Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
9
, No
.
1
,
J
an
ua
ry
201
8
, p
p.
1
52
~
156
IS
S
N:
25
02
-
4752
,
DOI: 10
.11
591/
ijeecs
.
v9.i
1
.p
p
152
-
156
152
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
Leaf Re
cogniti
on u
sing Textur
e Fe
atures fo
r Herbal
Plant
Identific
atio
n
Z
aida
h Ib
r
ah
i
m
*
1
, Nur
ba
i
ty Sabri
2
,
Nu
r
N
ab
il
ah A
bu M
angsh
or
3
1
Facul
t
y
of
Com
pute
r and
Ma
them
at
ic
a
l
Sci
ences
,
Univer
si
ti Te
kn
ologi
MA
RA,
40
450
Shah
Alam,
Sela
ngor
2,3
Facul
t
y
of
Co
m
pute
r
and
Mat
hemati
c
al Sci
en
c
es,
Unive
rsiti Tekno
logi
MA
RA, Cam
pus Jasin,
Mela
ka
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
A
ug
2
2
, 201
7
Re
vised
N
ov
1
2
, 2
01
7
Accepte
d
Dec
1
, 2
01
7
Thi
s
re
se
arc
h
i
nvesti
gates
th
e
appl
i
cation
of
te
xtu
re
f
eatur
es
for
l
ea
f
re
cognition
for
h
erb
al
p
la
nt
ide
nt
i
fic
a
ti
on.
Ma
lay
si
a
is
ric
h
with
h
er
bal
pla
n
ts
but
not
m
an
y
peopl
e
ca
n
identif
y
th
em
and
know
about
t
hei
r
uses
.
Preserva
ti
on
of
the
knowledg
e
of
th
ese
her
b
p
la
nts
is
importa
nt
since
i
t
ena
bl
es
the
g
en
era
l
pub
li
c
to
g
ai
n
useful
know
le
dge
whi
ch
th
e
y
ca
n
app
l
y
whene
ver
ne
ce
s
sar
y
.
Leaf
image
is
chose
n
for
pla
nt
re
cogni
t
io
n
since
it
is
ava
i
la
bl
e
and
vis
ibl
e al
l
th
e
ti
m
e
.
Unlike
flow
ers
t
hat
a
re
no
t
a
lwa
y
s
av
ai
l
able
or
roots
that
are
not
visibl
e
and
n
ot
e
as
y
to
obt
ai
n
,
l
ea
f
is
the
m
ost
abunda
n
t
t
y
p
e
of
dat
a
ava
i
la
bl
e
in
bota
ni
cal
re
fe
re
n
ce
co
lle
ct
ions.
A
compa
ra
ti
v
e
stud
y
has
bee
n
condu
c
te
d
among
three
popula
r
t
ext
ure
fe
at
ur
es
th
at
are
Histogram
of
Orie
nte
d
Gra
die
nts
(HO
G),
Loc
al
Bin
ar
y
Pattern
(LBP)
and
Speede
d
-
U
p
Robust
Feat
ure
s
(SU
R
F)
with
m
ult
ic
la
ss
Support
Vec
tor
Mac
hine
(SV
M)
cl
assifi
er.
A
ne
w
le
af
dat
ase
t
has
bee
n
constru
ct
ed
from
te
n
di
ffe
re
nt
her
b
pla
nts.
Exp
eri
m
e
nta
l
re
sul
ts
using
the
new
constr
uct
ed
da
ta
se
t
an
d
Flavi
a,
an
exi
sting
da
ta
se
t,
indi
c
at
e
tha
t
HO
G
and
LBP
produc
e
sim
il
ar
le
af
r
ec
ogni
ti
on
per
form
anc
e
and
they
ar
e
b
et
t
er t
han
SU
RF
.
Ke
yw
or
d
s
:
HOG
LBP
L
eaf
recog
niti
on
M
ulti
cl
a
ss SVM
SU
RF
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
:
Zai
dah I
br
a
hi
m
,
Faculty
of Com
pu
te
r
an
d
Ma
them
a
ti
cal
Scie
nces
,
Un
i
ver
sit
i Te
knol
og
i M
ARA
,
40450 S
hah A
l
a
m
, S
el
ango
r,
Em
a
il
:
zai
dah
@tm
sk
.u
itm
.ed
u.m
y
1.
INTROD
U
CTION
Ma
la
ysi
a
is
rich
with
he
rb
al
plants
that
are
no
t
on
ly
us
e
fu
l
in
co
ok
i
ng
bu
t
al
so
is
ben
efici
al
in
m
edical
.
This
us
ef
ul
knowle
dg
e
s
houl
d
be
le
t
kn
own
to
th
e
public
so
tha
t
they
can
gain
the
ben
e
fits.
Be
sides
that,
the
acce
ss
to
this
knowle
dg
e
s
hould
be
easy
and
fas
t.
Thu
s
,
the
de
velo
pm
ent
of
Ma
la
ysi
an
m
e
dicinal
herb
rec
ogniti
on
syst
em
is
a
necessit
y.
Be
sides
Ma
la
ysi
a
[
1],
re
searc
he
s
on
he
rb
al
pl
ant
rec
ogniti
on
hav
e
been co
nducte
d
in
o
t
her co
un
trie
s su
c
h
as
C
hin
a
[
2]
, In
done
sia
[3]
a
nd
Viet
nam
[4
]
.
The
m
os
t
p
opular
par
t
of
th
e
plant
t
hat
ha
s
bee
n
us
e
d
f
or
pla
nt
rec
ogni
ti
on
is
the
le
af
beca
us
e
th
e
le
af
is
avail
able
throu
ghout
the
ye
ar
an
d
it
is
easi
ly
visibl
e.
Flow
e
rs
are
on
ly
avail
able
at
certai
n
tim
e
wh
il
e
the
r
oots
a
re
not
easi
ly
acce
ssible.
Be
sides
that,
le
af
is
the
m
os
t
abu
nda
nt
ty
pe
of
data
avail
able
in
bo
t
anical
ref
e
ren
ce
c
ollec
ti
on
s
[
4].
P
rof
essionals
who
are
w
orkin
g
to
gethe
r
with
bo
ta
ny
identify
pl
ants
thr
ough
le
ave
s
identific
at
ion
[
5]
.
But
m
anu
al
identific
at
ion
in
the
fiel
d
is
qu
it
e
troubleso
m
e
and
so
m
et
i
m
es
tim
e
con
su
m
ing
.
This
is
bec
aus
e
the
kn
ow
le
dge
f
ro
m
the
bota
nist
is
re
qu
ired
for
the
pl
ant
rec
ogniti
on.
T
hus,
t
o
ov
erco
m
e
these
prob
le
m
s,
he
r
bal p
la
nt iden
ti
ficat
io
n
th
rou
gh leaf
recogn
it
io
n
a
pp
li
ca
ti
on
is c
riti
cal
.
A
rob
us
t
featu
re
is
need
e
d
to
cl
ea
nly
disti
ng
uis
h
am
on
g
the
dif
fer
e
nt
le
aves.
Lea
ves
ha
ve
va
rio
us
char
act
e
risti
cs
that
can
be
us
e
d
for
rec
ogniti
on
s
uch
as
sh
a
pe,
col
or
an
d
t
extu
re.
Sh
a
pe
f
eat
ur
e
has
be
e
n
us
e
d
in
[
6]
w
her
e
the
le
af
’s
s
ha
pe
co
ntour,
the
conve
xity
and
co
ncav
it
y
pro
per
ti
es
of
t
he
arch
es
are
m
easur
e
d.
Hu
in
var
ia
nt
m
o
m
ents
that
represe
nt
the
s
hap
e
of
the
le
af
ha
ve
bee
n
util
iz
ed
in
[7
]
.
But
the
sh
a
pe
can
be
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Leaf Rec
ogniti
on u
si
ng Text
ure
Fea
t
ur
es f
or Her
ba
l
Pla
nt
I
den
ti
fi
catio
n
(
Za
id
ah
Ib
r
ahim
)
153
aff
ect
ed
by
th
e
age
of
the
pl
ant
w
her
e
yo
unge
r
le
aves
m
ay
loo
k
sli
ghtl
y
diff
e
re
nt
com
par
ed
t
o
t
he
old
e
r
le
aves
a
nd
thi
s
can
p
rod
uce
dif
fer
e
nt
resul
ts
for
the
sam
e
plant
.
Applyi
ng
c
olor
is
ve
ry
chall
en
ging
sinc
e
alm
os
t al
l plan
t l
eaves
hav
e
si
m
il
ar co
lors.
Textu
re
featu
r
e
fo
r
le
af
rec
ogniti
on
has
be
en
inve
sti
gated
in
[
8]
with
ver
y
good
re
su
lt
s.
S
cal
e
Invar
ia
nt
Feat
ur
e
T
ra
ns
f
or
m
(S
I
FT)
te
xtur
e
featu
res
of
pl
ant
le
aves
have
bee
n
util
iz
ed
in
[
5]
but
ext
racti
ng
the
SIFT
featu
re
is
a
bit
slo
w.
S
pee
de
d
-
U
p
Ro
bust
Feat
ur
es
(SURF
)
t
extu
re
feat
ur
e
has
been
ap
plied
in
[
9]
bu
t
t
he
pe
rfo
r
m
ance
is
not
as
good
as
SI
F
T.
F
uzzy
Loca
l
Bi
nar
y
Patt
er
n
(LB
P)
te
xtu
r
e
featu
res
are
bein
g
us
e
d
in
[
3].
Histogram
of
Or
i
ented
G
rad
ie
nts
(
HOG)
ha
s
pro
duced
go
od
perform
ance
in
[10].
E
ve
n
t
hough
m
any
of
the
m
et
ho
ds
pro
duce
sat
isfact
or
y
resu
lt
s,
the
r
e
is
sti
ll
ro
om
fo
r
im
pr
ove
m
ent.
Since
diff
e
re
nt
featur
e
s
can
pro
duce
dif
fer
e
nt
plant
rec
og
niti
on
r
esults,
the
identific
at
ion
of
a
good
featur
e
or
feat
ur
es
is
crit
ic
al
fo
r
pla
nt
rec
ogniti
on.
Th
us
,
a
com
par
at
ive
stu
dy
is
necessa
ry
to
de
te
rm
ine
the
s
uitable
feat
ur
e
for
le
a
f
recog
niti
on
of
the
Ma
la
ysi
a
n
m
edici
nal
p
la
nt.
Th
is
res
earch
in
vestig
at
es
three
popula
r
te
xture
fe
at
ur
es
nam
ely
SU
RF
,
LBP
a
nd
H
OG.
A
m
ulti
c
la
ss
Suppo
rt
Vecto
r
Ma
chi
ne
(
SV
M
)
ha
s
bee
n
util
iz
ed
as
the
cl
assifi
er
of
th
ese
featu
res
f
or
le
af
rec
ogniti
on.
S
VM
is
ch
os
e
n
beca
us
e
i
t
is
widely
us
e
d
f
or
va
rio
us
obj
e
ct
recog
niti
on
s
s
uc
h
as
waterm
elo
n
seeds
exte
rior
qual
it
y
recogn
it
io
n
[
11]
,
re
m
ote
sensing
i
m
age
cl
assifi
cat
ion
[12] an
d b
rain wave
r
ec
ogniti
on [1
3].
2.
RESEA
R
CH MET
HO
D
Textu
re
featu
r
es
pro
vi
des
m
easur
em
ents
f
or
vis
ual
patte
rn
s
in
im
ages.
T
he
ide
ntific
at
ion
of
s
pecific
te
xtu
res
in
a
n
i
m
age
is
achi
eved
by
m
od
e
li
ng
te
xt
ur
e
a
s
two
-
dim
ensio
nal
gray
-
le
vel
var
ia
ti
ons
within
a
segm
ented
regi
on
.
It
can
be
descr
i
bed
base
d
on
c
oar
se
nes
s,
co
ntrast,
dir
ect
ion
s,
li
nelik
eness,
re
gu
la
rit
y
an
d
rou
ghness
[
14
]
.
H
O
G
co
unts
the
occ
urre
nce
s
of
gra
dient
ori
entat
ion
in
l
oc
al
iz
ed
re
gions
of
an
im
age
[15]
.
Fig
ure
1
il
lustr
at
es the
HOG
a
lgorit
hm
. Th
e i
m
ple
m
entat
ion
of HO
G
al
gorithm
is d
escribe
d
as
foll
ows:
1.
Divid
e
the im
age in
t
o
sm
al
l reg
io
ns
call
ed
c
el
ls;
2.
Com
pu
te
a h
ist
ogram
o
f
gr
a
di
ent d
i
recti
ons for t
he pixels
w
it
hin
the
cel
l;
3.
Discreti
ze eac
h ce
ll
into
a
ngul
ar
bin
s acc
ordi
ng to
t
he gra
di
ent orie
ntati
on;
4.
Group a
djace
nt cell
s into bloc
ks
;
5.
Norm
al
iz
e the b
loc
k hist
ogra
m
.
SU
RF
[
16]
is
achieved
by
rely
ing
on
int
egr
al
im
ages
and
Hess
ia
n
m
at
rix
-
base
d
m
easur
e.
T
he
i
m
ple
m
entat
io
n of SUR
F is
de
scribe
d
as
foll
ow
s:
1.
Find im
age in
te
rest points
u
si
ng d
et
e
rm
inant of Hessi
an
m
a
trix;
2.
Find m
ajo
r
i
nterest points
in sc
al
e sp
ace;
3.
Find feat
ur
e
di
recti
on
;
4.
Gen
e
rate fe
at
ure
vectors.
LBP [
17
]
label
s the
pix
el
s
of
an
im
age b
y t
hresh
old
i
ng the
neig
hborh
ood of
eac
h pixel a
nd conside
r
s
the
res
ult
as
a
bin
a
ry
num
ber.
Fig
ur
e
2
il
lustrate
s
a
sam
ple
co
m
pu
ta
ti
on
of
the
center
pix
el
with
it
s
neig
hbori
ng
pi
xels
w
hile
F
igure
3
dem
on
strat
es
t
he
ge
ner
al
flo
w
of
process
of
LBP.
T
he
ge
ner
al
i
m
ple
m
entat
io
n of LB
P is e
xpla
ined
b
el
ow:
1.
Divid
e
the e
xa
m
ined
wi
ndow
into
cel
ls;
2.
Fo
r
eac
h pixel
in the cell
,
com
par
e the p
i
xe
l t
o
each
of its
ei
gh
t
neig
hbor
s;
3.
Pr
od
uce
a
vect
or
of
bin
a
ry
nu
m
ber
s
wh
e
re
if
the
ce
nter
pixe
l’s
val
ue
is
gr
ea
te
r
tha
n
the
neig
hbor’s
val
ue;
change it
t
o 0 a
nd 1 ot
herwise.
4.
Com
pu
te
the
hi
stog
ram
o
f
the
f
re
quency
of e
ach
nu
m
ber
oc
currin
g;
5.
Norm
al
iz
e and
concat
enate t
he
h
ist
ogram
o
f
al
l cel
ls.
SV
M
w
as
init
ia
ll
y
dev
el
op
e
d
by
for
bin
a
ry
cl
assifi
cat
ion
[
18
]
bu
t
m
ul
ti
cl
ass
SV
M
has
been
pro
du
ce
d
to
cat
er
m
ulti
cl
ass
p
roblem
s.
A
com
par
ison
has
be
en
co
nducted
a
m
on
g
f
our
popu
la
r
ap
proache
s
f
or
m
ul
ti
cl
ass
SV
M
nam
ely
one
-
agai
ns
t
-
al
l
(
OAA),
one
-
ag
ai
ns
t
-
one
(OA
O)
,
decisi
on
di
rected
acy
cl
ic
gr
a
ph
(DDAG
)
an
d
adap
ti
ve
direc
te
d
acy
cl
ic
gr
aph
(ADAG
)
and
the
e
xp
e
ri
m
ental
resu
lt
s
ind
ic
at
e
that
OAO
pro
vid
es
the
be
st
resu
lt
[19].
Thu
s
,
this
res
earch
util
iz
es
t
he
O
AO
a
ppr
oa
ch
for
m
ulticlass
SV
M
w
he
re
i
t
consi
sts
of
m
ulti
ple
bin
ary,
li
near
S
VM
le
ar
ner
s
.
Cl
assifi
c
at
ion
is
pe
rform
ed
by
a
m
ax
-
wi
ns
vo
ti
ng
strat
egy
wh
e
re
e
ver
y
cl
assifi
er
assig
ns
the
instance
t
o
on
e of
the
tw
o
cl
asses.
A
fter th
at
,
the v
ote
f
or
t
he
assig
ne
d
cl
ass
is
increases
b
y
on
e
vote a
nd th
e cla
ss w
it
h t
he
m
axi
m
u
m
v
otes d
et
erm
ines the in
sta
nce cla
ssific
at
ion
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vol
.
9
,
No.
1
,
Jan
ua
ry
201
8
:
152
–
156
154
Figure
1. Dem
on
st
rati
on of
H
OG alg
or
it
hm
[
15]
Figure
2. A
sa
m
ple
co
m
pu
ta
ti
on
of the
inter
est
points
for
LBP
Figure
3. A
d
e
m
on
strat
ion
of
the g
e
ne
ral fl
ow
of
LBP
Figure
4
il
lust
rates
the
flo
w
of
process
for
le
af
recog
niti
on
.
On
ce
t
he
im
age
of
the
le
af
has
bee
n
captu
red,
pre
-
processi
ng
is
c
onduct
ed
on
the
im
age
fo
r
r
esi
zi
ng
a
nd
c
onve
rtin
g
the
c
olor
im
age
into
gray
-
scal
e
i
m
age.
T
hen,
the
te
xtu
r
e
featur
es
are
extracte
d
a
nd
entere
d
into
m
ulti
cl
ass
SV
M
fo
r
le
af
rec
ogniti
on.
The
t
hr
ee te
xture feat
ur
es
, na
m
el
y HO
G
,
S
URF a
nd LBP
are
bein
g
e
xtra
ct
ed
se
par
at
el
y.
Figure
4. Flo
w
of
process
f
or
le
af r
ec
ogniti
on
Flavia
dataset
[20]
co
ns
ist
s
of
va
rio
us
ori
en
ta
ti
on
s
of
le
a
f
i
m
ages
from
t
hirty
three
di
fferent
plant
sp
eci
fies.
Fi
gu
re
5
s
hows
s
om
e
sa
m
ple
i
m
a
ges
of
the
l
ea
ve
s
that
hav
e
be
en
us
e
d
in
t
his
exp
e
rim
ent
where
40
sam
ples f
ro
m
each
s
pecies
ha
ve been
used
for t
rainin
g
a
nd
10 sam
ples f
or
te
sti
ng
.
Figure
6
s
how
s
so
m
e
sam
ple
i
m
ages
f
ro
m
the
new
dataset
that
co
ns
ist
s
of
le
aves
f
ro
m
10
di
ff
e
ren
t
Ma
la
ysi
an
herb
plants
wh
e
re
20
sam
ples
from
each
sp
eci
es
ha
ve
been
use
d
for
trai
ni
ng
w
hile
5
sam
ples
f
or
te
sti
ng
. Fo
r bo
t
h datase
ts, th
e
i
m
ages w
ere
in
v
a
rio
us
or
ie
ntati
on
s a
nd size
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Leaf Rec
ogniti
on u
si
ng Text
ure
Fea
t
ur
es f
or Her
ba
l
Pla
nt
I
den
ti
fi
catio
n
(
Za
id
ah
Ib
r
ahim
)
155
Figure
5. Sam
ple i
m
ages o
f
th
e lea
ves fr
om
Flavia
dataset
[20
]
Figure
6. Sam
ple i
m
ages
of th
e lea
ves fr
om
n
ew
const
ru
ct
e
d dat
aset
3.
RESU
LT
S
A
ND
DI
SCUS
S
ION
Table
1
il
lustra
te
s
the
recog
niti
on
acc
ur
acy
r
esults
of
t
he
t
hree
te
xt
ur
e
feat
ur
es
wit
h
m
ulticlass
SVM
cl
assifi
er
us
in
g
new
co
ns
tr
uct
ed
dataset
w
hile
Table
2
sho
ws
the
res
ults
for
Flavia
data
set
.
The
rec
ogniti
on
accuracy
is
c
om
pu
te
d
by
div
i
ding
the
co
rr
ec
tl
y
reco
gniz
ed
le
af
of
the
te
sti
ng
im
ages
with
the
total
num
ber
of
te
ste
d
i
m
ages.
The
re
su
lt
s
of
t
hese
e
xp
e
rim
e
nts
are
c
onsta
nt
fo
r
both
datas
et
s.
By
loo
ki
ng
at
these
two
ta
bles,
we
ca
n
c
le
arly
see
that
H
O
G
and
LBP
pro
duce
sim
il
ar
cl
a
ssific
at
ion
res
ul
ts
fo
r
both
dat
aset
s
an
d
thei
r
resu
lt
s
are
m
uch
bette
r
tha
n
SU
RF
.
Since
HOG
a
nd
LBP
de
script
or
s
ope
rate
on
local
iz
ed
cel
ls,
the
m
et
ho
ds
uphol
d
inv
a
riance
t
o
ge
om
et
ric
and
phot
om
et
ric
transfor
m
at
ion
s
a
nd
r
obus
t
t
o
m
onotonic
gr
ay
-
scal
e
cha
ng
es
cause
d,
for
in
sta
nce,
by
il
lu
m
inati
on
var
ia
ti
ons. As
a res
ult, they
pe
rfor
m
b
et
te
r
t
han S
URF.
Table
1.
Res
ults o
f
m
ulti
cl
ass
SV
M
for new
const
ru
ct
e
d dat
aset
Multiclas
s Clas
sif
ier
Accurac
y
(
%)
HOG
99
LBP
99
SURF
74
Table
2.
Res
ults o
f
m
u
lt
ic
la
ss
SV
M
for
Fla
vi
a d
at
aset
Multiclas
s Clas
sif
ier
Accurac
y
(
%)
HOG
97
LBP
97
SURF
63
4.
CONCL
US
I
O
N
This
pa
per
e
va
luate
d
thr
ee
te
xture
featu
res
f
or
le
af
recog
niti
on
,
nam
ely
HO
G
,
LBP
an
d
SU
RF.
O
ne
of
the
im
po
rta
nt
cha
racteri
sti
c
for
scene
i
m
ages
su
c
h
as
herb
pla
nts
is
tolerant
to
il
lum
inati
on
ch
ang
e
s.
Ex
per
im
ental
resu
lt
s
on
t
hese
featu
res
e
xtra
ct
ed
f
ro
m
le
af
i
m
ages
in
a
ne
w
c
on
st
ru
ct
e
d
dataset
s
an
d
Flavia,
existi
ng
datase
ts
con
cl
ude
th
at
HOG
a
nd
LBP
are
b
et
te
r
tha
n
S
URF.
SU
RF
is
sensi
ti
ve
to
r
otati
on
a
nd
il
lu
m
inati
on
changes
w
her
ea
s
H
OG
a
nd
L
BP
are
no
t.
Be
sides
that,
si
nc
e
HOG
a
nd
L
BP
desc
riptors
op
e
rate
on
l
ocali
zed
ce
ll
s,
the
m
et
ho
ds
uphold
in
var
i
ance
to
ge
om
et
ric
and
photom
et
ric
transf
or
m
at
ion
s.
Th
us
,
t
hey
are
ver
y
s
uitable
for
scene
i
m
ages
li
ke
leaf
recog
niti
on
fo
r
herb
pla
nt
identific
at
ion.
Fu
tu
re
w
ork
is
to
perform
a co
m
par
at
ive
stu
dy
on the e
ff
ic
ie
nc
y per
form
ance of th
ese
f
eat
ures
for
m
ob
il
e app
li
cat
io
n.
ACKN
OWLE
DGE
MENTS
The
a
utho
rs
grat
efu
ll
y
ack
n
owle
dge
t
he
help
of
U
niv
e
rsiti
Tek
nolo
gi
M
ARA
f
or
s
ponsori
ng
thi
s
researc
h u
nd
e
r Le
sta
ri grant
600
-
IRMI/M
yR
A 5/3/LES
TA
RI (0
60
/
2017).
REFERE
NCE
S
[1]
Saini
n,
M
.
S.
,
a
nd
Alfre
d,
R.
(
2
014).
F
eat
ure
S
el
e
ct
ion
for
Mal
aysian
Me
d
ic
ina
l
Pl
an
t
Leaf
Sha
pe
Ide
n
ti
f
ic
at
ion
and
Classifi
ca
ti
o
n
.
Int
ern
a
ti
ona
l Confere
nc
e
on
C
om
puta
ti
ona
l
Sci
enc
e
and
T
ec
hno
log
y
.
1
-
6.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vol
.
9
,
No.
1
,
Jan
ua
ry
201
8
:
152
–
156
156
[2]
Xu,
H.,
Li,
F.,
Xue,
F.,
and
Sh
en,
L
.
(2010).
A
ppli
cation
of
Pa
tt
ern
Recogni
ti
o
n
to
the
HNMR
Spec
tra
of
Chin
ese
Me
dicinal
Herb
s
for
Cold
-
hot
Nature
Distingui
s
h
.
6th
Int
ern
a
ti
o
nal
Confer
ence
on
Natur
al
Com
puta
ti
on
(ICNC).
3278
-
3281.
[3]
Herdi
y
eni,
Y.
,
Ginanj
ar
,
A.
R.
,
Anggoro,
M
.
R.
L.,
Douad
y
,
S.
and
Zuhud,
E.
A.M.
(2015).
Me
dLeaf:Mob
ile
Bi
odivers
it
y
In
f
orm
ati
cs
Tool
for
Mapping
a
nd
Ide
ntifyi
ng
Indone
sian
Me
dic
inal
Pl
an
ts
.
7
th
Int
ern
atio
nal
Confer
ence
of
Soft
Com
puti
ng
a
nd
Pattern
R
ec
o
gnit
ion (S
oCP
aR).
5
-
9.
[4]
Le
,
T
.
,
Tr
an,
D.
and
Hoang,
V.
(2014).
Ful
l
y
aut
omatic
le
af
-
base
d
plant
ide
ntification,
applicat
io
n
for
Vi
et
namese
medic
ina
l
plan
t
search
.
Proce
edings
of
the
Fif
th
S
y
m
posi
um
on
Inform
at
ion
and
Com
m
unic
at
ion
Te
chno
log
y
.
146
-
154.
[5]
Pri
y
ank
ara,
H.
A.
C.
and
W
it
hana
ge
,
D.
K.
(2015).
Computer
Assisted
Pl
ant
Ide
nti
f
ic
a
ti
on
Sy
stem
for
Android
.
IEE
E
Moratuwa
Engi
ne
eri
ng
R
ese
arc
h
.
148
-
153
.
[6]
W
ang,
B.
,
Bro
wn,
D.,
Gao
,
Y.
and
La
Sal
le,
J.
(2013).
Mobi
l
e
Pl
an
t
Leaf
Id
ent
ification
Us
i
ng
Smar
t
Phones
.
Inte
rna
ti
ona
l
Co
nfe
re
nc
e
on
Im
a
ge
Proce
ss
ing
(I
CIP).4417
-
4421.
[7]
Luki
c
,
M.,
Tub
a,
E
.
and
Tub
a,
M.
(2017).
Lea
f
Re
c
ognit
ion
Algor
it
hm
using
Support
Ve
ct
or
Mac
hine
wi
th
Hu
Moment
s
and
Local
B
inary
P
at
te
rns
.
IEEE
15t
h
Inte
rna
ti
ona
l
S
y
m
posium
on
Applie
d
Mac
h
in
e
Intelli
g
ence
a
nd
Inform
at
ic
s.
485
-
490.
[8]
Jam
il
,
N.,
Hus
sin,
N.A.C.,
Nordin,
S.,
Aw
ang,
K.:
Aut
omatic
plant
ide
ntifi
ca
ti
on:
is
shape
the
ke
y
f
eat
ure?
Proce
dia Com
put.
Sc
i. 76, 436
–
4
42
(2015).
[9]
Ngu
y
e
n
,
Q.
K.,
Le
,
T
.
L.
and
Pham
,
N.
H.
(201
3).
Leaf
based
plant
ide
nt
if
i
cat
io
n
system
for
an
droid
using
su
rf
fe
atures
in
com
binat
ion
w
it
h
ba
g
of
words
model
and
sup
erv
ise
d
le
arning
.
Int
er
nat
ion
al
Conf
erence
on
Android
Te
chno
logi
es
for
Com
m
unic
at
ion
s (ATC).
40
4
-
40
7.
[10]
Xiao,
X.
Y.,
H
u,
R.
,
Zh
ang,
S
.
and
W
ang,
X
.
(2010).
HOG
-
based
Approach
for
Leaf
Classif
ic
ati
on
.
Ad
vance
d
Inte
lligen
t
Comp
uti
ng
Theori
es
a
nd
Application
.
Springer
Ber
li
n/
Heide
lb
erg
.
149
-
155.
[11]
Chen,
X.,
W
ang
,
L.,
Chen,
W
.
and
Gao,
Y.
(2
013).
Detect
ion
o
f
Wate
rm
el
on
See
ds
Ex
t
erior
Qualit
y
based
o
n
Mac
hine V
ision
.
TE
LKOM
NIK
A
,
Vol
.
11
,
No.
6,
2991
–
2996
.
[12]
Yang,
N.
,
L
i,
S.
,
L
iu,
J.
and
Ful
ingBi
an
.
(2014).
Sensit
i
vi
t
y
of
S
upport
Vect
or
Mac
hine
Classif
ic
ati
on
to
Vario
us
Tr
aini
ng
Fe
a
tures
.
T
EL
KO
MN
IKA
Indone
sian
J
ourna
l
of
E
le
c
trica
l
Engi
n
ee
r
ing.
Vol.
12
,
No.
1,
2
86
–
291
.
[13]
Zha
ng,
Z.,
Li,
Y.
and
Peng,
X.
(2016).
Brain
Wav
e
R
ec
ogni
tion
of
Word
Imaginati
on
based
on
Support
Ve
ctor
Mac
hine
s
.
T
ELK
OM
NI
KA
.
Vol.
14
,
No
.
3A,
27
7
–
281
.
[14]
Ta
m
ura
,
H.,
Mo
ri,
S.,
Yam
awa
k
i,
T.
(1
978)
.
Tex
tural
fe
atures
co
rr
esponding
to
vi
sual
perc
eption
.
IEE
E
Tra
ns
on
S
y
stems
,
Man
a
nd
C
y
b
ern
e
ti
cs
.
8,
460
–
472
.
[15]
Dala
l
,
N.
and
Tr
iggs,
B.
(2005).
Histogram
s
of
o
rient
ed
gradien
t
s
for
human
det
ec
ti
on
.
IEEE
Com
pute
r
Vision
and
Patt
ern
Rec
ogn
ition (CVPR).
886
-
893.
[16]
H.
Ba
y
,
From
W
ide
-
base
li
ne
Poi
nt
and
Li
n
e
Corr
esponde
nce
s
to 3D
,
Ph.D. The
si
s,
ET
H
Zurich, 2
006.
[17]
He,
D.
C
.
and
W
ang,
L.
(1990
),
Tex
ture
Unit
,
Text
ure
Sp
ec
tr
um
and
Text
ure
Anal
ysis
.
IEEE
Tra
nsa
ct
ions
o
n
Geosci
ence and Rem
ote
Sensing.
vol. 28, 509
-
5
12.
[18]
Corte
s,
C
.
and
V
apni
k,
V.
(1
995)
.
Support
-
ve
ct
or
net
work.
M
ac
hi
ne
L
ea
rn
ing. vol.
20
,
273
–
297
.
[19]
Hs
u,
C.
W
.
and
Li
n,
C.
J.
(2
002).
A
comparison
of
methods
for
multi
cl
ass
suppor
t
ve
ct
or
machine
s
.
IE
E
E
Tra
nsac
ti
ons on
Neura
l
N
et
work
s,
vol.
13
,
no
.
2
,
415
-
4
25.
[20]
W
u,
S.
G.,
Bao
,
F.
S.,
Xu,
E
.
Y.
,
W
ang,
Y.
X.,
Ch
ang,
Y.
F. and
Xiang,
Q.
L.
(2
00
7).
A
Leaf Re
cog
nit
ion
A
lgorit
hm
for
Pl
ant
cl
ass
if
ic
at
ion
Us
ing
Probabil
isti
c
N
eural
Net
work
.
IEE
E
7
th
Int
e
rna
ti
on
al
S
y
m
p
osium
on
Sign
al
Proce
ss
ing
and
I
nform
at
ion
T
ec
h
nol
og
y
.
11
-
16
.
Evaluation Warning : The document was created with Spire.PDF for Python.