Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
Vol
.
4
,
No
. 3,
J
une
2
0
1
4
,
pp
. 38
9~
39
7
I
S
SN
: 208
8-8
7
0
8
3
89
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJECE
Retin
a
l Blood Vess
els E
x
tracti
on
Bas
e
d on Curvel
et T
r
ans
f
orm
and by Combining Bothat
and Tophat Morphology
K. G
a
yathri,
D.
N
a
rmadh
a,
K. T
h
ilagavat
hi, K.
P
avit
h
r
a
,
M.
Pr
adeep
a
Departement of
Electronics and
Communi
cation Engineering,
C. Abdul
Hakeem
College of
Engineer
ing
and
Technolog
y
Vellore, Ind
i
a
Article Info
A
B
STRAC
T
Article histo
r
y:
Received
Mar 12, 2014
Rev
i
sed
May
6, 201
4
Accepted
May 25, 2014
Retinal image
contains vital inf
o
rmation
about
the health of th
e
sensor
y
par
t
of the vis
u
al s
y
s
t
em
. Extr
act
ing t
h
es
e feat
ures
is
the firs
t and m
o
s
t
im
portant
s
t
ep to
anal
ys
is
of ret
i
nal
im
a
g
es
fo
r various
applications of
medical or
human recognition. Th
e proposed method
c
onsists of prep
rocessing, con
t
ras
t
enhancement an
d blood vessels extraction
stag
es
. In preprocessin
g
, since th
e
green chann
e
l f
r
om
the coloured retin
al im
ages
has
the highes
t
contras
t
between th
e subbands so the green com
ponent
is selected. To
uniform the
brightness of image ad
aptiv
e histogram
equalization is used since it provides
an image with
a
uniformed, d
a
rk
er bac
kground
and brighter gr
ey
level of
the
blood vessels. Furthermore Curvelet transforms is used to e
nhance th
e
contrast of
an
image b
y
hig
h
lighting
its ed
ges in various
scales and
directions.
Ev
entually
the combination
of Both
at and
Tophat
morpholological function follo
w
ed b
y
lo
cal thresholding is provided to
classif
y
th
e bloo
d vessels. Hence the re
tin
al bloo
d vessels are sep
a
rated from
the b
ackground
image.
Keyword:
B
l
oo
d vessel
s
ext
r
act
i
o
n
Curvelet transform
M
o
r
p
h
o
l
o
gi
cal
f
unct
i
o
n
Pre
p
rocessing
Retinal im
age
Copyright ©
201
4 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
K. Gay
a
th
ri
Depa
rt
em
ent
of El
ect
r
oni
cs
a
n
d
C
o
m
m
uni
cat
i
on E
n
gi
nee
r
i
n
g
,
C
.
A
b
d
u
l
Ha
ke
em
C
o
l
l
e
ge o
f
En
gi
neeri
n
g
a
n
d Tec
h
nol
ogy
,
Vel
l
o
re
, I
n
di
a.
Em
a
il: kgayathri.be
.
ece@gm
a
il.com
1.
INTRODUCTION
On
e
of th
e m
o
st i
m
p
o
r
tan
t
intern
al co
m
p
onen
t
s of eye is called
retin
a,
wh
ich
cov
e
rs
all p
o
s
terior
co
m
p
art
m
en
t. An
y
d
a
m
a
g
e
in
retin
a lead
s to
sev
e
re
dis
eases. Disorde
r
s
in retina re
sulted from
s
p
ecial
di
seases a
r
e
di
agn
o
se
d
by
sp
eci
al
im
ages whi
c
h a
r
e
obt
ai
n
e
d
by
usi
n
g
o
p
t
i
c
im
agi
ng c
a
l
l
e
d Fu
n
dus
i
m
ag
e
.
Th
e Fundu
s i
m
ag
es are u
s
ed
fo
r d
i
agn
o
s
is by train
e
d
clin
ician
s
to
ch
eck
fo
r an
y abno
rm
alities o
r
an
y c
h
ange
i
n
t
h
e
ret
i
n
a.
They
are
capt
u
red
by
usi
n
g s
p
eci
al
de
vi
ces
cal
l
e
d o
pht
hal
m
oscopes. Eac
h
pi
xel
i
n
t
h
e
fu
n
d
u
s
im
age consi
s
t
s
of
t
h
ree
val
u
e
s
nam
e
l
y
R
e
d, Gree
n a
n
d B
l
ue, eac
h
val
u
e
bei
n
g
qua
nt
i
z
ed t
o
2
5
6
l
e
ve
l
s
. The
bl
o
od
vessel
s
are t
h
e i
m
port
a
nt
part
s
of t
h
e
ret
i
n
al
im
ages consi
s
t
i
n
g o
f
art
e
ri
es an
d art
e
ri
ol
es. C
h
e
c
ki
ng t
h
e
obtaine
d cha
n
ges in retinal im
ages in an especial peri
od can hel
p
the
physician
to diagnose the disease.
Ap
pl
i
cat
i
ons
o
f
ret
i
n
al
im
ages are di
ag
nosi
ng t
h
e
pr
og
res
s
of som
e
cardi
ovasc
ul
ar di
se
ases, di
ag
n
o
si
ng t
h
e
regi
on
wi
t
h
no
bl
o
od
vessel
s
(M
acul
a
)
,
usi
n
g suc
h
i
m
ages i
n
bi
om
et
ri
c appl
i
cat
i
ons a
nd
i
n
hel
p
i
ng a
u
t
o
m
a
t
i
c
laser s
u
rgery
on eye, etc.
On t
h
e
ot
he
r h
a
nd
, ext
r
act
i
n
g
t
h
e ret
i
n
al
bl
o
od
vessel
s
i
s
d
one i
n
s
o
m
e
cases by
phy
si
ci
a
n
m
a
nual
l
y
,
whic
h is
diffic
ult and tim
e consum
ing a
n
d is
accom
p
anie
d
by hi
gh m
i
stakes due t
o
m
u
ch
de
pende
n
ce
on the
physicians skill level. So the
exact ex
traction of the
blood
vessels from
t
h
e retinal im
ages necessitates usi
ng
al
go
ri
t
h
m
and i
n
st
rum
e
nt
s whi
c
h re
duce
t
h
e de
pen
d
e
n
c
y
on t
h
e f
u
nct
i
on a
nd el
i
m
i
n
at
e t
h
e erro
r f
act
ors
.
While ca
pturing the
im
age because
of
the
varia
b
ility of light reflection
coef
ficient in
diffe
re
nt pa
rts
of t
h
e
retin
al layer al
so
du
e to
th
e d
e
fects in
im
a
g
ing
syste
m
s t
h
ere
o
ccurs non
un
ifo
r
m
il
lu
min
a
tio
n
in
the retin
al
im
age, pi
xel
s
rel
a
t
e
d t
o
t
h
e bl
o
od
vessel
s
cann
o
t
be cl
assi
fi
ed caref
ul
l
y
.Thi
s im
pro
p
e
r cont
rast
i
s
due
t
o
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
4, No
. 3,
J
u
ne 2
0
1
4
:
38
9 – 3
9
7
39
0
di
ffe
re
nt
vesse
l
s
have di
f
f
ere
n
t
cont
rast
;
art
e
ri
es have
hi
gher co
n
t
rast th
an
v
e
in
s. In
ad
ditio
n
to
th
is, p
r
esen
ce
of
n
o
i
s
e,
fo
vea
an
d
opt
i
cal
di
sk,
wi
dt
h
of
t
h
e vessel
s
,
ef
fec
t
s of
l
e
si
on
s a
n
d
pat
h
ol
o
g
i
cal
chan
ges
sh
o
u
l
d
al
s
o
be c
o
nsidere
d
.
So, for ext
r
action of
blood
ve
ssels with hi
gh accuracy
,
we
need of a
n
e
ffe
ctive algorithm.
In th
e
p
r
op
osed
algorith
m
,
th
e fo
cu
s will
b
e
on
th
e ex
tractio
n of
b
l
ood
v
e
sse
ls con
s
titu
tes of
d
i
g
ital
co
lored im
ag
es of
retin
a as its in
pu
t
wh
ich
is th
en co
nv
ert
e
d t
o
gree
n c
h
a
nnel
i
m
age wi
t
h
best
c
ont
rast
.
Si
nce
t
h
e pre
p
r
o
cess
i
ng p
h
ase pl
ay
s an im
port
a
nt
rol
e
i
n
fi
na
l ex
traction
results. On
e o
f
th
e ad
v
a
n
t
ag
es of th
is
pha
se i
s
by
ap
pl
y
i
ng t
h
e a
d
a
p
t
i
v
e hi
st
og
ra
m
equal
i
zat
i
o
n
and C
u
r
v
el
et
t
r
ans
f
orm
on t
h
e im
age t
o
red
u
ce t
h
e
noi
se
an
d i
m
pr
ove
t
h
e
co
nt
ra
st
. The
r
ef
ore t
h
e ina
d
e
q
uacy of
pre
v
ious m
e
thods is
res
o
lved. Si
nce t
h
e
blood
vessel
s
are
di
st
ri
b
u
t
e
d i
n
di
ffe
rent
di
rect
i
o
n
s
,
appl
y
i
n
g
m
o
rp
hol
ogi
cal
o
p
e
r
at
i
on cause
s t
h
e bl
o
od
vessel
s
wi
t
h
hi
g
h
acc
uracy
t
o
be
sepa
rat
e
d f
r
o
m
t
h
e backg
r
ou
n
d
an
d
fi
nal
l
y
t
h
e co
nnect
e
d
c
o
m
pone
nt
s
wi
t
h
d
e
fi
ne
d
th
resh
o
l
d
,
frills in
th
e im
ag
e are rem
o
v
e
d
and
ex
tracted b
l
oo
d v
e
ssels are
o
b
t
ain
e
d
.
2.
LITERATU
R
E
REVIE
W
Recently
m
a
n
y
autom
a
ted detection techni
que
s are
const
a
ntly devised
and
im
ple
m
ented to help
op
ht
hal
m
ol
ogi
st
s det
ect
bl
o
o
d
vessel
s
by
a
p
pl
y
i
ng i
m
age p
r
oces
si
n
g
a
n
d
pat
t
e
rn
rec
o
gni
t
i
on t
ech
ni
ques
.
In 201
2, M.
Kalaiv
an
i, M. S. Jeyalak
s
h
m
i an
d Ap
a
r
na
.V
[
6
]
use
d
A
d
apt
i
ve
Hi
st
o
g
ram
Eq
ual
i
zat
i
o
n
for in
itial en
han
cem
en
t, fo
llo
wed
b
y
th
is th
e curvelet tran
sfo
r
m
s
to
th
e eq
u
a
lized
im
a
g
e and
th
e cu
rv
elet
coefficients are obtai
ned. T
h
e vessel e
x
trac
tion is don
e
b
a
sed
on t
h
res
h
ol
di
n
g
t
e
c
hni
q
u
e an
d t
h
e Ki
r
s
ch’s
te
m
p
lates. It in
vo
lv
es sp
atial filterin
g
of the i
m
ag
e
u
s
ing th
e tem
p
lates
in
eigh
t d
i
fferen
t
orien
t
ations. Th
e
mask
in
g of
r
e
du
nd
an
t r
e
g
i
on
s in
th
e ob
tain
ed
o
u
t
p
u
t
im
ag
e is carr
i
ed
ou
t
u
s
ing
bo
und
ary tech
n
i
qu
es.
In
ot
he
r rel
a
t
e
d w
o
r
k
, M
a
r
w
an D.
Sal
e
h an
d C
.
Eswa
ra
n [5]
p
r
o
p
o
sed t
h
e al
g
o
ri
t
h
m
has em
pl
oy
e
d
t
echni
q
u
es
, su
ch as bac
k
g
r
ou
nd
rem
oval
,
co
nt
rast
en
ha
nce
m
ent
,
h-m
a
xi
m
a
t
r
ansfo
r
m
a
ti
on, t
h
res
h
ol
di
ng
, et
c.
Aft
e
r
co
n
v
ert
i
ng
t
h
e
R
G
B
i
m
age t
o
gray
-s
cal
e, b
o
t
h
m
o
rph
o
l
o
gi
cal
t
o
p
-
hat
a
n
d
bot
t
o
m
-
hat
t
r
ansf
or
m
s
have
been
e
xpl
oi
t
e
d
t
o
per
f
o
rm
t
h
e co
nt
rast e
nhancem
ent. Other techniques
s
u
ch
a
s
h
-
ma
x
i
ma
t
r
a
n
s
f
o
r
m a
n
d
m
u
l
tilev
e
l th
resh
o
l
d
i
ng
h
a
v
e
b
een ex
p
l
o
ited to
d
ecrease
t
h
e in
ten
s
ity levels as m
u
ch
as po
ssib
l
e to
facilitat
e
t
h
e t
h
resh
ol
d
s
e
l
ect
i
on f
o
r
bi
n
a
ri
zat
i
on i
n
20
12
.
Iqbal, M.I et al [14] in 2007
used Col
o
r Space Conve
rsi
o
n, Edge
Zero Paddi
ng, Median Filtering
and
A
d
apt
i
v
e
Hi
st
o
g
ram
Equ
a
l
i
zat
i
on as p
r
e-p
r
oce
ssi
n
g
t
e
chni
que
s an
d t
h
ey
use
d
se
gm
ent
a
t
i
on t
o
g
r
o
up t
h
e
im
age i
n
t
o
reg
i
ons wi
t
h
sam
e
pro
p
e
r
t
y
or charact
e
r
i
s
t
i
c
s. M
e
t
hods
of i
m
age segm
entat
i
on i
n
cl
u
d
e s
i
m
p
l
e
th
resh
o
l
d
i
ng
,
K-m
ean
s Al
g
o
rith
m
an
d
Fu
zzy C-m
ean
s. Since it tak
e
s m
o
re ti
m
e
to
lo
ad th
e
d
a
ta.
An e
fficient re
tinal analysis
m
e
thod
based
on c
u
rv
elet transform
and m
u
lti structure e
l
em
ents wa
s
pr
o
pose
d
by
M
i
ri
et
al
[9]
i
n
2
0
1
1
,
he
descri
bed
t
h
at
gree
n c
h
an
ne
l
of t
h
e
o
r
i
g
i
n
al
col
o
re
d i
m
age wa
s
sel
ect
ed. O
b
t
a
i
n
t
h
e f
u
nd
us
regi
on m
a
sk usi
n
g Ot
s
u
a
l
go
ri
t
h
m
fol
l
o
wed
by
m
o
rp
hol
ogi
cal
cl
osi
ng a
n
d
m
u
l
tip
ly
its re
su
lt i
m
ag
e wit
h
FDCT v
i
a wrapp
i
ng
, th
en
m
odify the curvelet coe
ffi
ci
e
n
t
s
and
obt
ai
n
e
d t
h
e
enha
nce
d
i
m
age. The
n
s
ubt
ra
ct
s t
h
e est
i
m
a
ted bac
k
gr
ound from
the enha
nced im
ag
e. T
h
ere
b
y m
odified top-
h
a
t tran
sform
s
u
s
ing
th
e mu
ltistru
c
ture ele
m
en
ts
m
o
rpho
log
y
were ap
p
lied
and
b
y
p
r
o
v
i
d
i
ng
open
i
ng
fun
c
tion
th
e imag
e were reco
n
s
t
r
u
c
ted
.
In
o
r
d
e
r to
e
limin
ate th
e re
m
a
in
ed
false edg
e
s, ap
p
l
y len
g
t
h
filterin
g
alo
n
g
with
CC
A [2
] lo
cally b
u
t
th
e im
ag
e resu
lted
fro
m
To
pHat fu
n
c
tion can
in
clud
e all n
e
g
lig
ib
le chan
g
e
s
in
th
e
grey levels ex
istin
g in th
e im
ag
e (su
c
h as
n
o
i
se).
Pri
y
a R
et
al
[8]
i
n
2
0
11
u
s
ed p
r
ep
r
o
cess
i
ng t
ech
ni
q
u
es
l
i
k
e Gray
scal
e C
onve
rsi
o
n, A
d
a
p
t
i
v
e
Histog
ram
Eq
u
a
lizatio
n
,
Match
e
d
Filter R
e
spo
n
s
e and
propo
sed
a m
e
th
od
for feat
u
r
e ex
traction
based
on
Area
o
f
o
n
pi
xel
s
, M
e
a
n
a
n
d St
a
nda
rd
D
e
vi
at
i
on.
Al
s
o
i
n
2
0
1
2
,
Jas
p
reet
Kau
r
a
n
d
Dr.
H
.
P.Si
n
h
a [
3
]
p
r
esen
ted a Filter b
a
sed
app
r
o
a
ch
with
m
o
rp
ho
log
i
cal filters is u
s
ed
to
seg
m
en
t th
e v
e
ssels. Th
e
m
o
rp
ho
log
i
cal filter is tu
n
e
d
t
o
m
a
tch
th
at p
a
rt o
f
v
e
ssel to b
e
ex
tracted
in
a g
r
een
ch
ann
e
l i
m
ag
e. To
classify
th
e p
i
x
e
ls in
to v
e
ssels an
d
no
n
v
e
ssels lo
cal th
resho
l
di
n
g
based o
n
g
r
a
y
l
e
vel
co-occ
ur
rence m
a
t
r
i
x
as i
t
cont
ai
ne
d i
n
f
o
r
m
at
i
on o
n
t
h
e
di
st
ri
b
u
t
i
o
n
o
f
gray
l
e
vel
f
r
eq
uency
a
n
d e
d
g
e
i
n
f
o
rm
at
i
on
have
bee
n
p
r
es
ent
e
d.
I
n
201
2, Pai
n
ta
m
ilselv
i
et.al
[
4
] car
r
i
ed ou
t
b
l
ood
v
e
ssels ex
traction
i
n
fiv
e
step
s. First th
e RGB
im
age was conve
rt
ed i
n
t
o
g
r
ay
scal
e. Secon
d
l
y
m
o
rp
hol
ogi
cal
o
p
eni
n
g
and cl
osi
n
g
ope
rat
i
o
n i
s
used t
o
red
u
ce sm
all
noi
se
. In t
h
e
t
h
i
r
d st
ep t
o
obt
ai
n t
h
e
ves
s
el
st
ruct
ure a
uni
q
u
e t
ech
n
i
que cal
l
e
d t
o
p hat
t
r
ans
f
o
r
m
a
ti
on
was
use
d
.
I
n
t
h
e
f
o
urt
h
st
ep, t
h
e
resultan
t
i
m
ag
e was ob
tain
ed
after
b
i
n
a
risatio
n and
t
h
res
hol
di
n
g
.
F
i
nal
l
y
con
n
ect
e
d
c
o
m
pone
nt
a
n
al
y
s
i
s
was
us
ed t
o
o
b
t
a
i
n
an
im
age w
h
i
c
h
was
free
f
r
om
noi
se
.
The re
st
of t
h
e pape
r i
s
o
r
g
a
ni
zed as
f
o
l
l
o
ws:
I
n
Sect
i
o
n
3 p
r
op
ose
d
m
e
t
h
o
d
s i
s
de
sc
ri
be
d w
h
i
l
e
sect
i
on 3.
1.
1
& 3.
1.
2
e
x
am
ines g
r
een cha
nnel
sel
ect
i
on
and i
m
age enh
a
ncem
ent
usi
ng ada
p
t
i
v
e hi
st
og
ram
equal
i
zat
i
o
n, I
n
Sect
i
o
n 3.
1.
3. desc
ri
bes
C
ont
rast
en
ha
nc
em
ent
usi
n
g
F
D
C
T
a
n
d
sectio
n 3.2 pr
esen
ts the
m
e
t
hod
f
o
r e
x
t
r
act
i
o
n
o
f
vess
el
s fr
om
col
o
re
d r
e
t
i
n
al
i
m
age. I
n
sect
i
o
n
4 t
h
e
resul
t
s
of
t
h
e al
go
ri
t
h
m
ov
er a
n
extensi
v
e
datas
e
t are
prese
n
te
d a
n
d conc
l
u
si
ons
are
re
vi
ew
ed i
n
sect
i
o
n
5.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Ret
i
nal
Bl
oo
d
Vessel
s
Ext
r
act
i
on
Base
d
o
n
C
u
rvel
et
Tr
an
sf
o
r
m
an
d
by C
o
mbi
n
i
n
g B
o
t
h
a
t
…
(
K
. G
a
yat
h
ri)
39
1
3.
PROP
OSE
D
METHO
D
The
pr
o
pose
d
sy
st
em
i
n
t
h
i
s
w
o
r
k
c
o
n
s
i
s
t
s
of
f
o
l
l
o
wi
n
g
st
ep
s p
r
e
p
r
o
cessi
ng
an
d
bl
oo
d
vessel
s
ext
r
act
i
o
n.
The
bl
oc
k
di
ag
ram
o
f
ret
i
n
al
bl
oo
d
vessel
s
e
x
t
r
a
c
t
i
on i
s
s
h
ow
n
i
n
Fi
g
u
r
e
1.
Fi
gu
re
1.
B
l
oc
k
di
ag
ram
of re
t
i
n
al
bl
o
o
d
ves
s
el
s ext
r
act
i
o
n
t
echni
q
u
e
3.1. Prepr
o
ces
sing
3.
1.
1.
Green Channel
Selecti
o
n
If t
h
e three cha
nnels
of a RGB coloure
d
retinal im
age are obs
erved, the
red c
h
annel s
h
ows a
poorly
cont
rast
ed
ret
i
n
al
vasc
ul
at
ur
e
on t
op
of t
h
e
cho
r
oi
dal
vasc
ul
at
ure
.
The
G
r
een c
h
a
nnel
s
h
o
w
s
wel
l
con
t
rast
ed
arteries
a
n
d veins with
a
clear dark
f
ovea
i
n
t
h
e ce
nt
re.
The
bl
ue c
h
a
nnel
sh
o
w
s a
noi
si
er
i
m
age of
t
h
e
vasc
ul
at
ure. S
o
t
h
at
gree
n chan
nel
has t
h
e
best
cont
rast
by
expe
ri
ence
i
s
show
n i
n
Fi
gu
re 2.
Hen
ce i
t
i
s
selected
fo
r fu
rth
e
r
wo
rk
.
Fi
gu
re
2.
(a
) R
e
d c
h
an
nel
,
(
b
)
g
r
een
cha
n
nel
and
(c
)
bl
ue c
h
annel
3.
1.
2.
Ad
ap
ti
v
e
Hi
st
ogr
am
E
qual
i
z
ati
o
n
We in
itially
worked
o
n
t
h
e co
lou
r
retinal i
m
ag
e. To
redu
ce th
e effect of d
i
fferen
t lig
h
t
n
i
ng
co
nd
itio
ns and to
un
ifo
r
m
il
lu
min
a
tio
n
Adap
tiv
e h
i
st
o
g
ra
m
Eq
u
a
lizatio
n
is
u
s
ed
. It
is an
enh
a
n
c
e
m
en
t
t
echni
q
u
e
capa
b
l
e
o
f
i
n
crea
si
ng
t
h
e
l
o
cal
C
ont
rast
al
so
i
t
i
m
proves t
h
e
br
i
ght
ne
ss
of
an
im
age. It
di
f
f
er
s f
r
om
Input RGB Imag
e
Green channe
l
s
e
le
ction
Adaptive histogr
am equalization
Contrast
enhan
c
ement using FD
CT
Edge d
e
tection u
s
ing morphological
operation followed b
y
local
thresholding
Prep
ro
ces
s
i
n
g
Blood vessel
extraction
Blood
vessels ex
tracted image
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
4, No
. 3,
J
u
ne 2
0
1
4
:
38
9 – 3
9
7
39
2
ordina
ry histogram
equaliza
tion in the
res
p
ective that adaptive m
e
t
hod com
putes several histograms each
co
rresp
ond
ing
to
d
i
stin
ct sectio
n
of
t
h
e im
age and
uses
the
m
to re
distribut
e th
e ligh
t
n
e
ss v
a
lu
es
o
f
im
a
g
e.
So
that contrast
of the im
age was
ad
j
u
sted
t
o
th
e li
m
i
t 0
an
d 1 hen
ce th
e b
l
oo
d
v
e
ssels ar
e
h
i
gh
lig
h
t
ed
.
3.1.3. Contras
t
Enhanc
e
m
ent using
Fas
t
Discrete Curvel
et Transform
Cu
rv
elet tran
sfo
r
m
is d
e
v
e
lo
ped
to
ov
erco
me th
e
li
mita
tio
n
o
f
wav
e
let an
d
Gabo
r tran
sform
s
[1
0
]
.
Alth
oug
h,
wavelets are
wid
e
l
y
u
s
ed
in
feat
ure ex
traction
bu
t it fails to
h
a
n
d
l
e
ran
d
o
m
ly
o
r
ien
t
ed
edg
e
s
o
f
th
e
o
b
j
ect and
th
e
sin
g
u
l
arities of th
e
ob
j
ect.
Gab
o
r
filters
o
v
e
rco
m
e th
e li
m
i
t
a
tio
n
o
f
wav
e
l
e
t tran
sform
an
d
d
eal
with
the
o
r
iented
ed
g
e
s,
b
u
t
it lo
ses t
h
e sp
ectral info
rm
ation
of the
image. Curvel
et trans
f
orm
is use
d
t
o
o
v
e
rco
m
e th
ese p
r
ob
lem
s
o
f
t
h
e
wav
e
let an
d Gab
o
r
filters.
It can ob
tain
t
h
e co
m
p
lete sp
ectral in
fo
rm
ati
o
n of
th
e im
ag
e an
d
h
a
nd
le
with
t
h
e differe
n
t
orie
ntations
of the
im
age edges.
The idea of curvelet is to repres
ent
a cur
v
e
as a super
p
osi
t
i
on o
f
fu
nct
i
o
ns o
f
vari
ous l
e
ngt
hs an
d
wi
dt
h
s
o
b
ey
i
n
g t
h
e scal
i
ng
l
a
w wi
dt
h
≈
l
e
ngt
h
2
. Th
is can
be don
e by first d
e
co
mp
o
s
i
n
g
th
e imag
e int
o
subba
n
ds i.e. s
e
parating the
object into a s
e
ries of
di
sjoint scales. The
n
, each scal
e is analyzed by a local
ri
d
g
el
et
t
r
ans
f
o
r
m
.
The
newl
y
const
r
uct
e
d
and
i
m
prove
d
versi
o
n
o
f
t
h
e
cur
v
el
et
t
r
a
n
sf
orm
i
s
kn
o
w
n
as Fast
Discrete Curve
l
et Transform
(FDCT).
The
n
e
w co
nst
r
uct
e
d
versi
o
n i
s
fast
er, si
m
p
l
e
r and
l
e
ss redu
nda
nt
t
h
an
the ori
g
inal curvelet trans
f
orm
,
which bas
e
d on
Ridgelet.
As m
e
ntioned, according to Cand'
es
et al.
[15]
t
w
o
im
pl
em
ent
a
t
i
o
ns
of
F
D
C
T
a
r
e p
r
o
p
o
sed:
1.
Unequally s
p
aced Fast
Fourier T
r
ansform
(USFFT
)
2. Wr
ap
pi
n
g
F
unct
i
o
n
Bo
th
im
p
l
e
m
e
n
tatio
n
s
of
FDCT d
i
ffer m
a
i
n
ly in
cho
o
s
i
n
g
th
e sp
atial grid
t
h
at used to
tran
slate
curvelet at each scale and a
n
gle. Bot
h
di
gital transform
a
tions
return a tabl
e of di
gital curvelet coe
ffi
cients
in
d
e
x
e
d b
y
scale, o
r
ien
t
atio
n
an
d lo
cation
param
e
ters. Here, we u
s
e th
e
wrapp
i
ng
m
e
th
o
d
to
im
p
l
e
m
e
n
t th
e
Fast Discrete Cu
rv
elet Transform
(FDCT) o
n
th
e retin
al
i
m
ag
e wh
ich is a
two
d
i
men
s
ion
a
l sig
n
a
l. Th
e
wra
p
pi
n
g
im
pl
em
ent
a
t
i
on i
s
sim
p
l
e
r, fast
er a
nd
has l
e
ss co
m
put
at
i
onal
com
p
l
e
xi
ty
t
h
an exi
s
t
i
ng a
p
p
r
o
aches.
Wrapp
i
ng
b
a
sed
curv
elet tran
sfo
r
m
is a
mu
lti-scale p
y
ramid
wh
ich
con
s
ists of d
i
fferen
t
o
r
ien
t
ation
s
and
p
o
s
ition
s
at a lo
w frequ
e
n
c
y lev
e
l. Basically,
m
u
ltireso
l
u
tio
n
d
i
screte cu
rv
elet tran
sform
in
th
e sp
ectral
d
o
m
ain
u
tilize
s
th
e ad
v
a
n
t
ages o
f
fast Fou
r
i
e
r Transfo
r
m
(FFT). Du
ring
FFT, bo
th
th
e i
m
ag
e an
d
th
e cu
rv
elet
at
a gi
ve
n sca
l
e and
o
r
i
e
nt
at
i
on a
r
e t
r
a
n
s
f
o
r
m
e
d i
n
t
o
t
h
e
Fo
uri
e
r
d
o
m
a
in.
At
t
h
e e
n
d
of t
h
i
s
com
put
at
i
o
n
process
,
we
obtain a set of
curvelet
coeffi
cients by applying inverse FFT
to
th
e sp
ectral p
r
o
d
u
c
t. Th
is set
contains c
u
rvel
et coefficients
in asce
ndi
ng
o
r
der
o
f
t
h
e
scal
e
s
an
d
ori
e
nt
at
i
ons
[
1
1]
.
Fi
gu
re
3.
St
eps
i
n
F
D
C
T
vi
a
wra
p
pi
n
g
m
e
t
hod
In orde
r t
o
obt
ain the c
u
rvele
t
coefficients
for
an im
age the below
steps a
r
e
perform
e
d seque
ntially.
1)
Ap
pl
y
t
h
e
2
D
FFT a
n
d
obt
ai
n F
o
uri
e
r
sam
p
l
e
s
^
f
[
n1,n
2
]
,
-
n
/
2
<
n
1
,
n2
<
n
/
2
(1
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Ret
i
nal
Bl
oo
d
Vessel
s
Ext
r
act
i
on
Base
d
o
n
C
u
rvel
et
Tr
an
sf
o
r
m
an
d
by C
o
mbi
n
i
n
g B
o
t
h
a
t
…
(
K
. G
a
yat
h
ri)
39
3
2)
For eac
h scale
j a
n
d angle l, form
the produc
t
~
U
j,l[n
1
,
n2
]
^
f
[
n1,
n2
]
(2
)
Wh
ere, j, l [n1
,
n2
] is th
e
d
i
screte lo
calizin
g
wi
n
dow.
3)
Wra
p t
h
i
s
p
r
od
uct
ar
o
u
n
d
t
h
e
ori
g
i
n
a
n
d
obt
a
i
ned
^
f
[
n1, n2
]
=
W
(
~
U
j,l
^
f
)
[
n1,
n2
]
(3
)
W
h
er
e, th
e
r
a
ng
e
f
o
r
n
1
is
n
o
w
0
<
n1
<
L1,j
an
d 0 <
n
2
<
L2
ar
e co
nstant.
4)
Apply the i
nve
rse
2D FFT
to
each
f(j,l)
, hence collecting the discrete
coe
f
ficients C
D
(
j
, l,
k
)
.
Sin
ce th
e Curvelet tran
sfo
r
m
is well ad
ap
ted
to re
pres
ent the
im
ages
contai
n
i
ng
ed
g
e
s, it is a go
od
candi
dat
e
fo
r edge e
nha
ncem
ent
.
Fu
rt
he
rm
ore t
h
e co
nt
rast
of an i
m
age i
s
enha
nced
usi
n
g m
odi
fi
ed cur
v
el
e
t
coef
fi
ci
ent
s
.
S
ubs
eq
ue
nt
l
y
ou
r
pr
op
ose
d
m
e
tho
d
t
o
a
n
al
yze
th
e retin
al im
a
g
e co
nsists of t
h
e
fo
llowing
st
ep
s:
i)
Ap
pl
y
i
ng
F
D
C
T
vi
a
wra
p
pi
n
g
m
e
t
hod
,
we
o
b
t
a
i
n
a
set
o
f
s
cal
es S
j
a
n
d
di
r
ect
i
onal
ban
d
s
C
i
coefficients.
ii)
For
each
direc
tional ba
nd, C
{
1}{1} t
h
e m
i
nim
u
m
thre
shold value
we
re
determ
ined and re
place all the
coefficients
with these
val
u
es.
iii)
R
econst
r
uct
t
h
e en
hance
d
i
m
age
us
ing t
h
ese
m
odified c
o
efficients.
Figure
4. Flow Chart t
o
classi
fy
Blood
Vess
els from
retinal im
age
Load RGB
im
age
Make red
and
blue ch
annel infor
m
ation=0,
keep gr
een
ch
an
nel
inform
ation
Reduces
the brig
htness of an
image using
Adaptive histogr
am equalization
Enhanc
e th
e
con
t
ras
t
of
an
im
age
b
y
FDCT
ST
OP
Improved morpholog
y
function
= Tophat
(SE) – Both
at (SE)
Apply
FDCT via wrapping metho
d
.
Get th
e se
t of
sc
ales,
S
j
and
dir
e
ctional
bands C
i
Calculate minimum value C{1}{1}and
Repla
ce al
l
th
e c
o
effic
i
ents
b
y
i
t
.
Reconstruct th
e
enhanc
ed im
age using
this
m
odified curv
el
e
t
co
effi
cien
ts
.
Apply
local ther
sholding with
fixed
threshold
Displa
y blood
v
e
ssel ex
tra
c
ted
i
m
a
ge
ST
ART
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
4, No
. 3,
J
u
ne 2
0
1
4
:
38
9 – 3
9
7
39
4
3.
2. E
x
tr
ac
ti
o
n
o
f
B
l
o
o
d
Ve
ssel
s
M
o
r
p
h
o
l
o
gy
i
s
a b
r
oad
set
o
f
i
m
age pr
oce
ssi
ng
o
p
e
r
at
i
o
ns t
h
at
p
r
oc
es
s i
m
ages base
d
on
s
h
apes
.
M
o
r
p
h
o
l
o
gi
cal
operat
i
ons a
p
pl
y
a st
ruct
uri
ng el
em
ent
t
o
an input im
age, creating a
n
out
put im
age
of the
sam
e
size [7]. In a m
o
rphol
ogical operation, the val
u
e
of eac
h pi
xel
i
n
t
h
e o
u
t
p
ut
im
age i
s
based o
n
a
co
m
p
ariso
n
o
f
th
e co
rrespon
din
g
p
i
x
e
l in
t
h
e in
pu
t im
ag
e with
its n
e
i
g
hbo
rs. Man
y
m
o
rp
ho
log
y
fun
c
tio
ns are
ap
p
lied
in
featu
r
e ex
t
r
actio
n
(e.g
., op
en
ing
)
, b
u
t
th
e p
r
ob
lem o
f
th
is fu
n
c
t
i
o
n
is th
at th
e p
i
x
e
ls in
th
e resu
lted
im
age can i
n
clude
all ne
gligi
b
le cha
n
ges i
n
th
e grey lev
e
ls
ex
istin
g in
t
h
e
i
m
ag
e.
I
n
ou
r pro
p
o
s
ed
algor
ith
m
,
im
prove
d m
o
rp
hol
ogy
f
unct
i
o
n i
s
use
d
a
n
d
i
t
i
s
de
fi
ne
d as
,
Im
pr
oved
f
u
nct
i
on
= i
m
su
bt
r
a
ct
{(
I
0
– (
I
0
◦
SE)
– (
I
0
•S
E)
–
I
0
)}
(4
)
Whe
r
e, I
0
is th
e im
ag
e to
b
e
pro
cessed
,
◦
–o
pe
ni
n
g
op
erat
or
,•
–cl
o
si
n
g
o
p
e
r
at
o
r
, S
E
i
s
t
h
e di
s
k
sha
p
e
d
structuring element. A struct
uri
ng el
em
en
t
is a
matrix
co
n
s
isting
of on
ly 0
'
s an
d
1
'
s th
at can
h
a
ve an
y
arbi
t
r
a
r
y
shap
e and si
ze. T
h
e pi
xel
s
wi
t
h
val
u
es o
f
1
defi
ne t
h
e nei
g
h
b
o
r
h
oo
d. T
h
e cent
e
r
pi
xe
l
of t
h
e
stru
cturing
elemen
t, called
the orig
in,
id
en
tifies
th
e p
i
x
e
l
bein
g
pro
cessed.
Th
e Toph
at tran
sfo
r
m
is u
s
ed fo
r ex
tracting
sm
a
ll o
r
narrow, bright or
da
rk feat
ures i
n
an im
age. It
is represen
ted
as,
h =
I
0
– (I
0
◦
SE)
(5
)
Th
e B
o
th
at tran
sfo
r
m
,
also
called
clo
s
ing
resid
u
e
, is
u
s
ed to
ex
tract valleys su
ch
as
d
a
rk
li
n
e
s and
dar
k
s
pot
s
.
It
i
s
a p
r
oces
s w
h
i
c
h i
s
do
ne
b
y
t
h
e su
bt
ract
i
on
o
f
t
h
e
ori
g
i
n
al
i
m
age fro
m
t
h
e cl
osi
n
g
resul
t
.
There
f
ore,
t
h
e
bl
o
o
d
vess
el
s of t
h
e ret
i
n
a
,
act
ual
l
y
consi
d
ere
d
as
dar
k
l
i
n
es are e
x
t
r
a
c
t
e
d by
a
ppl
y
i
ng t
h
e
bot
hat
t
r
a
n
sf
or
m
.
The b
o
t
t
o
m
-
h
a
t tran
sform
is exp
r
essed
as th
e
fo
llowing
eq
u
a
tion
,
h =
(
I
0
•
SE)
–
I
0
(6
)
In
o
u
r
pr
o
pos
e
d
w
o
rk
, m
o
rp
hol
ogi
cal
ope
r
a
t
i
on i
s
pe
rf
o
r
m
e
d by
hi
ghl
i
ght
i
n
g i
t
s
bac
k
g
r
ou
n
d
t
o
a
lin
e size o
f
7
.
Ten
ro
tated
stru
cturing
ele
m
e
n
ts
are applied with a radial resol
u
tion
of
15
. The struct
uri
n
g
ele
m
en
t len
g
t
h sho
u
l
d
b
e
chosen
su
ch
t
h
at it
m
u
st b
e
sm
al
ler th
an
t
h
e lowest p
i
x
e
ls presen
t in
th
e
set. Th
en
t
h
e hi
g
h
l
i
g
ht
ed bac
k
gr
ou
n
d
i
s
subt
ract
e
d
u
s
i
ng T
o
p
h
at
a
nd B
o
t
h
at
t
r
an
sfo
r
m
a
t
i
on so t
h
e bl
o
o
d
vess
el
s ar
e
sh
own
m
u
ch
cl
early when
com
p
ared
to
the
o
r
i
g
in
al im
ag
e.
As a resu
lt th
ere o
c
cur so
m
e
frills in
th
e fi
nal ed
g
e
i
m
ag
e d
u
e to in
trin
sic
no
ise
p
r
esen
t in th
e retin
al i
m
ag
e.
It
i
s
com
p
l
e
t
e
l
y
rem
oved a
n
d c
o
n
v
e
r
t
e
d i
n
t
o
bi
na
ry
i
m
ag
e with
l
o
cal th
resho
l
d
i
ng tech
n
i
qu
e. The fin
a
l im
ag
e is d
i
sp
layed
with
ex
tract
ed b
l
oo
d v
e
ssels ar
e
sh
own
in
b
l
ack
an
d the b
a
ck
gro
und
as in
wh
ite
4.
E
X
PERI
MEN
T
AL RES
U
L
T
S
Th
e au
to
m
a
tic
ex
traction
o
f
b
l
ood
v
e
ssel
s
fro
m
retin
al i
m
ag
e was ev
alu
a
ted on
t
h
e
p
u
b
licly
avai
l
a
bl
e DR
I
V
E an
d STAR
E dat
a
bases
.
The ex
peri
m
e
nt
s were i
m
pl
ement
e
d usi
n
g t
h
e M
A
TLAB
v
e
rsi
o
n
7
.
5
so
ftware initiall
y Cu
rv
eLab
too
l
bo
x was
in
stalled
in
Matlab
.
So
m
e
o
f
t
h
e v
a
l
u
es are differen
t
fo
r
d
i
fferent
im
ages, as t
h
ey
we
re est
i
m
at
ed
fr
om
di
ffere
n
t
ori
e
nt
at
i
ons
a
n
d
bac
k
gr
o
u
n
d
set
t
i
ngs.
The pr
o
p
o
s
ed
m
e
t
hod
ol
o
g
y
descri
bes
t
h
e vari
ous
t
ech
ni
que
s used
f
o
r
cont
rast
en
ha
n
c
em
ent
an
d
edge
detection in RGB retina
l
im
age. The Figure 5 s
h
o
w
s
th
e in
pu
t co
l
o
u
r
ed
retin
al image. Si
nce the
RGB
i
m
ag
e h
a
s
h
i
gh
er
d
i
m
e
n
s
io
n it was resized
to
less th
an
h
a
lf
o
f
th
e
o
r
ig
in
al si
ze.
In pre
p
roces
sing, gree
n
chan
nel
i
n
Fi
g
u
re
6 sh
ow
s t
h
e best
back
g
r
o
u
n
d
co
nt
rast
than
o
t
her two
ch
ann
e
ls; so
it was selected
for furth
e
r
p
r
o
cess. Si
n
ce
th
e non
un
ifo
r
mit
y
o
f
illu
m
i
n
a
tio
n
g
e
nerates th
e
frills in
t
h
e fin
a
l edg
e
imag
e,
so it is n
e
cessary
to
un
ifo
r
m
th
e i
m
ag
e illu
m
i
n
a
tio
n
.
Th
is enh
a
n
cem
en
t is d
o
n
e
b
y
u
s
i
n
g ad
ap
tiv
e h
i
st
og
ram
eq
u
a
lizatio
n
;
it
m
a
kes t
h
e vess
el
s appear
bri
g
ht
er t
h
a
n
t
h
e b
ackg
r
ou
n
d
i
s
sho
w
n i
n
Fi
g
u
r
e
7. Aft
e
rwa
r
d
by
usi
ng m
u
l
t
i
s
cal
e
an
d
th
e m
u
ltid
irectio
n
a
l Cu
rvelet tran
sform
ed
g
e
s o
f
an
i
m
ag
e are enh
a
n
c
ed
th
ereb
y in
creasin
g
th
e co
n
t
rast.
At the
outset
of FDCT
via
wrappi
ng, a
set
of scales
S
j
a
n
d
di
rect
i
o
nal
ba
n
d
s C
i
co
ef
f
i
c
i
en
ts
a
r
e
o
b
t
a
i
n
e
d
.
F
o
r
each
directional band, C{1}
{1} the
m
i
nimum
threshold
value is
dete
rmined
and re
place all the coefficients
with these m
odified c
o
efficient valu
es aft
e
r th
at im
ag
e
were
recon
s
tru
c
ted
u
s
ing
IFDCT is shown
i
n
Fi
gu
re 8.
B
y
i
n
t
r
o
duci
n
g i
m
prove
d m
o
r
p
hol
ogi
cal
f
unct
i
o
n
wi
t
h
s
t
ruct
u
r
i
n
g el
e
m
ent
s
t
h
e bl
o
od
vessel
s
a
r
e
ext
r
act
ed
.
It
i
s
per
f
o
r
m
e
d by
hi
g
h
l
i
ght
i
n
g
i
t
s
bac
k
gr
ou
n
d
t
o
a l
i
n
e
si
ze o
f
7.
Ten
r
o
t
a
t
e
d
st
ruct
u
r
i
n
g el
e
m
ent
s
are appl
i
e
d wi
t
h
a radi
al
resol
u
t
i
on
of 1
5
. T
h
e hi
g
h
l
i
g
ht
ed
back
g
r
o
u
nd i
s
subt
ract
ed wi
t
h
To
phat
an
d
B
o
t
h
at
t
r
ans
f
o
r
m
a
ti
on,
so t
h
at
bl
o
o
d
vessel
s
al
one
a
r
e s
h
o
w
n m
u
ch clear tha
n
ba
ckground
pixel
s
. T
h
e im
age is the
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Ret
i
nal
Bl
oo
d
Vessel
s
Ext
r
act
i
on
Base
d
o
n
C
u
rvel
et
Tr
an
sf
o
r
m
an
d
by C
o
mbi
n
i
n
g B
o
t
h
a
t
…
(
K
. G
a
yat
h
ri)
39
5
con
v
e
r
t
e
d t
o
a
bi
na
ry
i
m
age wi
t
h
L
o
cal
t
h
resh
ol
di
ng
of
fixe
d size.
He
nce the
res
u
ltant im
age where the
extracted bl
ood
vessels a
r
e in
b
l
ack and
t
h
e
b
ackgr
oun
d in
w
h
ite ar
e sh
own
in Figur
e
9
.
To
facilitate th
e p
e
rfo
r
m
a
n
ce
o
f
retin
al
v
e
ssel ex
tr
action
al
go
rith
m
s
, we h
a
v
e
selected th
e PSNR and
RMSE as
perform
ance m
eas
ures
. T
h
ose
m
e
asure
s
are
estimated as follows.
Fi
gu
re
5.
I
n
p
u
t
R
G
B
i
m
age
Fi
gu
re
6.
G
r
ee
n c
h
an
nel
represen
tatio
n
Fi
gu
re
7.
A
d
a
p
t
i
v
e hi
st
o
g
r
a
m
eq
u
a
lized
im
ag
e
Fi
gu
re
8.
C
u
rv
el
et
enha
nce
d
i
m
age
Fi
gu
re 9.
Se
gm
ent
e
d bl
o
o
d
ve
ssel
s
Peak
Sign
al to
No
ise Ratio
(PSNR):
PSNR ev
alu
a
tes t
h
e int
e
nsity cha
nges
of an im
age between t
h
e
ori
g
inal
and the
proces
sed im
age.
PSNR= 20
log
10
(
255
/MSE)
2
(7
)
M
ean S
q
uare
d
Err
o
r
(M
S
E
):
M
S
E(M
ean
S
q
uare
d E
r
r
o
r) i
s
com
put
ed
vi
a,
MSE =
mn
1
m
i
n
j
11
||I
0
(
i
,
j
)-I
p
(i,
j
)|| (8
)
RMSE =
MSE
(9
)
Where
,
MSE
are m
ean sq
u
a
red
er
ro
r of
im
a
g
e,
I
o
is th
e orig
in
al im
ag
e and
I
e
is the
enha
nced im
age.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
4, No
. 3,
J
u
ne 2
0
1
4
:
38
9 – 3
9
7
39
6
Table I .
Performance Analysis
Module
PSNR (dB)
RMSE
Pr
epr
o
cessing 30.
236
28.
419
E
nhancem
ent
Using FDCT
31.
56
22.
89
E
x
tr
action of bloo
d Vessels
using m
o
r
phological
oper
a
tion
34.
48
22.
67
5.
CO
NCL
USI
O
N
Here
we p
r
ese
n
t
a n
ovel
m
e
tho
d
t
o
de
vel
o
p
a qui
c
k
al
g
o
ri
t
h
m
for cl
assi
f
y
i
ng t
h
e
bl
o
o
d
vessel
s
i
n
retinal im
ages. It has conside
r
ed the c
r
iteria for asse
ssi
n
g
t
h
e m
e
t
h
o
d
s us
ed f
o
r en
ha
nci
ng t
h
e c
o
nt
rast
of t
h
e
im
ages and e
x
t
r
act
i
n
g t
h
e
bl
o
od
vessel
s
.
Si
nce t
h
e E
d
ge en
ha
ncem
ent
pl
ay
s an i
m
port
a
nt
rol
e
i
n
fi
nal
extraction res
u
lts, applying
histogram
equal
i
zation on retinal im
age will
have a noticeable effect on bot
h
h
a
v
i
n
g
t
h
e reti
n
a
l im
ag
es with
un
ifo
r
m
ill
u
m
in
atio
n
as well as i
m
p
r
ov
ing
th
e accu
r
acy o
f
th
e
final ed
ge
im
age. Considering the aforesaid attr
ib
u
t
es o
f
th
e Cu
rvelet tran
sform
,
it was seen
th
at, th
is d
e
v
e
lo
ped
instrum
e
nt has
serve
d
s
u
cces
sful in e
n
hanci
ng t
h
e cont
rast
of the im
ages. In t
h
e m
e
thod of com
b
ination
of
t
o
p
h
at
an
d
b
o
t
h
at
m
o
rp
hol
og
y
fu
nct
i
o
n wi
t
h
st
ruct
uri
ng
ele
m
ents, the st
ructure elem
ents act with m
o
re
powe
r
in
reco
gn
izing th
e edg
e
s.
Of cou
r
se, th
ere were
so
m
e
frills in
th
e ed
g
e
im
ag
e due to
th
e ch
ang
e
s in
illu
m
i
n
a
tio
n
o
f
th
e b
a
ck
groun
d. Th
ese frill
s were
rem
o
v
e
d
effectiv
ely b
y
lo
cal th
resh
o
l
d
i
ng
with
d
e
fi
n
e
d
v
a
lu
e. Con
s
i
d
ering
th
at
th
e al
g
o
rith
m
can e
x
tract blood
ves
s
els from
the re
tinal
im
ages with high
acc
uracy
in
n
ear
ly goo
d time, it can
b
e
u
s
ed
as th
e f
a
st an
d r
e
liab
l
e m
e
th
o
d
.
A
c
tually au
to
m
a
ted
an
alysis of
fu
ndus
im
ages req
u
i
r
e
s
segm
ent
a
t
i
o
n
of i
m
age i
n
t
o
regi
ons
suc
h
as
opt
i
c
di
sk,
f
o
v
ea, vessel
s
, an
d bac
k
gr
o
u
n
d
r
e
t
i
n
a.
The t
e
c
hni
que
desc
ri
be
d
her
e
can
pe
rf
orm
part
o
f
t
h
i
s
e
x
t
r
act
i
o
n
pr
oc
ess i
.
e.
bl
o
o
d
vessel
s
e
x
t
r
act
i
on.
I
n
fut
u
re, i
m
pro
v
e
d p
r
ep
r
o
cessi
ng t
e
c
hni
que
s sho
u
l
d
be
use
d
on t
h
e p
r
o
p
o
s
e
d al
g
o
ri
t
h
m
s
. Suc
h
t
ech
ni
q
u
e
s
coul
d
co
n
t
ribu
te to fu
rt
h
e
r im
p
r
ov
emen
ts on
th
e
alg
o
rith
m
s
, resu
ltin
g in
m
o
re ro
bu
st and
mo
re precise
d
e
t
ectio
n
th
at ev
en
tu
ally can
b
e
accep
ted
for th
e clin
ical p
u
rp
o
s
es.
REFERE
NC
ES
[1]
E. Annie
Edel Quinn and K. Go
kula
Krishnan, “Retinal Blood V
e
ssel Segmen
tation using Curvelet Tr
ansform and
Morphological Reconstruction”,
IEEE In
t. Co
nf. on Em
erging T
r
ends in Computing, Com
m
unication and
Nanotechnolog
y
, pp.570-575
, 20
13.
[2]
Saleh Shahbeig, “Automatic an
d quick blood
v
e
sse
ls extra
c
t
i
on
algori
t
hm
in re
tinal
im
ages”
,
The Institu
tion of
Engineering and
Technolog
y (
I
ET
)
Journal on I
m
age Processin
g
, Vol. 7
,
Issue
4, pp
. 392–400
,
Jan 2013.
[3]
Jaspreet Kaur, D
r
. H.P. Sinha, “An effi
cien
t blo
od vessel detection Algorithm fo
r retina
l
im
ages using local Entr
op
y
thresholding
”,
In
ternational Jour
nal of Engin
eer
ing Research
&
Technolog
y (
I
JERT)
, Vol. 1 Issue
4, June 2012.
[4]
Paintamilselv
i
and Shy
a
mala, “A novel method to
detect th
e fovea of
fundus r
e
tin
al image”,
2012 Internation
a
l
journal for resea
r
ch and develop
m
ent in
eng
i
neer
ing (
I
JRDE)
, Vo
l.1: Issue.1, pp-
21-25, June-July 2012.
[5]
C. Eswaran and
Marwan D. Saleh,
“An automated blood vessel extracti
on algor
ithm in fundus i
m
ages”,
2012 IEEE
International Co
nference on
Bi
o
i
nformatics and
Biomedicine (
B
IBM)
, pp. 1
-
5, Oct 2012.
[6]
M
.
Kala
ivani
,
M
.
S
.
J
e
yal
a
ks
hm
i, V. Aparn
a
,
“
E
xtract
i
on of R
e
tin
al Blood
Vessels Using Curvelet Tr
ansform an
d
Kirsch’s Templates”,
In
ternatio
n
a
l Journal of
Emerging Techno
logy and Ad
van
ced Eng
i
neering (
I
JETAE)
, Vol.
2,
issue 11, pp. 360
-363, Nov 2012
.
[7]
NagaRaju
, C
.,
NagaMani, S., Prasad,
G
.
R., Sunitha, S., “Morphological
edge
d
e
tection
algorith
m
based on multi-
s
t
ructure
el
em
en
ts
of diff
eren
t d
i
r
ect
ions
”,
Int. J. I
n
f. Commun. Technol. Res. (
I
JIC
T
)
, pp. 37–43, 2
011.
[8]
R. Priy
a
and P.
Aruna, “Review
of auto
mated diagnosis of diab
etic r
e
tinop
at
h
y
u
s
ing the support vector machine”,
International Jo
urnal of
A
pplied Engineering
Research
, Vol.
1 No
4, pp.844-863
, J
a
n 2011.
[9]
M.S. Miri and
A. Mahloojif
ar,
“
R
etinal Im
age
Anal
y
s
is
Using Curvelet
Trans
f
orm
and Multistructure E
l
em
e
n
ts
Morpholog
y
b
y
Reconstruction”,
IEEE Trans. on
Biomed. Eng
, V
o
l. 58
, no
. 5
,
pp
.1183-1192, May 2011.
[10]
J
o
ao V.B. S
o
are
s
, Roberto M
.
Ces
a
r J
r
., H
e
rbe
r
t F
.
J
e
lin
ek,
an
d M
i
chae
l J
.
Cr
ee, “
R
e
tina
l
Ves
s
el S
e
gm
entat
i
o
n
Using the 2D
Gabor Wavelet
an
d
S
upervis
ed C
l
as
s
i
fica
tion”
,
I
E
EE Trans. Med
.
I
m
ag.
, 2010
.
[11]
Jean-Luc Starck
, Fionn Murtagh, Jalal
M. Fadili (
2010), “Sparse Image and Signa
l Processing: Wavelets, Curvelets
,
and Morphological Diversity
”
.
U
n
ited
States of America:
Cambridge university pr
ess
.
[12]
B. Zhang, L
.
Z
h
ang, Le
. Zhan
g and F. Karray, “
R
et
inal
Ves
s
el Extra
c
tion b
y
Match
e
d Filte
r with First-Order
Derivative o
f
Gaussian”,
Computers in Biology a
nd Medicine
, Vo
1.40, Issue 4, 20
10.
[13]
Bai, M.R
., Kris
hna, V.V., Sree
Devi, J., “A new morphologi
cal approach for noise rem
oval cu
m edge detectio
n”,
Int.
J.
Comput.
S
c
i.
Is
s
u
es
(
I
JCSI)
, pp. 187–190
,
2010.
[14]
M.I. Iqbal, A.M. Aibinu
,
M. Nilsson and M.J.E.
Salami, “Automatic Diagnosis
of Diab
etic Retinopath
y
usin
g
Fundus Images
using Di
gital Signal and Image
processing Tech
niques”,
In
t. con
f
. on Robotics,
Vision, Informatio
n
and Signal Processing
, 2007
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Ret
i
nal
Bl
oo
d
Vessel
s
Ext
r
act
i
on
Base
d
o
n
C
u
rvel
et
Tr
an
sf
o
r
m
an
d
by C
o
mbi
n
i
n
g B
o
t
h
a
t
…
(
K
. G
a
yat
h
ri)
39
7
[15]
Candes, E.J., D
e
manet,
L., D
onoho, D.L., Ying, L., “Fast disc
rete Curvelet tr
an
sforms”,
SIAM Multiscal
e
Mod
e
l.
Simul
, pp
. 861–8
99, Mar
c
h 2006
.
[16]
Mendonca, A.M
., Campilho, A., “Seg
menta
tion
of retinal blood
vessels b
y
comb
ining the detection of centerlin
es
and morphological r
econstru
c
tion
”
,
I
E
EE
Trans. Med. Imag
., pp.
1200–1213, 200
6.
[17]
Available at:
http://www.
isi.
uu.
n
l/Research/Databases/DRIVE/.
[18]
Availab
l
e
at
:
htt
p
://curv
e
l
e
t.o
r
g/software.h
tm
l
[19]
http://www.
mathworks.
com.
[20]
http://www.pudn.com.
BIOGRAP
HI
ES OF
AUTH
ORS
K. Ga
yathr
i
gr
aduated her
B.E. in
Electronics
and communication eng
i
neer
ing from
C.Abdul Hakeem College of En
gineer
ing and
Technolog
y
,
M
e
lv
isharam. She is currently
pursuing the M.E. (Applied Electronics) from
C.Abdul Hakeem College of Engineer
ing
and Techno
log
y
, Melvisharam
.
She has published man
y
p
a
pers in Natio
nal and
International co
nference
proceedings. She attended man
y
wo
rkshops Robotics and
embedded s
y
stem design. Her research in
te
r
e
s
t
s
include Im
age P
r
oces
s
i
ng,
P
a
tter
n
Recognit
i
on, M
e
dical im
aging
and Multiscale
m
e
thods in im
age processing.
She is an
act
ive m
e
m
b
er
o
f
IET
E
.
D.
Narmadha
r
eceived B.E (ECE) in Ranganath
an Engineerin
g College, Coimbatore.
She is currently
pursuing Master
of Engine
ering
in Applied Electr
onics, C.Abdul
Hakeem
College of
Engineering
and Technolog
y
Melvis
haram. She has
presented man
y
papers in
National and International Conferences. Sh
e attend
ed workshop in Impleme
n
tation
a
l
As
pects
of M
i
crocontroll
ers
.
Her
Area of re
sear
ch includes Imag
eProcessing, Networking,
Embedded S
y
s
t
em. She ia
an active member of
IETE.
K. Thilagavathi
Started Career towards Engineer
ing in 2011 as B.E (ECE) stud
ent. She
is curretly
doin
g
Master of En
gineer
ing in
A
pplied
Electron
ics at C
.
Abdul
Hakeem
College of
Engineering
and
Technolog
y
,
Me
lvis
haram. She
attended man
y
wor
k
shop in
Embedded S
y
s
t
em and VLSI
domains. She
has pr
esented
in man
y
national
and
Interna
tiona
l co
nferenc
e
s
.
Her
Area of int
e
r
e
s
t
includ
es
I
m
age P
r
oces
s
i
ng, VLS
I
,
Embedded S
y
s
t
em Design, Ver
ilog, Network
i
ng.
She ia an
activ
e
member of IETE.
K. Pavithra
is
currently
purs
u
ing her B.E.
(E
CE) at C.Ab
dul Hakeem Colleg
e
of
Engineering and
Technolog
y
,
M
e
lvishar
a
m. Sh
e has presented man
y
papers in National
Level Technical
S
y
mposium. She attend
ed ma
n
y
workshops. Her
area of in
ter
e
st includes
Digital Im
age Processing, Digit
a
l Ele
c
tron
ic
s, Microcontro
llers
a
nd Electron
i
c Circuit
Designs.
M. Prad
e
e
p
a
graduated B.E. in Electron
i
cs
and communi
cation engineering fro
m
Bharadhid
a
san University
in 20
02 and rece
ived
the M.E. (Com
p
u
terr & Com
m
u
nica
tion)
from Anna univ
e
rsity
in 2004. She is currently
p
u
rsuing the Ph.D. under the ar
ea
of Signal
P
r
oces
s
i
ng. S
h
e
is
pres
entl
y
working as
As
s
o
ciat
e P
r
ofes
s
o
r at E
l
e
c
troni
cs
and
Communication Engineering, C
.
Abdul Hakeem
College
of Engineering and
Technolog
y
,
Melvisharam. She has published man
y
resear
ch papers in International Journ
a
l and
Interna
tiona
l Co
nferenc
e
s
.
Her
res
earch
int
e
res
t
s
include S
i
gn
al
P
r
oces
s
i
ng. S
h
e is
an
act
ive lif
etim
e M
e
m
b
er
of
IS
T
E
.
Evaluation Warning : The document was created with Spire.PDF for Python.