Int
ern
at
i
onal
Journ
al of
A
d
vances in
A
p
p
li
ed Sciences
(
IJ
A
AS
)
Vo
l.
7
, No
.
4
,
Decem
ber
201
8
, p
p.
361
~
368
IS
S
N:
22
52
-
8
8
14
,
DOI: 10
.11
591/ija
as
.v
7
.i
4
.
pp361
-
368
361
Journ
al h
om
e
page
:
http:
//
ia
e
score.
c
om/j
ourn
als/i
ndex.
ph
p/IJAA
S
Lossless
4D M
edi
ca
l
Imag
es Comp
ression Usin
g A
daptiv
e Inter
Sli
ce
s Fil
ter
in
g
Le
il
a
Be
lhade
f,
Z
ou
li
kh
a M
ekkaki
a M
aa
z
a
Depa
rt
m
ent
d
’In
form
at
ique
,
L
ab
ora
toi
r
e
SIM
PA
,
Facu
lté
des
Ma
thé
m
at
iqu
es
et d
’Inform
at
ique,
U
nive
rsit
é
des
Sci
enc
es
et
d
e la
Technol
ogie
d
’Oran
Mo
hamed
Boudia
f
,
US
TO
-
MB,
BP
1505,
E
l
Mnaou
er,
31000
Oran
,
Algér
ie
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ma
y
2
1
, 201
8
Re
vised
A
ug
5
, 201
8
Accepte
d
Aug
2
9
,
201
8
Rec
en
t
lossless
4D
m
edi
cal
images
compress
ion
works
ar
e
b
a
sed
on
the
appl
i
ca
t
ion
of
t
ec
hniqu
es
orig
in
at
ed
from
vid
eo
compress
ion
to
eff
ic
i
entl
y
el
iminate
red
un
danc
i
es
in
d
iffe
r
ent
dimensions
of
imag
e.
In
th
i
s
context
w
e
pre
sent
a
new
a
pproa
ch
of
los
sless
4D
m
edi
c
al
images
compres
sion
which
consists
to
appli
ca
t
ion
of
2D
wa
vel
e
t
tr
ansform
i
n
spatial
direct
io
ns
foll
owed
or
not
b
y
e
it
he
r
li
f
ti
ng
tra
nsfo
rm
or
m
oti
on
c
om
pensa
ti
on
in
inter
sli
ces
dire
c
ti
on,
the
o
bta
in
ed
slices
ar
e
code
d
b
y
3D
SP
IHT.
Our
app
roa
ch
was
compare
d
with
3D
SP
IHT
with/
without
m
oti
on
compensat
ion
.
The
r
esult
s
show
our
appr
oa
ch
offe
rs be
t
te
r
p
erf
orm
anc
e
in
lo
ss
le
ss
compress
ion
rate.
Ke
yw
or
d:
3D SP
IHT
4D m
edical
i
m
age
In
te
ger
wa
vele
t t
ran
s
form
Loss
le
ss
com
pr
essio
n
Moti
on
com
pen
sat
io
n
Copyright
©
201
8
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
:
Lei
la
Bel
had
ef
,
Dep
a
rt
m
ent d
’Infor
m
at
iqu
e, L
aboratoire
S
IMPA,
Fac
ulté de
s Mat
hém
at
iqu
es et d
’Info
rm
a
ti
qu
e,
Un
i
ver
sit
é
des Sci
ences et
de l
a Tech
no
l
og
ie
d’O
ran Mo
ha
m
ed
Bo
ud
ia
f
, UST
O
-
MB
BP 1505
, El M
naoue
r,
3100
0 Or
a
n, Al
gér
ie
Em
a
il
:
leila
.b
el
had
e
f@u
niv
-
ust
o.dz
1.
INTROD
U
CTION
4D
m
edical
i
m
ages
re
prese
nt
vo
l
um
ino
us
da
ta
wh
e
re
eac
h
i
m
age
is
com
po
se
d
by
a
set
of
volum
es
represe
nting
t
he
3D
im
ages
of
a
hu
m
an
body
par
t
at
dif
fer
e
nt
instants
t,
each
3D
im
age
is
i
n
tur
n
c
om
po
s
ed
by
a
set
of
sli
ces.
In
the
li
te
ratu
r
e
a
lot
of
los
sle
ss
4D
m
edica
l
i
m
ages
com
p
ressio
n
w
orks
are
reali
zed
to
reduc
e
i
m
ages
siz
e
wi
thout
data
l
os
s
in
or
der
to
a
void
dia
gnos
is
error
s
.
C
onsid
erin
g
4D
im
age
as
a
set
of
volum
es
evo
l
ving
in
ti
m
e,
the
r
ecent
lo
ssless
4D
m
edical
i
m
ages
co
m
pr
ession
te
ch
niques
a
pp
ly
th
e
t
echn
i
qu
es
of
vid
e
o
com
pr
essio
n
s
uch
a
s
m
otion
com
pen
sat
i
on
i
n
order
t
o
ef
fici
ently
rem
ov
ing
i
nter
sli
ces
or
volum
es
redu
nd
a
ncies.
Ba
sed
on
the
a
dv
a
nce
d
vid
e
o
cod
i
ng
H
.26
4/AV
C
Sa
nch
ez
et
al
.
[
1]
desc
ri
bed
tw
o
c
odin
g
m
e
tho
ds
for
lossless
4D
m
e
dical
im
ages
c
om
pr
essio
n.
T
he
first
m
et
ho
d
ex
plo
it
s
t
he
re
dundancies
am
ong
the
2D
sli
ces
in
(each
vo
l
um
e)
third
dim
ension
.
T
he
se
co
nd
m
et
ho
d
e
xp
l
oits
the
sim
il
ari
ties
betwee
n
t
he
sli
ces
i
n
t
he
f
ourth
dim
ension
ti
m
e
.
Re
dunda
nci
es
thr
ough
the
four
t
h
dim
ensi
on
are
m
or
e
num
ero
us
tha
n
re
dundancies
t
hroug
h
the thir
d dim
ension
; t
her
e
fore
this is t
he se
co
nd m
et
ho
d wh
i
ch perm
it
te
d
bette
r
com
pr
essi
on r
at
io
.
Fo
r
m
or
e
ex
pl
oit
the
r
ed
unda
ncies
in
t
he
four
dim
ension
s
of
m
edical
i
m
a
ges
the
sam
e
auth
or
s
of
th
e
pr
ece
de
nt
wor
k
pro
pos
ed
a
ne
w
c
om
pr
essio
n
te
c
hniq
ue
ba
sed
on
H
.26
4/AV
C
[
2]
a
s
i
n
pr
ece
de
nt
wor
k.
S
o
i
n
this
com
pr
essi
on
te
ch
nique
m
ulti
-
fr
am
e
m
oti
on
com
pen
sat
i
on
is
ap
plied
fi
rstly
in
dim
ens
ion
t
o
eac
h
vo
l
um
e.
The
obta
ined
r
efere
nce
a
nd
r
esi
du
al
sli
ces
a
re
treat
ed
sec
ondly
al
s
o
by
m
ulti
-
fr
am
e
m
otion
c
om
pen
sat
ion
i
n
tem
po
ral
dim
e
ns
io
n,
t
hus
the
redu
nd
a
ncies
are
re
duced
in
an
d
dim
ension
s
.
T
he
final
resid
ual
sli
ces
and
m
ot
ion
vecto
rs
are
c
om
pr
esse
d by ent
ropy c
od
i
ng.
Her
e
m
ulti
-
fr
a
m
e
m
otion
co
m
pen
sat
ion
ap
pl
ie
d
to
a
set
of
sli
ces
con
sis
ts
to
co
ns
ide
r
f
irst
sli
ce
as
ref
e
ren
ce
sli
ce
intra
c
od
e
d
an
d
−
1
sli
ces
as
i
nter
sli
ces
w
hich
are
inte
r
c
ode
d
by
va
riable
blo
c
k
m
at
chin
g
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2252
-
88
14
IJ
A
AS V
ol
.
7
,
No.
4
,
Decem
ber
2018
:
361
–
368
362
and
m
ulti
-
fr
am
e
m
otion
com
pen
sat
io
n
w
he
re
each
blo
c
k
of
inter
sli
ces
can
be
pr
e
dicte
d
f
ro
m
m
ulti
ref
eren
ce
sli
ces in
dim
e
ns
io
n
i
n
it
s f
ir
s
t app
li
cat
io
n
a
nd
dim
ension
in it
s secon
d
a
pp
li
cat
io
n.
Ma
rtin
et
al
.
pro
posed
in
[3
]
a
ne
w
m
et
hod
of
4D
m
edical
i
m
ages
com
pr
essi
on
al
so
base
d
on
t
he
H.264
vid
e
o
c
odin
g wh
e
re
pr
e
dicte
d
sli
ces ar
e g
e
ner
at
e
d
by
us
in
g
s
patio
-
te
m
po
ral ref
e
rence
sli
ces from
an
d
tem
po
ral
neig
hbor
hood,
wh
e
r
e
dif
fer
e
ntly
to
pr
ece
de
nt
te
chn
i
qu
e
w
hich
reali
zes
pre
dicti
on
in
dim
e
ns
io
n
fo
ll
owe
d
by
pr
edict
ion
i
n
di
m
ension,
this
t
echn
i
qu
e
reali
z
es
sp
at
io
-
te
m
po
ral
pr
e
dicti
on
wh
e
re
eac
h
bloc
k
of
inter
sli
ce
ca
n
be
pr
e
dicte
d
f
r
om
m
ulti
ref
er
ence
sli
ces
i
n s
patio
-
te
m
po
ral
neig
hborh
ood,
this
pe
rm
it
s
to
ob
ta
i
n
a good
res
ults in c
om
pr
essio
n com
par
ed pre
di
ct
ion
in
dim
ensio
n or p
re
dicti
on
in
ti
m
e d
i
m
ension.
All
the
pr
e
vious
ci
t
ed
w
orks
(
based
on
H.264
vid
e
o
c
od
ing)
don’
t
i
nte
gr
at
e
pro
gr
es
sive
lo
ssy
t
o
lossless
decodi
ng
in
their
sch
e
m
es.
Othe
r
te
chn
i
qu
e
s
a
pp
ly
wa
velet
tran
sf
or
m
ind
ee
d
it
c
an
be
c
om
bin
ed
with
i
m
age
com
pr
es
sion
te
ch
nique
as
f
ractal
to
im
pro
ve
ti
m
e
codi
ng
[
4]
al
s
o
w
a
velet
tra
ns
f
or
m
is ver
y
ef
fecti
ve
i
n
m
edical
i
m
age
an
d
vid
e
o
co
m
pr
ession
t
hus
a
lot
of
w
orks
are
pro
pose
d
as
in
[
5]
the
a
uthors
pr
ese
nte
d
c
olor
vid
e
o
m
edical
com
pr
essio
n
te
chn
i
qu
e
by
us
ing
geo
m
et
ric
wav
el
et
i
n
orde
r
to
el
im
inate
eff
ic
ie
ntly
sim
i
l
arit
y
i
n
each
f
ram
e
com
po
sed
the
vi
deo
seq
ue
nce
with
good
im
age
qu
al
it
y
bu
t
without
e
xp
l
oiti
ng
t
he
te
m
poral
redu
nd
a
ncies.
The
fi
rst
pro
pose
d
w
orks
ba
sed
on
wav
el
e
t
transfor
m
in
li
te
ratur
e
f
or
l
os
sle
ss
4D
m
edical
i
m
ages
com
pr
essio
n
do
no
t
i
nteg
rate
m
otion
com
pe
ns
at
io
n
in
com
pr
essi
on
sc
he
m
e
thu
s
i
n
[
6]
the
aut
hors
c
om
par
ed
three
te
c
hn
i
qu
es
us
i
ng
JP
EG
2000.
The
fi
rst
te
ch
nique
c
onsist
s
to
a
pp
ly
2D
wa
velet
tra
ns
f
or
m
to
eac
h
sli
ce
com
po
sed
the
vo
l
um
es
and
c
om
pr
ess
it
sep
aratel
y
by
us
i
ng
JP
EG
2000.
The
sec
ond
te
c
hn
i
qu
e
is
re
pre
sented
by
two
var
ia
nts:
the
first
va
riant
is
app
li
cat
ion
of
1D
wa
ve
le
t
transfor
m
i
n
di
m
ension
,
t
he
seco
nd
va
ri
ant
is
app
li
cat
io
n
of
1D
wa
velet
transfor
m
in
dim
ensio
n
a
nd
for
bo
t
h
va
riants
t
he
obta
ine
d
sli
ces
are
c
om
pr
esse
d
by
JPE
G
2000.
The
t
hi
rd
te
ch
ni
qu
e
is
a
ppli
cat
ion
of
1D
wave
le
t
transfor
m
in
dim
ension
flo
wed
by
1D
wa
velet
trans
form
in
dim
ension
the
ob
ta
i
ned
sli
ces
are
c
om
pr
essed
by
JPE
G2000.
The
ex
pl
oitat
ion
of
te
m
po
ral
dim
ension
in
s
econd
a
nd
t
hir
d
te
ch
ni
qu
e
pr
ov
i
des
bette
r
r
esult
s
in
com
pr
essio
n
rate
be
cause
t
he
nu
m
erous
si
m
il
arities in te
m
po
ral
dim
ension.
Du
e
to
e
ff
ic
ie
nc
y
of
t
he
m
oti
on
com
pen
sat
i
on
in
the
vi
deo
com
pr
essio
n
s
chem
e
to
redu
ce
tem
po
ral
redu
nd
a
ncies
Kasim
et
a
l.
[7
]
pro
po
se
d
a
wav
el
et
s
c
om
pr
essio
n
te
ch
nique
of
4D
m
e
dical
i
m
ages
based
on
m
ot
ion
com
pen
sat
io
n
an
d
int
eger
wa
velet
transfo
rm
,
her
e
the
4D
im
age
i
s
com
pr
essed
as
a
seq
uen
ce
of
3D
i
m
ages.
T
his
te
chn
i
qu
e
c
on
sist
s
to
achie
ve
3D
m
otion
com
pe
ns
at
io
n
in
orde
r
to
el
im
inate
eff
ic
ie
ntly
sim
i
l
arit
ie
s
existi
ng
betwe
en
volum
es
in
te
m
po
ral
dire
ct
ion
,
afte
r
t
he
res
ulti
ng
vo
l
um
es
are
dec
orr
el
at
ed
by
3D
i
ntege
r
wav
el
et
tra
ns
f
or
m
.
The
ob
ta
ined
data
are
cod
e
d
by
3D
SPIHT.
This
m
et
ho
d
al
lo
ws
lossless
c
od
i
ng
a
nd
pro
gr
essi
ve
lo
s
sy t
o
los
sle
ss
de
cod
i
ng.
The
fir
st
ste
p
is
m
otion
est
im
at
ion
inter
3D
i
m
ages.
Wh
ere
a
group
of
3D
im
ages
are
rep
rese
nte
d
by
one
key
im
age
(the
first
i
m
age
in
a
gr
oup
is
c
om
pr
ess
ed
without
m
otion
c
om
pen
sa
ti
on
)
an
d
−
1
inte
r
i
m
ages.
Eac
h
c
urren
t
im
age
(a
t
posit
ion
)
to
be
pr
e
dicte
d
is
di
vid
ed
i
n
c
ubes
f
or
eac
h
c
ub
e
i
t
m
os
t
si
m
il
ar
cube
in
ref
e
re
nce
im
age
(at
posit
ion
−
1
)
is
f
ound.
T
he
diff
e
re
nce
bet
ween
t
he
po
sit
ion
of
cu
rr
e
nt
c
u
be
a
nd
it
m
os
t
si
m
il
ar cu
be
is
m
ot
ion
vecto
rs
and the
d
i
ff
e
re
nce
betwee
n
t
hose c
ubes
gen
e
rates the
resid
ua
l im
age.
Th
us
the
res
ults
of
this
ste
p
a
re
3D
key
im
a
ge,
−
1
3D
resid
ua
l
i
m
ages
an
d
corres
pondin
g
m
ot
ion
vecto
rs.
To
rec
on
st
ru
ct
the
i
niti
al
gr
ou
p
o
f
im
ages,
the
key
i
m
age
an
d
the
corres
pondin
g
m
ot
ion
vecto
rs
is
use
d
to
ob
ta
in
pr
e
dicte
d
im
age
w
hi
ch
per
m
it
s
to
reconstr
uct
t
he
first
inter
im
a
ge
by
a
dd
i
ng
t
he
pr
e
dicte
d
im
age
to
the f
i
rst re
sid
ua
l im
age,
the
foll
ow
e
d i
nter
im
ages at
po
sit
i
on
are
rec
ons
tructe
d
as
the
f
i
rst
on
e
b
y
u
si
ng i
nter
i
m
age
at
posit
ion
−
1
and
c
orrespondin
g
m
otion
vecto
rs.
The
se
cond
ste
p
c
onsi
sts
to
a
pp
ly
3D
integer
wa
velet
trans
form
to
ke
y
i
m
age
an
d
r
esi
du
al
im
ages
an
d
in
fi
nal
st
ep
t
he
key
a
nd
resi
du
al
im
ages
are
c
od
e
d
by
us
i
n
g
3D
SP
I
HT,
al
so
t
he
m
otio
n
ve
ct
or
s
are
ent
r
op
y
co
de
d.
Thi
s
te
ch
nique
perm
it
s
to
reali
ze
lossless
c
om
pr
essio
n
and
pro
gr
e
ssiv
e
lossy
to
lossl
ess
dec
od
i
ng
by
transm
issi
on
a
par
t
of
bit
stream
of
key
im
ages
fo
ll
owe
d
by
a
par
t
of
bit
stream
of
each
inte
r
i
m
age
this
perm
it
s
to
i
m
pr
ov
e
progressi
vely
the
i
m
age
qu
al
it
y
in
deco
m
pr
ession.
All
ci
te
d
wor
ks
of
4D
m
edical
i
m
age
com
pr
ession
try
to
el
im
inate
the
re
dundancies
by
us
i
ng
di
ff
e
ren
t
pr
e
dicti
on
te
c
hniq
ues
in
and
dim
ension
s
ei
ther
by
a
pp
ly
in
g
m
otion
com
pen
s
at
io
n
or
by
app
ly
in
g
wav
e
le
t
trans
form
o
r b
y app
ly
in
g bo
t
h
m
otion
c
om
pen
sat
io
n
fl
ow
e
d by wa
velet
tr
ansfo
rm
.
In
this
pap
e
r
w
e
propose
a
ne
w
l
os
sle
ss
c
ompressi
on
a
ppr
oa
ch
base
d
on
a
dap
te
d
filt
erin
g
in
inte
r
sli
ces
directi
on
t
o
im
pro
ve
t
he
el
im
inati
on
of
re
d
unda
ncies
in
4D
m
edical
i
m
ages.
T
his
a
ppr
oa
ch
c
onsist
s
t
o
ap
ply
2D
inte
ger
wa
velet
tra
ns
f
or
m
to
each
sli
ces
f
ollow
e
d
or
not
by
ei
the
r
wa
ve
le
t
filt
er
or
m
ot
ion
com
pen
sat
i
on
in
inter
sli
ces
dire
ct
ion
.
The
obta
ined
sli
ces
are
cod
e
d
with
3D
SPIHT.
T
he e
xperim
e
ntal
resu
lt
s
are
com
par
ed
to
two
a
ppr
oac
he
s:
3D
SP
IHT
usi
ng
m
otion
co
m
pen
sat
ed
te
m
poral
filt
er
i
n
i
nter
sli
ces
dire
ct
ion
a
nd
3D
S
PI
H
T
(w
it
ho
ut
m
otion
c
om
pen
sat
io
n),
the
ob
ta
ine
d
l
os
sle
ss
c
ompressi
on
rates
sh
ow
our
a
ppr
oach
ou
t
perfor
m
s
two
oth
e
r
a
ppro
ac
he
s.
T
he
r
est
of
the
pa
pe
r
is
organ
iz
e
d
as
f
ollow
s
.
T
he
sect
i
on
2
prese
nts
pr
e
dicti
on
te
ch
niques
us
e
d
in
vi
deo
com
pr
essio
n.
The
sect
io
n
3
detai
ls
the
pro
po
s
ed
a
ppr
oac
h
of
com
pr
essi
on.
I
n
the
sect
i
on
4
the
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
A
AS
IS
S
N:
22
52
-
8938
Lo
ssless
4D M
edical
Ima
ges C
ompressi
on
Using A
daptiv
e I
nter S
li
ces
F
il
te
ring
(Lei
la
Bel
hadef
)
363
exp
e
rim
ental
resu
lt
s
for
l
os
sl
ess
c
om
pr
essio
n
a
re
com
pa
re
d
t
o
oth
e
r
com
pr
essi
on
a
ppr
oa
ches.
I
n
fi
nal
sect
ion
con
cl
us
io
ns
a
r
e presente
d.
2.
MOTIO
N CO
MPEN
SA
TI
O
N AND TE
M
POR
AL FILT
ERING
The
pr
i
ncipals
vid
e
o
c
om
pr
ession
te
c
hniq
ue
s
us
e
d
i
n
4D
m
edical
i
m
ages
c
om
pr
essio
n
are
represe
nted
by
tw
o
ty
pes
of
te
chn
i
qu
e:
H.2
64
/
AV
C
vi
deo
cod
i
ng
an
d
wa
velet
vid
e
o
co
di
ng
.
T
hese
te
ch
niques
a
re
base
d
on
m
ot
ion
c
om
pen
sat
io
n
to
el
im
i
nate
the
re
dund
ancies
as
in
H
.
264/AVC
c
ode
r
the
m
otion
c
om
pen
sat
ion
is
a
pp
li
ed
by
usi
ng
the
blo
ck
m
at
ching
t
o at
ta
in
t
he m
i
nim
u
m
pr
edict
ion
er
ror
w
hich
is
the
m
ini
m
um
diff
eren
ce
be
tween
current
blo
c
k
(to
be
pre
dicte
d)
a
nd
re
fer
e
nce
bl
oc
k
in
s
earch
wind
ow,
al
so
the
ref
e
r
ence
bl
oc
k
s
c
an
be
determ
ined
f
rom
m
u
lt
i refer
e
nce im
ages w
it
h varia
ble b
l
oc
k
siz
e.
The
lossless
4D
com
pr
essio
n
te
ch
niques
base
d
on
H
.26
4/AVC
offe
r
good
com
pr
es
sion
rate
but
without
pe
rm
i
t
ti
ng
the
pr
ogre
ssive
dec
odin
g,
th
e
al
te
r
nativ
e
is
t
he
wa
vele
t
te
ch
niques
w
hic
h
produce
s
cal
able
flu
x
an
d
reali
ze
good
decorre
la
ti
on
of
sig
nal
as
presente
d
i
n
[
7]
w
her
e
the
pro
gr
essi
ve
de
cod
i
ng
is
obta
ined
by
reord
e
rin
g
of
bi
t
stream
,
the
pro
gr
e
ssive
dec
od
i
ng
can
be
a
lso
ob
ta
ine
d
by
us
in
g
te
m
po
r
al
scal
abili
ty
bu
t
it
is
no
t
us
e
d here
c
ause t
he
m
otion
c
om
pen
sat
io
n
a
nd w
a
velet
t
ran
s
f
or
m
is achieve
d
se
par
at
el
y.
In
orde
r
t
o
obta
in
te
m
po
ral
scal
abili
ty
and
eff
ic
ie
nc
y
el
im
inate
tem
po
r
al
redu
nd
a
ncie
s
the
recent
wav
el
et
vid
e
o
cod
i
ng
a
pp
ly
t
he
wa
velet
tran
sform
in
m
otion
tra
j
ect
or
y
by
us
in
g
m
o
ti
on
c
om
pen
sat
ed
te
m
po
ral
filt
ering
(MCT
F)
.
I
n
wa
velet
vid
e
o c
odin
g t
he M
CTF
ca
n
be a
ppli
ed
befor
e
spa
ti
al
dec
orrelat
ion
with
wa
velet
trans
form
(t+2D
)
or
a
fter
sp
at
ia
l
decorrelat
io
n
i
n
wav
el
et
do
m
ai
n
(2D+t
or
in
-
band)
fl
ow
e
d
by
e
ntr
op
y
c
od
i
ng
.
The
fi
rst
works
in
vi
deo
c
ompressi
on
[8
]
,
[
9]
integrate
d
Ha
ar
m
otion
com
pensat
ed
te
m
p
or
al
filt
erin
g
wh
e
re
t
he
im
a
ges
a
re
se
par
at
ed
in
eve
n
im
a
ges
a
nd
od
d
im
ages,
t
he
e
ve
n
i
m
ages
are
lo
w
pass
filt
ered
a
nd
the
odd
im
ages
ar
e
hi
gh
pass
filt
ered,
t
he
obta
ined
lo
w
pas
s
i
m
ages
are
al
s
o
sepa
rated
in
e
ven
im
ages
an
d
odd
i
m
ages
an
d
filt
ered
ti
ll
the
te
m
po
ral
le
vel
de
com
po
sit
ion
i
s
done
.
T
he
filt
erin
g
is
reali
zed
in
te
m
po
ral
tr
ajecto
ry
determ
ined
by
m
otion
c
om
pen
sat
io
n
to
re
du
ce
t
he
e
nerg
y
in
ob
ta
i
ned
high
pa
ss
filt
ered
im
age
s
.
With
the
e
m
erg
ence
of
l
ifti
ng
sc
hem
e
fo
r
the
cal
c
ulati
on
of
t
he
integ
er
wa
velet
coeffic
ie
nts,
ot
her
works
we
re
car
ried
a
s
in
[10]
t
he
a
ut
hors
de
fine
d
two
filt
ers:
m
otion
com
pen
sat
ed
Haa
r
li
ftin
g
filt
er
an
d
m
ot
ion
com
pen
sat
ed
5/
3
li
fting
f
il
te
r
; t
he use
of m
oti
on co
m
pen
sat
e
d 5/3
li
ftin
g fil
te
r gave
the
bes
t resu
lt
s.
The
sc
hem
e
lif
ti
ng
is
ac
hiev
ed
us
i
ng
tw
o
ste
ps
pre
dicti
on
an
d
update:
the
predict
io
n
ste
p
retai
ns
diff
e
re
nce
bet
ween
tw
o
sam
ples
of
si
gn
al
(
high
pass
filt
ering)
a
nd
t
he
up
date
ste
p
retai
ns
th
e
a
ver
a
ge
there
fore
an
a
ppr
ox
im
at
i
on of si
gn
al
(lo
w pass
filt
erin
g).
In
[11]
the
aut
hors
int
rod
uce
d
the
m
otion
c
om
pen
sat
ed
tr
un
cat
e
d
5/
3
li
f
ti
ng
filt
er
w
he
re
only
the
pr
e
dicti
on
ste
p
of
li
ftin
g
is
pe
rfor
m
ed,
th
us
each
high
pa
ss
filt
ered
im
a
ge
is
obta
ine
d
by
determ
inin
g
the
pr
e
dicti
on
er
ror f
ro
m
neig
hbor
re
fer
e
nce
im
a
ges f
orwa
r
d an
d b
ack
ward.
We
pro
pose
an
a
dap
ti
ve
predict
ion
i
n
order
to
am
el
iorate
los
sle
ss
4D
m
edical
im
a
ges
com
pr
essio
n
rate,
we
a
pp
l
y
ei
ther
5/
3 t
runcated
li
fting
f
il
te
r or
m
ot
ion
c
om
pen
sat
e
d, if
one
of them
m
ini
m
i
ze p
red
ic
ti
on e
rror i
n
i
nter
sli
ces directi
on.
3.
PROP
OSE
D CO
MP
RESSI
ON SC
HEME
The
first
ste
p
of
ou
r
sc
hem
e
is
form
at
ion
of
GOS
(
Gro
up
O
f
Sli
ces)
from
4D
m
edical
i
m
age
f
ollo
w
e
d
by
ap
plica
ti
on
of
2D
i
ntege
r
wav
el
et
tra
ns
f
orm
(2
D
I
WT)
i
n
sp
at
ia
l
direct
ion
s
(
,
)
of
eac
h
sl
ic
es,
the
seco
nd
ste
p
is
inter
sli
ces
filt
erin
g
of
each
GOS
a
nd
t
he
final
ste
p
is
co
di
ng
of
ob
ta
ine
d
sli
ces
,
m
otion
vecto
rs
a
nd
et
iqu
et
te
s.
As
s
how
n
in
Fig
ure
1
p
rop
os
ed
com
pr
ession
sc
he
m
e
.
Figure
1. Pro
pose
d
c
om
pr
ession sc
hem
e
We
descr
i
be
t
he
se steps
in
t
he
foll
ow
i
ng poi
nts.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2252
-
88
14
IJ
A
AS V
ol
.
7
,
No.
4
,
Decem
ber
2018
:
361
–
368
364
3
.
1.
C
on
s
tructi
on
of GO
S
and sp
at
i
al t
r
an
s
fo
rm
The
GOS
ar
e
f
or
m
ed
from
the
sli
ces
of
4D
m
edical
i
m
age
(
,
,
,
)
as
s
hows
in
Fig
ur
e
2(a)
.
This
const
ru
ct
io
n
pe
rm
i
ts
to
obta
in
a
set
of
G
OS
re
pr
ese
ntin
g
4D
m
edical
i
m
a
ge
with
eac
h
G
OS
is
com
po
se
d
with
near
est
sli
ces in s
patia
l direct
ion
acr
os
s
tim
e.
Figure
2. (a
) 4
D
Im
age c
om
po
se
d by 3 v
olum
es. (
b)
C
onstructio
n of G
O
S co
m
po
se
d by
16 s
li
ces
(in
gr
ay
)
In
Fig
ur
e
2(b
)
the
sli
ces
in
gray
represent
t
he
first
GO
S
wi
th
16
sli
ces
of
4D
im
age
com
po
s
ed
by
3
vo
l
um
es,
thu
s
al
l
GO
S
of
4D
i
m
age
are
al
so
form
ed
with
fol
low
sli
ces
in
the
sam
e
directi
on
us
e
d
f
or
t
he
first
GOS
(ac
r
os
s
ti
m
e),
as
the
sec
ond
G
OS
sta
rts
with
sli
ce
17.
Nex
t
t
he
form
a
ti
on
of
G
OS
ea
ch
sli
ce
is
tra
nsfo
rm
ed
by
2D
intege
r
wav
el
et
tra
nsfo
rm
us
ing
li
ft
ing
sc
hem
e
[1
2]
in
sp
at
ia
l
di
recti
on
s
(
,
).
T
his
tran
sf
or
m
is
rev
e
rsible; i
t al
lows
pro
duci
n
g
inte
ger wa
ve
le
t coeff
ic
ie
nts which
pe
rm
i
ts
lossless c
om
pressi
on
.
3
.
2
.
I
nt
er
Sli
ces Fil
tering
Af
te
r
s
patia
l
transfo
rm
each
ob
ta
ine
d
GOS
is
filt
ered
by
t
he
pro
pose
d
filt
er
in
inter
sli
c
es
directi
on
base
d
on
tr
un
c
at
ed
li
fting
sc
hem
e
and
m
otion
com
pen
sat
io
n.
T
he
m
otion
com
pen
sat
ed
tru
ncated
5/3
l
ifti
ng
filt
er
pro
posed
in
[
11]
is
reali
zed
by
ap
plica
ti
on
of
t
he
tr
uncat
ed
li
fting
sc
hem
e
and
m
oti
on
com
pen
sat
ion
i
n
the
sam
e
tim
e
,
the
t
runcated
li
fting
sc
hem
e
achieve
s
on
l
y
pr
e
dicti
on
ste
p
of
li
ftin
g
s
c
hem
e
.
The
m
oti
on
com
pen
sat
ed
t
r
un
cat
e
d 5/3
li
ft
ing
filt
er (
MC
Trunc
5/3)
for t
he
bl
ock
S
k
[
m
, n
]
ca
n be
for
m
ula
te
d
as:
[
,
]
=
[
,
]
−
⌊
0
.
5
×
(
−
1
[
−
1
,
−
1
]
+
+
1
[
−
2
,
−
2
]
)
⌋
(1)
Wh
e
re
S
k
-
1
,
S
k+
1
: refer
e
nce
sli
ces, a
nd
S
k
: i
nter
sli
ce.
R
: resid
ual sli
c
e (
high
pass
sli
ce).
(
d
1m
, d
1n
): m
otion vect
or of
blo
ck
S
k
[
m,
n
]
t
o
a
posit
ion i
n
S
k
-
1
.
(
d
2m
, d
2n
): m
otion vect
or of
blo
ck
S
k
[
m,
n
]
t
o
a
posit
ion i
n
S
k+1
.
.
: co
rr
es
ponds t
o rou
nd operat
or.
The
m
otion
co
m
pen
sat
ion
is
r
eal
ise
d
by
us
i
ng
blo
c
k
m
otion
m
od
el
so
eac
h
inter
sli
ce
(to
be
predict
e
d)
is
div
i
ded
in
bl
ock
s
an
d
eac
h
blo
c
k
is
filt
er
ed
by
MC
T
runc
5/3
.
T
his
fi
lt
er
achieve
s
bi
-
directi
onal
m
otio
n
com
pen
sat
ion
t
hu
s
f
or
eac
h
c
urren
t
blo
c
k
in
S
k
it
m
os
t
si
m
il
ar
blo
c
k
is
de
te
rm
ined
in
S
k
-
1
and
in
S
k+1
a
nd
t
he
diff
e
re
nce
be
tween
the
posit
ion
of
cu
rr
e
nt
blo
c
k
a
nd
t
hes
e
blo
c
ks
repre
sents
m
otion
ve
ct
or
s,
the
c
rite
ria
of
si
m
il
arities
us
ed
is
m
ini
m
um
su
m
of
abs
olu
te
di
ff
e
ren
c
es
(SAD
).
T
hus
each
inter
sl
i
ce
is
rep
la
ced
by
tw
o
m
ot
ion
fiel
ds
a
s sho
ws
in
Fig
ur
e
3(a)
and
re
sidu
al
sli
ce.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
A
AS
IS
S
N:
22
52
-
8938
Lo
ssless
4D M
edical
Ima
ges C
ompressi
on
Using A
daptiv
e I
nter S
li
ces
F
il
te
ring
(Lei
la
Bel
hadef
)
365
(a)
(b)
Figure
3. Moti
on f
ie
lds:
(a)
predict
ed
b
l
ock
with
MC
T
run
c 5
/3
, (b
) pro
pose
d
a
ppro
ac
h wit
h 1:
unfilt
ered
bl
oc
k(
E
1
),
2:
pr
e
dic
te
d
bl
ock w
it
h
tru
ncated
5/3 l
ifti
ng transf
or
m
(
E
2
), 3:
predict
ed bloc
k wit
h
m
ot
ion
c
om
pen
sat
ed
(
E
3
)
In
ou
r
sc
hem
e
each
i
nter
sli
ce
(to
b
e predict
e
d)
is divi
ded
in
bl
ock
a
nd
eac
h
blo
c
k
is
ei
the
r
unfilt
ere
d
or
filt
ere
d
by
one
of
fo
ll
owin
g
predict
io
n
te
c
hn
i
qu
e
s:
T
runc
at
ed
5/3
li
fting
trans
form
or
m
otion
c
om
pen
s
at
ion
.
This
filt
ering c
an be
represe
nted by th
ree
f
oll
ow
i
ng pre
dicti
on er
rors:
The u
nf
il
te
red
blo
c
k
S
k
[
m,n
]
:
1
[
,
]
=
[
,
]
(2)
The
tr
uncat
ed
5/3
li
ftin
g
t
ransform
(
without
m
otion
co
m
pen
sat
io
n)
:
2
[
,
]
=
[
,
]
−
⌊
0
.
5
×
(
−
1
[
,
]
+
+
1
[
,
]
)
⌋
(3)
The
m
otion
c
om
pen
sat
ed
pre
dicti
on
:
3
[
,
]
=
[
,
]
−
−
1
[
−
,
−
]
(4)
Wh
e
re
(
d
m
, d
n
)
:
m
otion
vecto
r
of
blo
c
k
S
k
[
m
, n
]
t
o
a
posit
ion
in
S
k
-
1
.
Be
tween
the
th
ree
pr
e
dicti
on
error
s
the
m
ini
m
u
m
is
cho
se
n
for
determ
ining
t
he
pr
e
dicti
on
e
rror
of
thi
s
blo
c
k
S
k
[
m,n
]
a
nd it
s pred
ic
ti
on m
et
ho
d.
The
pre
dicti
on
er
r
or
E
1
re
pr
e
sents
the
un
filt
ered
bl
ock
S
k
[
m,n
]
,
the
predi
ct
ion
er
ror
E
2
represe
nt
s
pr
e
dicti
on
ste
p
in
5/
3
li
ftin
g
trans
f
or
m
and
t
he
predict
io
n
e
rror
E
3
is
the
di
ff
ere
nce
bet
w
een
blo
c
k
S
k
[
m
,n
]
a
nd
it
m
os
t
si
m
il
ar
blo
c
k
in
sli
ce
S
k
-
1
as
sh
ow
s
i
n
Figure
3(
b)
,
t
he
m
otion
vect
or
(
d
m
,
d
n
)
repres
ents
the
dis
plac
e
m
ent
wh
ic
h
al
lo
ws m
ini
m
iz
ing
E
3
.
So
each
i
nter
sli
ce
is
re
placed
by
on
e
m
otion
fiel
d
(forwar
d)
a
nd
/
or
e
ti
qu
et
te
(t
o
in
dicat
e
w
hic
h
pr
e
dicti
on er
ror
is
us
e
d)
a
nd
r
esi
du
al
sli
ce.
Figure
4. 3 l
ev
el
s m
otion
com
pen
sat
ed
tr
uncat
ed
5/
3
li
ftin
g fil
te
r
f
or
G
O
S w
it
h 1
6 sl
ic
es
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2252
-
88
14
IJ
A
AS V
ol
.
7
,
No.
4
,
Decem
ber
2018
:
361
–
368
366
Figure
4
pr
e
se
nts
3
deco
m
posit
ion
le
vels
of
MC
Tru
nc
5/
3
f
or
GOS
wit
h
16
sli
ces;
the
nu
m
ber
of
ob
ta
ine
d sl
ic
es
of
MC
T
runc
5/3 i
s t
he sa
m
e
to
ou
r
s
chem
e w
hic
h
is
c
om
po
se
d by
2 re
fere
nce
sli
ces (
1,
9) a
nd
14 r
esi
du
al
sli
ces (R
1,
R
2,
…
R14),
t
he
sli
ce
(17) r
e
pr
e
sents
the
first r
e
fere
nce slic
e in
the
n
e
xt GOS
.
Howe
ver
for
m
ot
ion
co
ding,
the
a
ppr
oach
with
MC
T
run
c
5/3
pr
oduces
tw
o
m
otion
fiel
ds
a
nd
ou
r
appr
oach
on
e
m
ot
ion
fiel
d
(forwar
d)
an
d
et
iqu
et
te
f
or
eac
h
predict
e
d
bloc
k
by
E
3
an
d
only
et
iqu
et
te
for
eac
h
pr
e
dicte
d
blo
c
k by
E
1
or
E
2
.
3
.
3
.
Re
ference
a
n
d residu
al
sli
ces, motion
vectors
and e
t
iquette
s c
od
in
g
The
obta
ined
ref
e
ren
ce
an
d
resid
ual
sli
ces
com
po
se
d
G
OS
a
re
c
oded
by
3D
S
PIH
T
[
13
]
(S
et
Partit
ion
in
g
in
Hierarc
hical
Tr
ees)
an
d
gen
e
r
at
ed
bit
stream
is
cod
e
d
by
arit
hm
etic
cod
er
,
3D
SPIHT
is
use
d
as
in
vid
e
o
co
ding
an
d
volum
etr
ic
im
age
co
di
ng,
it
is
a
bit
plan
c
od
e
r
an
d
it
pe
rm
it
s
to
re
duce
data
s
iz
e
by
exp
l
oiti
ng the
inter
de
pe
nd
e
nc
es of
sub
bands
in th
ree
dim
ension
s
,
an
d
i
nter
sli
ces.
Co
nc
ern
i
ng the
m
oti
on
vecto
rs
a
nd eti
qu
et
te
s a
re c
od
ed by arit
hm
et
i
c co
der.
4.
E
X
PERI
MEN
TAL RES
UL
TS
We
te
ste
d
t
he
pro
po
se
d
c
ompressi
on
m
et
h
od
with
five
na
ti
ve
4D
m
edi
cal
i
m
ages
CT
(
Com
pu
te
d
Tom
og
ra
ph
y
)
of
he
art
from
two
re
fe
ren
ces
[14][
15
]
s
how
n
i
n
Fig
ur
e
5
.
Each
one
is
c
om
po
sed
by
10
vo
l
um
es
and
th
e
s
patia
l
res
olu
ti
on
of
sli
ces
is
512×
512
c
od
e
d
on
16bits
pe
r
pi
xel.
H
oweve
r
the
vo
l
um
es
of
ea
ch
4D
m
edical
i
m
age
ha
ve
diff
e
re
nt
siz
es:
data
of
re
fer
e
nce
[1
4]
Im
age1
(14
1
sli
ces/
volum
e),
Im
age2
(
16
9
sli
ces/
vo
lum
e)
and
Im
age3
(
170
sli
ces/
volu
m
e)
data
of
re
f
eren
ce
[
15]
Im
age4
(
136
sli
ces/
vo
l
um
e)
and
Im
age5
(12
0
sli
ces/
volum
e).
Figure
5. First
sli
ces of each
4D m
edical
i
m
a
ge
We
c
om
par
e
th
e
ex
per
im
ental
res
ults
of
ou
r
appr
oach
w
it
h
MC
Trunc 5
/3 ap
pr
oach
(3D
SP
I
HT
wit
h
MC
Tr
unc 5
/
3) and
3D SP
IHT with
ou
t m
oti
on
c
om
pen
sat
ion. F
or
all
thes
e appr
oach
es
GOS are
c
om
po
se
d b
y
16 sli
ces f
or
m
ed
as
sho
ws
in
Figure
2 (b).
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
A
AS
IS
S
N:
22
52
-
8938
Lo
ssless
4D M
edical
Ima
ges C
ompressi
on
Using A
daptiv
e I
nter S
li
ces
F
il
te
ring
(Lei
la
Bel
hadef
)
367
Af
te
r
eac
h
G
O
S
is
tra
ns
f
or
m
ed
with
3 l
e
vels
2D
inte
ger
wa
velet
tra
ns
f
orm
in
s
patia
l
dir
ec
ti
on
s
(
,
)
we
us
e
5/
3
filt
er.
I
n
i
nter
sli
ces
di
recti
on
of
GOS
our
ap
pro
ach
us
es
the
propose
d
filt
er
e
q
uatio
ns
(
2,
3
and
4)
,
the
MC
Tr
unc
5/3
a
ppr
oach
use
s
eq
uatio
n
(
1)
with
al
so
t
h
r
ee
le
vels
of
de
com
po
sit
ion
for
both
.
Howe
ve
r
3D
SPIHT
with
out
m
otion
c
om
pen
sat
io
n
us
es
i
n
inte
r
sli
ces
di
recti
on
1D
int
eger
wa
velet
tr
ansfo
rm
with
thre
e
le
vels
of
decom
po
sit
ion
of
t
he
sam
e
us
ed
filt
er
in
s
patia
l
directi
ons
(
5/3
)
.
T
he
ob
ta
in
ed
s
ubba
nds
of
th
ree
appr
oach
es
are
coded
w
it
h 3
D
SP
I
HT
.
The
li
ftin
g
sch
e
m
e
is
ver
y
ap
plied
in
m
edical
i
m
age
com
p
ressio
n
with
tr
aditi
onnel
filt
ers
as
5/3
but
oth
e
r
ne
w
filt
ers
ca
n
be
te
ste
d
i
n
sp
at
ia
l
dir
ect
ion
s
[
16
]
.
T
he
m
otion
vector
s
a
re
determ
i
ned
by
blo
c
k
m
ot
ion
co
m
pen
sat
ion
m
od
el
with
fu
l
l
search
m
et
hod,
the
searc
h
is
eff
ect
uated
with
bloc
k
siz
e
of
16×
16
pix
el
s
in
ou
r
appr
oach
a
nd
MC
Tr
un
c
5/
3
a
ppr
oac
h.
T
he
T
able
1
li
s
ts
the
obta
ine
d
a
ve
rag
e
bit
rate
for
t
he
lo
ssless
com
pr
essio
n u
sing t
w
o
fi
rst volum
es o
f
eac
h 4
D
m
edical
im
age.
Table
1.
4D L
ossl
ess Com
pr
es
sion R
esults
in
bit pe
r pixel
(bpp)
4
D M
ed
ical
I
m
ag
e
s
3
D SPI
H
T
MC T
run
c 5/3
Prop
o
sed
M
eth
o
d
I
m
ag
e1
5
.19
5
.23
5
.10
I
m
ag
e2
5
.92
5
.97
5
.89
I
m
ag
e3
5
.39
5
.45
5
.36
I
m
ag
e4
5
.03
5
.06
4
.98
I
m
ag
e5
4
.53
4
.55
4
.35
The re
su
lt
s
s
how o
ur
m
et
ho
d
ou
t
perform
s
the
oth
e
r m
et
ho
ds
for a
ll
te
st
i
m
ages.
3D
SP
IHT
giv
es
lo
w
aver
a
ge
bit
rate
tha
n
MC
Tr
unc
5/3
of
0.7
6%
,
th
us
t
he
i
nt
egr
at
io
n
of
th
e
m
otion
c
om
pen
sat
ed
tr
un
cat
ed
5/3
li
fting
filt
er
i
n
3D
SP
IHT
penal
ise
s
the
com
pr
essi
on
rate
i
n
MC
T
runc
5/
3
a
ppr
oach.
H
ow
e
ve
r
the
pro
po
s
ed
inter
sli
ces
filt
er
im
pr
ov
e
s
th
e
com
pr
essio
n
rate
(lo
w
bit
ra
te
)
with
ave
ra
ge
of
1.45
%
co
m
par
ed
to
3D
SPIHT
,
as w
el
l a
net i
m
pr
ov
em
ent is reac
hed of th
e
ord
e
r
of 2.2%
com
par
ed
t
o
M
C Tr
un
c
5
/
3.
The
a
ver
a
ge
im
pro
vem
ent
by
pro
po
se
d
te
c
hniqu
e
c
om
par
ed
to
3D
SP
I
HT
is
ob
ta
i
ned
by
usi
ng
im
age1
thu
s
we
us
e
it
in
f
ollow
i
ng
te
st.
The
e
xp
eri
m
ental
resu
lt
s
pr
ese
nted
in
T
able
2
a
re
ob
t
ai
ned
by
te
sti
ng
fi
ve
diff
e
re
nt sizes
of
im
age1
, thu
s li
ne 2
r
e
pr
ese
nts lo
ssless c
om
pr
ession res
ul
ts of 2
first
volum
es o
f
im
age1
,
li
ne
3
of 4
fir
st
volum
es
of
Im
age1
an
d
s
o
on
to the
la
st
li
ne
of 1
0
first v
ol
um
e
s
of
im
age1
,
t
he
se
dif
fer
e
nt
si
zes
of
Im
age1
p
e
rm
it
to
obta
in
dif
fere
nts
GO
S
b
y
re
or
ga
nizing t
he sl
ic
es as in Fi
gure
2 (
b).
Table
2.
4D L
ossl
ess Com
pr
es
sion R
esults
in
bit pe
r pixel
(bpp) fo
r dif
fer
e
nt sizes
of I
m
a
ge1
4
D M
ed
ical
I
m
ag
e
s
3
D SPI
H
T
MC T
run
c 5/3
Prop
o
sed
M
eth
o
d
2
Vo
lu
m
es
I
m
ag
e1
5
.19
5
.23
5
.10
4
Vo
lu
m
es
I
m
ag
e1
5
.08
5
.10
4
.99
6
Vo
lu
m
es
I
m
ag
e1
5
.07
5
.05
4
.96
8
Vo
lu
m
es
I
m
ag
e1
4
.98
5
.00
4
.92
1
0
Vo
lu
m
es I
m
ag
e
1
4
.99
4
.96
4
.90
The
obta
ine
d
c
om
pr
essio
n
rate
by
pro
posed
m
et
ho
d
is
bette
r
tha
n
ot
her
s
chem
es
fo
r
al
l
i
m
ages
te
st.
W
e
no
te
that
t
he
bi
t
rate
is
re
duce
d
for
al
l
com
pressi
on
sc
hem
e
s
w
hile
t
he
nu
m
ber
of
te
m
po
ral
sli
ces
(at
diff
e
ren
t
tim
e
and
with
t
he
sam
e
posit
ion
)
com
po
s
e
d
G
OS
incre
ase.
The
re
su
lt
s
s
how
a
var
ia
ti
on
of
the
com
pr
es
sion
rate
as
a
f
un
ct
i
on
of
t
he
num
ber
of
volum
es.
The
im
pr
ove
m
ent
of
this
ra
te
by
our
a
ppr
oach
is
of
t
he
order
of
1.73%
an
d
1.8
5%
com
par
e
d
to
3D
SP
IHT
and
MC
T
runc
5/3
re
sp
ect
ive
ly
,
wh
il
e
it
does
no
t
e
xcee
d
0.1
1%
betwee
n
the
lat
te
r
tw
o
m
et
ho
ds.
5.
CONCL
US
I
O
N
In
this
pa
per
w
e
pro
pose
d
a
ne
w
a
ppr
oac
h
of
lo
ssless
c
ompressi
on
of
4D
m
edical
i
m
ag
es,
it
c
onsist
s
in
c
onstr
uction
of
GOS
a
nd
ea
ch
sli
ces
c
om
po
se
d
G
OS
is
de
correla
te
d
in
s
pa
ti
al
directi
ons
(
,
),
after
pro
po
s
ed
inter
sli
ces
filt
erin
g
is
perf
or
m
ed
with
m
ini
m
um
of
th
ree
pre
dicte
d
e
rrors
f
or
eac
h
blo
c
k
c
om
po
sed
inte
r
sl
ic
es.
The
th
ree
pre
dicti
on
e
rror
s
are
re
pr
ese
nte
d
by
un
filt
ered
bl
ock,
tru
nc
at
ed
5/3
li
ftin
g
filt
er
an
d
m
ot
ion
com
pen
sat
ed
t
he
ai
m
is
to
reduce
the
siz
e
of
resid
ual
sli
ces.
The
obta
ine
d
sl
ic
es
are
co
ded
with
3D
SP
I
H
T.
The
pro
po
se
d
a
ppr
oach
pr
ov
i
des
lossless
c
om
pr
ession
im
pr
ove
m
ents
by
reducin
g
th
e
bit
r
at
e
s
com
par
ed
to
3D
SPIHT
with
MC
trun
c
5/3
and
3D
SP
IHT
with
ou
t
m
otion
c
om
pen
sat
ion
,
t
his
en
ha
ncem
ent
is
due
to
the
integrati
on
of
the
three
predi
ct
ion
te
ch
niqu
es
in
inter
sli
ces
filt
er.
As
f
ut
ur
e
wor
k,
our
ap
proach
offe
rs
th
e
po
s
sibil
it
y
to
obta
in
l
os
sy
to
lossless
dec
od
i
ng
by
ex
plo
it
in
g
inte
r
sli
ces
sc
al
abili
ty
al
so
ot
her
4D
m
edical
i
m
age
m
od
al
it
ie
s can
be
te
ste
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2252
-
88
14
IJ
A
AS V
ol
.
7
,
No.
4
,
Decem
ber
2018
:
361
–
368
368
REFERE
NCE
S
[1]
V.
Sanchez,
P.
Nasiopoulos,
R.
Abugharbi
eh
,
"Loss
le
ss
Com
pr
ession
of
4D
Medical
Im
age
s
u
sing
H.264/
AV
C,
"
IEE
E
Inte
rnat
io
nal
Con
f
ere
nce
on
Ac
oust
ic
s,
Sp
ee
ch
,
and
Signal P
roce
ss
ing
,
Tou
louse,
2006
,
pp
.
1116
-
1119.
[2]
V.
Sanchez,
P.
Nasiopoulos,
R.
Abugharbi
eh
,
"
Eff
icient
lossles
s
compress
ion
of
4D
m
edi
c
al
i
m
age
s
base
d
on
th
e
adva
nc
ed
v
ide
o
codi
ng
sch
eme,
"
IEEE
Tr
ans.
Inf
orm
ati
on
Techn
o
logy
In
Bi
omed
ic
in
e
,
V
ol
.
12,
N
O.
4,
JU
LY
200
8,
pp.
132
-
138
.
[3]
U.
Marti
n
,
A.
Ka
up
,
"
Ana
ly
sis
o
f
compress
ion
of
4
D
vol
umetri
c
me
dic
al
image
data
sets
using
mult
i
-
vi
ew
(
MVC)
vi
de
o
codi
ng
me
thods,
"
M
at
hemat
ic
s
o
f
Data/Im
age
Pa
tt
ern
Rec
ogn
it
io
n,
Com
pre
ss
ion
and
En
cr
y
pti
on
with
Appli
ca
t
ion
s
XI,
Proc.
of
SP
I
E
.
Vol
.
7075
,
70
757,
2008.
[4]
Y.
Feng,
H.
Lu
,
X.
Z
eng,
"
A
Frac
tal
Im
age
Com
pre
ss
ion
M
et
hod
Based
on
Multi
-
W
ave
l
et
",
TEL
KOMNIKA
(
Tele
communic
ati
on,
Computing, E
l
ec
troni
cs
and
Control)
,
Vol.
13
,
No.3
,
Sep
te
m
ber
2015,
pp.
996
-
1005
.
[5]
Y.
Habc
hi
,
M.
B
el
adgh
am,
A.
A.
Ta
l
eb
,
"
RGB
Medic
a
l
Video
Com
pre
ss
ion
Usi
ng
Geom
et
ri
c
W
ave
le
t
and
SP
IHT
Coding
",
In
te
rn
ati
onal
Journal
of
Elec
tric
al
an
d
Computer
Eng
ine
ering
(
IJE
CE
)
,
Vol.
6,
No.
4
,
Augus
t
2016,
p
p.
1627
-
1636
.
[6]
H.
G.
L
al
gudi
,
A.
B
il
gin
,
M.
W
.
Marc
ellin,
A.
Tabesh,
M
.
D
.
Nad
ar
and
T.
P.
Trou
ard
,
"
Four
-
dime
nsional
compress
ion
of
fM
RI
using
JPEG2000,
"
Medi
c
al
Im
agi
ng
2005
:Image
Proc
essing,
Proc
.
of
SP
IE
.
Vol
.
5747
.
[7]
A.
A.
Kass
im,
P.
Yan,
W
.
S.
Le
e
,
K.
Sengupta,
"M
oti
on
Com
pensa
t
ed
Loss
y
-
to
-
Los
sless
Com
pre
ss
i
on
of
4D
Medi
cal
Im
age
s
Us
ing
In
te
ger
W
ave
l
et
T
ran
sform
s,"
IEEE
Tr
ans.
Information
Te
chnol
og
y
In
Bi
omed
ic
in
e
,
VO
L
.
9
,
NO
.
1,
2005,
pp
.
132
-
1
38.
[8]
J.
-
R.
Ohm
,
"Thr
ee
-
dimensiona
l
s
ubband
cod
ing
w
it
h
m
oti
on
comp
ensa
ti
on
,
"
IE
EE
Tr
ans.
on
Image
Proce
ss
ing
,
199
4,
pp.
559
-
571
.
[9]
S.
-
J.
Choi
and
J
.
W
oods,
"M
oti
o
n
-
compensat
ed
3
-
d
subband
cod
ing
of
vid
eo,
"
I
EE
E
Tr
ans.
on
I
mage
Proce
ss
in
g
,
1999,
pp
.
155
-
1
67
.
[10]
A.
Seck
er
and
D.
T
aubman,
"M
ot
ion
-
compensat
e
d
highly
sc
al
ab
le
vide
o
compress
ion
using
an
ada
p
ta
ti
v
e
3d
wav
ele
t
tra
nsform
base
d
on
li
f
ti
ng,
"
IEEE
,
2001
.
[11]
L.
Luo,
J.
Li,
S.
Li,
Z.
Zhu
ang,
and
Y
-
Q
.
Zha
ng
,
"
Mo
ti
on
-
compe
nsated
l
if
t
ing
wa
ve
l
et
and
i
ts
app
li
cation
in
vi
de
o
codi
ng,
"
in
IEEE
Int
ern
ationa
l Confere
nc
e
on
Multi
m
edi
a
and
Expo
,
Augus
t
20
01.
[12]
I.
Daube
chi
es
an
d
W
.
Sw
el
dens,
"F
ac
tori
ng
wav
e
le
t
tra
nsform
s
int
o
li
f
ti
ng
steps
,
"
Journal
of
Fouri
er
Anal
ysis
and
Appl
ic
a
ti
ons
,
19
98.
[13]
B.
J.
Kim
and
W
.
A.
Pear
lman
,
"An
Embe
dd
ed
Wav
elet
Vi
deo
Coder
Us
ing
Thr
ee
-
Dimensional
Set
Parti
ti
on
ing
in
Hierar
chi
cal Tr
ee
s (
SPIHT
)
,
"
In
I
EE
E
Data Com
pre
ss
ion
Confer
en
ce
,
1997,
pp.
22
1
-
260.
[14]
J.
Vand
emeule
b
rouc
ke,
S.
Rit
,
J
.
K
y
bic,
P.
C
la
r
y
s
se,
and
D.
Sarrut
,
"S
pat
iotem
pora
l
m
oti
on
esti
m
at
i
on
for
r
espirator
y
-
cor
relate
d
imagi
ng
of
th
e lungs,"
In
M
ed
Ph
ys
,
20
11,
pp
.
166
-
178
.
[15]
htt
p://m
ida
s.ki
twar
e.com/comm
unity
/view/47
.
[16]
A.
Haz
ar
at
h
ai
ah
,
B.
Prabha
k
ara
Rao,
"
Medi
ca
l
I
m
age
Com
pre
ss
ion
using
Li
ft
ing
base
d
New
W
a
vel
e
t
Tr
ansform
s
"
,
Inte
rnational
Jo
urnal
of El
e
ct
ri
c
al
and
Comput
er
Engi
n
ee
ring
(
IJE
CE)
,
Vol.
4,
No.
5
,
Octob
er
20
14,
pp
.
741
-
750
.
Evaluation Warning : The document was created with Spire.PDF for Python.