
t
ha
n
the
centrali
zed
a
ppro
ac
h.
T
he
ene
rg
y
co
nsum
pti
on
of
nodes
ha
s
a
sta
ble
balance,
w
hic
h
is
he
lpf
ul
to
i
m
pr
ove
the
li
fetim
e
of
network
a
nd
exte
nd
t
he
li
fe
cy
cl
e.
In
the
T
a
bl
e
1
,
PS
NR
val
ues
ex
presse
d
in
dB
are
us
ed
as
a
m
easur
e
of im
age
qu
al
it
y
. Ta
ble
2 s
how x
il
inx IS
E
dev
ic
e
util
iz
at
ion
f
act
or
f
or N
MF
, SVD, P
NMF m
et
hods
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
RTL
im
ple
men
t
ation of i
mage c
ompressi
on te
chn
i
qu
e
s in
W
SN
(
S. Ar
una Deept
hi
)
1755
(a)
(b)
(c)
Figure
8
(a). E
nergy
distrib
ution o
f NMF im
age c
om
pr
ession net
wor
k
(
b)
En
e
rg
y
distrib
ution o
f PNMF
(
c)
Energy
distrib
ution o
f SV
D
i
m
age co
m
pr
es
sion net
work
Table
1.
PS
NR V
al
ue
for T
hree C
om
pr
essio
n
Me
thods
with t
he Im
age I
nput
size o
f 512
x512
Co
m
p
r
ess
io
n
M
eth
o
d
PSNR Valu
e
SVD
1
2
.00
4
8
NMF
1
7
.00
4
8
PNMF
2
0
.17
2
0
Table
2.
Xili
nx I
SE
D
e
vice
Util
iz
at
ion
Factor f
or
NMF, SV
D, P
NMF Met
ho
ds
Log
ic Utilization
NMF,
PN
MF,
SV
D
Nu
m
b
e
r
o
f
Slices
1817
Nu
m
b
e
r
o
f
Slices i
n
Flip Flo
p
s
1776
Nu
m
b
e
r
o
f
4 in
p
u
t L
UTs
3356
Nu
m
b
e
r
o
f
bo
n
d
ed
I
OBs
77
Nu
m
b
e
r
o
f
GCL
K
s
1
Maxi
m
u
m
Fr
eq
u
en
cy
8
0
.51
2
MHz
The
desi
gn
is
base
d
blo
c
k
spl
it
t
ing
co
nce
pt
with
blo
c
k
siz
e
of
8x
8
the
de
vice
util
iz
at
ion
facto
r
will
be
sim
i
la
r
to
al
l
the
three
m
eth
ods.
Xili
nx
I
SE
14.
1
ver
si
on
with
fam
ily
Virtex
6
-
XC
6VLX
757
an
d
pa
ckag
e
FF78
4
is
us
e
d.
Figure
9
s
how
R
TL sch
em
at
ic
of N
M
F, SV
D
, PNMF m
et
hods
.
Figure
9
.
R
TL
schem
at
ic
o
f
N
MF, S
VD, PN
MF m
et
ho
ds
4.
C
O
NC
L
US
I
O
N
In
this
pa
per
,
the
diff
e
re
nt
im
age
c
om
pr
ession
m
et
ho
ds
li
ke
SVD,
NMF
,
and
P
NMF
i
n
WSNs
a
r
e
desig
ne
d.
An
at
tem
pt
is
m
ade
to
co
nn
e
ct
Ma
tl
ab
and
V
eril
og
H
DL.Im
ages
are
c
om
pr
esse
d
an
d
tra
ns
fe
rre
d
us
in
g
Ma
tl
ab
and
RTL
sc
hem
at
ic
is
gen
erated
us
i
ng
Ver
i
lo
g
H
DL.
Im
age
qu
al
it
y,
com
pr
essio
n
rati
o,
s
ign
a
l
to
noise
rati
o
a
nd
ene
rg
y
c
on
su
m
ption
are
the
m
os
t
vital
m
et
rics
cal
culat
ed
us
i
ng
Ma
tl
ab.
Sim
ulatio
n
res
ults
sh
ow
t
hat
the
PN
MF
-
base
d
i
m
age
com
pr
es
sion
m
echan
is
m
pr
ovide
d
be
tt
er
PSN
R
a
nd
SVD
bas
ed
i
m
age
com
pr
essio
n
pro
vid
es
al
le
vi
at
e
the
ener
gy
con
su
m
ptio
n
of
cam
era
nodes
w
hic
h
are
the
key
r
oles
in
the n
et
wor
k.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
3
,
June
2019
:
1750
-
1756
1756
REFERE
NCE
S
[1]
Abdulla
h
AB,
R
aga
b.
K&
z
aman
.
N
,
“
W
ire
l
ess
Sensor
Networks
and
En
erg
y
Eff
i
ci
en
c
y
,
”
publi
sh
ed
b
y
IGI
Globa
l,
pp
1
-
18,
2012
.
[2]
Ahonen.
T,
P
ie
t
i
kai
nen
,
M.Z
h
ao,
G&Mata
s.J
,
.
“
Rota
ti
on
-
inva
r
iant
image
and
vi
deo
desc
r
ipt
ion
with
local
b
inar
y
pat
t
ern
f
ea
rtur
es
,
”
IE
EE Tr
ansacti
ons on
Image
P
roce
ss
ing
,
1465
-
1477,
21(4)
,
201
2.
[3]
Bengi
o.
Y.Courv
il
le.
A.C
&
Vince
nt
.
P
.
,
“
Uns
uper
vised
feature
l
ea
rning
and
de
e
p
le
arn
ing
:
A
R
evi
ew
and
new
Perspec
ti
v
es
,
”
Depa
rtmant
of co
m
pute
r
scie
n
ce and ope
ra
ti
on
res
ea
rch
,
Univ
ersity
of
Montre
al, M
ontre
al Ca
n
ad
a
[4]
Brunel
lo
.
D,
C
alvagno,
G
.
Mian,
G.A&Rinaldo.R
,
“
Lossless
co
m
pre
ss
ion
of
vi
deo
using
te
m
pora
l
informati
on
,
”
IEE
E
Tr
ansacti
o
n
s on
Image Proce
ss
ing
,
pp
.
132
-
139,
2003
.
[5]
Chang.
N.B
,
L
iu.M
,
“
Optimal
co
m
pet
it
ive
a
lgori
t
hm
s
fo
r
opportuni
stic
spe
ct
rum
a
cc
ess
,
”
IE
EE
Jo
urnal
on
sel
ec
t
e
d
areas i
n
commu
nic
ati
ons
,
pp
.
11
83
-
1192,
2008
.
[6]
Cheng,
C
,
Chen
.
X,
W
ang.
S &
Y
ang.
Y
,
“
Am
ult
ila
y
er
Im
prove
d
R
B
M ne
twork
bas
ed
image compress
ion
m
et
hod
in
W
ire
le
ss
sensor
net
works
,
”
2016
.
[7]
Dargi
e.
W
&
Poellabaue
r
.
C
,
“
Fundam
ent
al
s
of
W
ire
le
ss
senso
r
Networks:
the
ot
y
and
Pra
ct
i
ce,
”
John
W
il
e
y
a
nd
Sons
Lt
d.
W
est
S
uss
e
x,
UK
:doi
:1
0.
1002/9780470
666388.
[8]
Ding.
C.
He
,
X
&
Sim
on
H
,
“
O
n
the
equi
vql
en
ce
of
Nonnega
t
i
ve
Matri
x
fa
ct
o
riz
a
ti
on
and
Spect
r
al
cl
ust
eri
ng
,
”
Proce
ed
ings o
f
Confe
renc
e
data
mining
,
pp
.
606
-
610
,
2005
.
[9]
Dipa
y
an
,
“
Dee
p
le
arn
ing
with
Ha
doop
,
”
Publ
ishe
d
b
y
Pack
t
Publ
i
shing,
2017
.
[10]
Gao.
Y,
L
i,
W
.
X
,
L
.
Sun.G.L,
&
Zha
ng.
C
“
Reser
ch
on
data
com
pre
ss
ion
al
gori
th
m
base
d
on
pre
dic
ti
on
codi
ng
f
or
wire
le
ss
sensor
nodes,
”
Proc
ee
d
ings
of
int
ernational
f
orum
on
i
n
formation
te
chn
ology
and
a
ppli
c
ati
ons
(
IFI
TA
-
09)
vol.
1,
pp
.
283
-
2
86.
[11]
Huang.
K.Sidiro
poulos,
N
&
Swa
m
i
,
“
A
non
nega
ti
v
e
m
at
rix
fa
ctoriz
a
ti
on
rev
isit
e
d:
Uniquene
ss
a
nd
al
gorit
hm
for
s
y
m
m
et
ric
d
ec
o
positi
on
,
”
IE
EE
transacti
ons on s
ignal
proce
ss
in
g
,
pp
.
211
-
224
,
2014
.
[12]
Pri
y
ank
a
Mek
ala;Je
ffre
y
F
an;
W
en
Cheng
L
ai
a
nd
Ching
W
en
Hs
ue
,
“
Gesture
rec
ogni
ti
on
usin
g
neur
a
l
n
et
wor
ks
base
d
o
n
HW
/SW
cosimula
ti
on
pla
tform
,
”
Hindawi
journal
Adv
anc
es
i
n
software
engi
nee
ring
,
20
13
htt
p://dx.doi.org/10.1155/2013/
70
7248
.
[13]
Kahu.
S.
&
Rah
at
e
.
R
,
“
Im
age
c
om
pre
ss
ion
usin
g
SV
D
,
”
Inte
rnational
journal
of
advanc
em
ents
in
res
earc
h
and
te
chno
logy
,
No
.
8
,
pp
.
244
-
248
,
2013
.
[14]
Kim
.
J
&
K
y
ung
.
C.
M
.
“
A
lossle
ss
embedde
d
co
m
pre
ss
ion
using
significant
bit
t
runc
ation
for
HD
vide
o
codi
ng
,
”
IEE
E
transacti
o
ns on
systems f
o
r v
ide
o
t
ec
hnolo
gy
,
pp
.
848
-
860
,
2010.
[15]
Ta
o
L
i.
&
Chris
Ding
,
“
The
R
elati
onship
s
amon
g
var
ious
Nonnega
t
ive
Mat
rix
f
ac
tor
iz
a
ti
on
,
”
Si
xt
h
in
te
rnationa
l
conf
ere
n
ce on
D
ata
mining
,
”
htt
p
:/
/doi.org/
10
.
110
9/ICDM.2006.
1
60
,
2006
.
[16]
Li
.
H,
L
i.Z
&
W
en.
C
,
“
Fast
m
o
de
dec
ision
a
lgo
rit
hm
for
int
erf
r
am
e
codi
ng
in
fully
sc
alable
vid
eo
codi
ng
,
”
IEEE
transacti
ons on circui
ts
a
nd
syst
ems f
or v
id
eo tec
hnology
,
PP
889
-
898
[17]
Ma.
Z.
W
ang
.
W
,
Xu.M
&
Yu.
H
“A
dvanc
ed
scre
en
con
te
nt
c
oding
using
col
or
ta
ble
and
in
dex
m
ap”
IEEE
transacti
ons on i
mage
proce
ss
ing
,
pp
.
4399
-
4412
.
[18]
Minoli
.
D,
Sohra
b
y
.
K,
Zn
at
i
.
T
,
“
W
ire
le
ss
senso
r
Networks,
Tec
hnolog
y
,
Protoc
ols
and
applicati
ons,”
John
W
ile
y
&
Sons
,
In
c
Hob
oken,
Nj
:doi
:10
.
1002/047011276X.
[19]
Sw
ami.
A,
W
in
Hong,
Zha
o.
Q
,
“
W
ire
le
ss
Senso
r
Networks
signal
proc
essing
an
d
comm
unic
at
io
ns
per
spec
ti
v
es,
”
John W
il
e
y
&
S
ons L
td.
W
est
Sus
sex,
UK
.
[20]
W
ang.
K.J&Z
uo
.
C.
T
,
“
Im
prove
m
ent
s of
non
n
e
gat
iv
e
m
at
rix
fa
c
tori
z
at
ion
for im
age
ext
r
ac
t
ion,
”
Appl
.
Re
s.Compu
t
.
[21]
Xiang
Yu.L
,
X
ia
o
-
Chun.
Y
&
Ya
-
Zhe
.
W
,
“
Faci
ng
the
wir
ele
ss
sensor
net
works
strea
m
ing
dat
a
compress
ion
te
chno
log
y
,
”
Co
mputer
Scienc
e
vol.
34
,
no
.
2
,
pp
.
141
-
143
,
2007
.
[22]
Baghouri
Mos
tafa,
Ch
akkor
sa
a
d,
Hajr
aoui
Ab
der
rah
m
ane
,
“
Firefly
al
gori
thm
to
improve
th
re
shold
distri
but
e
d
ene
rg
y
eff
ic
i
ent
cl
uster
ing
al
gori
thm
for
het
ero
g
enous
W
ire
le
ss
senso
r
Networks
,
”
IAES
Inte
r
na
t
iona
l
Journal
of
Artifi
c
ia
l
I
nt
el
l
ig
enc
e
,
vo
l.
6,
No.
3
,
pp.
91
-
99,
20
17
.
[23]
K.Pani
m
ozhi
,
G
.
Maha
dev
an
,
“
QO
S
Fra
m
e
work
for
a
m
ult
i
stack
base
d
het
er
og
enous
W
ire
le
ss
sensor
net
work,
”
Inte
rnational
jou
rnal
of El
e
ct
ri
ca
l
and
C
omput
er
Engi
ne
ering
(
IJ
ECE
)
,
vol
.
7
,
n
o.
5p
,
pp
.
2713
-
2720,
2017
.
[24]
Em
y
Se
t
y
ani
ngs
ih,
Agus
Har
jok
o
“
Surve
y
of
H
y
brid
Im
age
co
m
pre
ss
ion
Te
ch
nique
s
,
”
Int
ern
a
ti
onal
J
ourna
l
o
f
El
e
ct
ri
ca
l
and
C
om
pute
r
Engi
n
e
eri
ng
(IJEC
E)
,
v
ol
.
7
,
No
.
4
,
pp
.
2206
-
2214
,
201
7
.
Evaluation Warning : The document was created with Spire.PDF for Python.