Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
V
o
l.6
,
No
.5
,
Octo
b
e
r
2
016
, pp
. 21
67
~
2
175
I
S
SN
: 208
8-8
7
0
8
,
D
O
I
:
10.115
91
/ij
ece.v6
i
5.1
025
4
2
167
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
Wood Cl
assifi
cati
on B
a
sed on
Edge Detections and Texture
Features Selection
Achm
ad Fahr
uroz
i
1
, Sarifu
ddin
Madend
a
1
, Ern
a
s
t
uti
1
, D
j
at
i K
e
ra
mi
2
1
Computer Science Dep
a
rtment,
Gunadarma University
, Indonesia
2
Departement o
f
Math
ematics,
University
of
In
donesia, Indones
i
a
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Feb 21, 2016
Rev
i
sed
Ju
l 19
,
20
16
Accepte
d
J
u
l 30, 2016
One of the p
r
op
erties of wood is a m
echanical
property
,
includ
es: hardness,
strength, cleav
age resistance, etc. Am
ong these properties ther
e that can be
m
eas
ured or es
tim
ated b
y
vis
u
al obs
ervation
on cross-sectional areas of
wood, which
is based on in
ter-f
iber dens
ity
,
f
i
ber size, and
lin
es that build
the
a
nnua
l rings.
In this pape
r,
we propose
d a
ne
w wood qua
lity
classification method based
on edge
detections. Edge
d
e
tection is applied to
the wood test images with the
aim to im
provin
g
the
characteris
tics of wood
fibers so as to
make it easier
to di
stinguish th
eir quality
.
Gray
Level Co-
occurren
ce Matr
ix (GLCM) used to obt
ain wood textur
e features, while th
e
wood quality
classification don
e
b
y
Naïve B
a
yes classifier.
Found in our
experimental res
u
lts
that the
f
i
rst-order
edg
e
d
e
tection
is likely
to
provide a
good accur
a
c
y
r
a
te and pr
ecisio
n
. The second o
r
der edge de
te
cti
on is high
l
y
dependen
t
on th
e choice o
f
par
a
meters a
nd tends to giv
e
worse classificatio
n
results, as filter
i
ng the or
iginal wood
image, thus blurri
ng ch
aracteristics
related
to wood
density
.
S
e
l
ect
i
on of fe
atures
o
b
tain
ed from
co
-occurren
c
e
m
a
trix is
also qu
ite
aff
ect
ed
the
c
l
assific
a
tion
resu
lts.
Keyword:
Ed
ge det
ect
i
o
n
GLCM
Naïve
-
bayes cl
assifier
Texture feature
s
Wood
q
u
a
lity classificatio
n
Copyright ©
201
6Institute of
Ad
v
anced
Engineeri
ng and Scien
c
e.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
Achm
ad Fa
hr
u
r
ozi
,
C
o
m
put
er Sci
e
nce
Depa
rt
m
e
nt
,
Gun
a
d
a
rm
a Univ
ersity,
10
0
M
a
rg
o
nda
R
o
ad
, Dep
o
k
, West
Ja
va, I
n
d
one
si
a.
Em
a
il: ach
m
a
d
.
fah
r
u
r
o
z
i12
@
g
m
ai
l.co
m
1.
INTRODUCTION
Th
e d
e
term
in
atio
n
of wood
q
u
ality
is
esp
ecially
i
m
p
o
r
tan
t
fo
r a m
a
n
u
f
act
urer
of
wood
, as a m
ean
s o
f
quality control whic
h se
rve
s
to
determ
ine the a
p
pr
opriate price
for each wood
product a
n
d
maintai
n
co
nsu
m
ers’ con
f
i
d
en
ce. Th
e
q
u
a
lity of
wood
is
n
a
turally
in
fl
u
e
n
c
ed
b
y
sev
e
ral factors, in
clu
d
i
n
g
th
e
den
s
ity
o
f
th
e
wood
stru
ct
u
r
e fi
b
e
r, sh
elter, weat
h
e
r an
d
so
forth. Th
erefo
r
e,
woo
d
will h
a
v
e
sev
e
ral of m
ech
an
ical
pr
o
p
ert
i
e
s, t
e
xt
ures
, s
h
ape
s
a
n
d
col
o
rs
. T
h
e
det
e
rm
i
n
at
i
o
n
of
w
o
od
q
u
al
i
t
y
can be
bas
e
d
on
seve
ral
t
h
i
n
g
s
,
i
n
cl
udi
ng sl
op
e of t
h
e
wo
o
d
f
i
ber [
1
]
,
de
fect
s i
n
t
h
e w
o
o
d
,
especi
al
l
y
knot
s [2]
o
r
t
h
e m
e
chani
cal
p
r
o
p
e
r
t
i
e
s o
f
w
ood
[3
],
[4
].
H
a
rdn
e
ss and
stif
f
n
ess
are mechanical properties that fre
qu
ently observe
d
and consi
d
ered to
d
e
term
in
e q
u
a
lity
o
f
wo
od
[4
], and
h
a
v
e
p
o
s
itiv
e co
rrel
atio
n
to
d
e
n
s
ity o
f
wood
[5
]. So
th
e alleged
of
o
b
s
erv
a
tio
ns are b
a
sed
on
the d
e
n
s
ity b
e
tween
con
s
titu
ent o
f
ann
u
a
l ring
s. Th
e nu
m
b
er o
f
lin
es th
at
b
u
ild
an
nu
al r
i
n
g
s
of f
o
u
r
-
s
eason
s
w
ood
usu
a
lly m
o
r
e
th
an
th
e
w
ood
in
two-
season
s r
e
g
i
on
or tropical. The distance
th
e lin
es
on
ann
u
a
l
ring
s
d
e
term
in
ed
b
y
large-sm
al
l an
d
d
e
n
s
ity th
e co
n
s
ti
tu
en
t
fib
e
rs
o
f
t
h
e
wood
.
At th
e sim
p
le
st lev
e
l, th
e assessm
en
t o
f
q
u
a
lity o
f
woo
d
carried
th
ro
ugh
v
i
su
al ob
serv
ation
by
expe
rt
s w
h
o ar
e expe
ri
ence
d
[6]
.
B
u
t
i
t
i
s
t
e
nd t
o
be l
e
ss efficien
t in
term
s o
f
tim
e an
d
effo
rt.
Ov
er th
e ti
m
e
,
th
e assessm
en
t o
f
th
e q
u
a
lity o
f
wood
can
also
b
e
do
ne in
th
e lab
o
r
ato
r
y, wh
ere testin
g
can
b
e
eith
er
d
e
stru
ctiv
e or
n
ond
estru
c
tiv
e
[
3
]. A
s
t
h
e d
e
velo
p
m
en
t o
f
cur
r
e
n
t
tech
no
logy, th
er
e ar
e lo
t o
f
r
e
sear
ch
h
a
s b
een
done to
re
place the function
of t
h
e hu
m
a
n eye as a tool of vis
u
al observ
ations,
one of
whom
is use digital
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:20
88-
870
8
I
J
ECE
Vo
l. 6
,
N
o
. 5
,
O
c
tob
e
r
20
16
:
216
7–2
175
2
168
cam
e
ras. B
y
usi
ng a di
gi
t
a
l
cam
era we can obt
ai
n a
di
g
ital im
age of an object.
Di
gital image obtaine
d can be
read a
n
d p
r
oc
essed
by
a co
m
put
er, t
o
su
bse
que
nt
l
y
us
e in the nece
s
s
ary resea
r
ch. Digital im
age-base
d
researc
h
has
b
een de
vel
o
pe
d fo
r co
nsi
d
e
r
ing the cha
r
acteristics of the object
in a digital image that contains
th
e obj
ect.
Non
d
e
stru
ctiv
e meth
od
s
related
to
woo
d
qu
ality tests h
a
v
e
been
d
o
n
e
su
ch as b
y
u
s
i
n
g
a wo
od
im
age base
d
o
n
t
h
e sl
o
p
e
o
f
t
h
e w
o
od
fi
ber
[
1
]
o
r
c
o
nsi
d
e
r
i
n
g t
h
e
n
u
m
b
er
of
de
fect
s a
n
d
t
h
ei
r
di
st
ri
b
u
t
i
o
n
s
[7
]. Th
is stu
d
y
fo
cu
ses on
th
e classificatio
n
o
f
wood
qu
ality u
s
in
g
a wood
i
m
ag
e, bu
t is o
b
s
erv
e
d
a test wo
od
t
e
xt
ure
by
u
s
i
ng t
e
xt
u
r
e feat
ures
o
f
w
o
od
.
Text
u
r
e
f
eat
ur
es ob
tain
ed
by u
s
ing
Gray-Level Co-occ
urre
nce
Matrix
(GLC
M) as an
ex
tr
acto
r
wh
ich
is im
p
l
e
m
en
ted
in Matlab
so
ftware.
Tex
t
u
r
e featu
r
es
of
GLC
M
th
at
ofte
n used in
researc
h
of texture a
n
alysis are C
ont
ra
st, Correlation, Inve
rse
D
i
ff
er
en
ce Mo
m
e
n
t
(I
D
M
),
Ent
r
opy
,
E
n
er
gy
an
d
H
o
m
e
geni
t
y
[
8
]
-
[
1
3]
. B
e
f
o
re
feat
u
r
e e
x
t
r
act
i
o
n,
edge
det
ect
i
o
n
carri
e
d
,
beca
use t
h
e
characte
r
istics of the
edges
in a
n
im
age can
be
use
d
to
represen
t
th
e im
ag
e to
furth
e
r an
alysis and
i
m
p
l
e
m
en
tatio
n
[1
4
]
, an
d
also
r
e
info
r
ces the o
b
s
er
vatio
n
o
f
f
i
b
e
r
d
e
n
s
ity an
d
annu
al rin
g
s
o
n
th
e
o
f
w
ood
were ob
serv
ed
in
th
is st
u
d
y
.
2.
THEORITICAL CONCEPTS
2.
1.
Edg
e
detections
Ed
ge det
ect
i
o
n
i
s
a
m
e
t
hod
u
s
ed t
o
gai
n
an
edge c
o
nt
ai
ned
i
n
an im
age o
b
ser
v
e
d
. E
dge
st
at
es l
i
m
it
whe
r
e t
h
ere
ha
s drast
i
c
cha
n
g
e
of g
r
ay
l
e
vel
[15]
. I
n
ge
ne
r
a
l
,
edge det
ect
i
on m
e
t
hod i
s
di
vi
de
d i
n
t
o
t
w
o
,
i
e
first-o
r
d
e
r ed
ge d
e
tectio
n
meth
od
s (in
c
l
u
des Ro
b
e
rts,
Prewitt an
d
Sobel) an
d
secon
d
-ord
er edg
e
d
e
tection
m
e
thod
(Canny and La
placia
n
of
Gaussian
filters). R
obe
rt
s detector usi
n
g a
2x2 m
a
sk or
kernel,
whe
r
e the
d
i
fferen
tial soug
h
t
b
y
d
i
ago
n
a
lly (4
5 and
13
5 d
e
g
r
ees) as illu
strated in
t
h
e
Fig
u
re
1
.
Fi
gu
re
1.
R
e
l
a
t
i
ons
hi
p
bet
w
ee
n i
m
age pi
xel
s
and
m
a
sk of
R
obe
rt
s
ope
rat
o
r
Th
us, R
obe
rt
s
o
p
erat
o
r
det
e
ct
s t
h
e ed
ges
i
n
a di
a
g
onal
di
rect
i
o
n. T
h
e
pi
xel
i
n
t
e
nsi
t
y
val
u
es i
n
matrices as Rob
e
rts
op
erat
o
r
resu
lts at po
sitio
n (
x,
y
)
on
th
e
th
is p
a
p
e
r is calcu
l
ated
b
y
th
e
form
u
l
a:
r
x
,
y
|p
1
–
p
4
|
|p
3
–p
2
|
(1
)
Mean
wh
ile, Prewitt an
d So
b
e
l are ed
g
e
op
erato
r
s t
h
at use a 3x
3 k
e
rn
el as
sh
own
in th
e Fig
u
re
2
.
Fi
gu
re
2.
Ke
rn
el
of
(a
) P
r
ewi
t
t
ope
rat
o
r a
n
d
(
b
)
So
bel
o
p
erat
ors
From
Fi
g
u
re
2
,
we
ca
n see
t
h
at
bot
h P
r
ewi
t
t
and
S
obel
o
p
erat
or
det
ect
s
t
h
e e
d
ges i
n
vert
i
cal
a
n
d
ho
ri
zo
nt
al
di
re
ct
i
ons. C
a
nny
edge
det
ect
i
o
n
per
f
o
r
m
sm
o
o
t
h
i
n
g i
n
a
d
va
nce u
s
i
n
g Ga
u
ssi
an f
u
n
c
t
i
on
and i
s
claim
e
d as the best edge
detection m
e
thod because it can capture both
e
dge
s of the strong and wea
k
edge
s,
but
has c
o
m
p
l
e
x com
put
at
i
on
and t
i
m
e cons
u
m
i
ng [1
4]
,[
1
6
]
.
Laplacian of Gaus
sian (L
oG) has a kernel size is
v
a
r
i
ed
an
d r
e
qu
ir
e
s
(si
g
m
a
) as a
param
e
ter, w
h
ere
sigm
a is pa
ram
e
ter used in
L
o
G
’
s
fo
rm
ula [14]
:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE I
S
SN
:
208
8-8
7
0
8
Woo
d
Cla
ssifica
tio
n Ba
sed on Edg
e
Detection
s
a
n
d
Textu
r
e Fea
t
u
r
es
S
e
lectio
n
(Achma
d
Fah
r
u
r
o
z
i)
2
169
LoG(
x
,
y
) =
1
(2
)
Gene
ral
l
y
, t
h
e
rel
a
t
i
ons
hi
p
b
e
t
w
een
val
u
e
of si
gm
a and t
h
e si
ze
of t
h
e
ker
n
el
t
h
at
i
s
oft
e
n u
s
ed i
s
si
ze=cei
l
i
ng (
3
*si
g
m
a
)*2+
1.
2.
2.
Gray-le
v
el co-occurrence
m
a
tri
x
(GLCM)
Gray-Level C
o
-occ
urrence
Matrix
(G
LC
M)
is a m
e
th
o
d
of
ten used
in t
h
e text
ure
analysis [17],
pr
o
pose
d
by
Haral
i
c
k i
n
[1
8]
on
19
7
3
t
h
at
i
n
cl
udes 1
4
feat
ures t
e
xt
u
r
e. GLC
M
p
r
o
duce
s
feat
u
r
es
whi
c
h
descri
be well
t
h
e relations
hip
of adjacency a
m
ong pixels in
an texture im
age. The
s
e features extracte
d
from
a
co-occ
urrence matrix,
where
th
e con
s
tru
c
tion
of th
ese m
a
trices d
e
p
e
nd
s
o
n
sev
e
ral p
a
ra
m
e
ters, in
clud
ing
:
sp
atial d
i
stan
ce (
d
),
di
rect
i
o
n o
r
a
ngl
e
or
i
e
nt
at
i
on (
θ
) an
d
gray-limit
s
(
G
). Sp
atial
d
i
stan
ce d
e
term
in
es
adjace
ncy
di
st
ance am
ong
pi
xel
s
o
f
an i
m
age,
whi
l
e
t
h
e
di
rect
i
o
n det
e
r
m
i
n
es ho
w t
o
nei
g
hb
o
r
i
n
g pi
xel
s
o
n
the
specified direction. Gray-limits
para
m
e
ter determ
ines the size of the
co-occ
urrence
matrix are generated,
as well p
a
rtitio
n
e
d
i
n
ten
s
ity v
a
lu
e con
t
ained
in th
e en
tire i
m
ag
e to
p
r
o
d
u
ce a m
a
tri
x
th
at is “conv
erted
”
.
Fi
gu
re
3 s
h
o
w
s an e
x
am
pl
e of t
h
e rel
a
t
i
o
n
s
hi
p
bet
w
ee
n co-occ
urrence matrix
cons
truction a
nd t
h
e s
e
lected
param
e
ter valu
es.
Fig
u
re
3
.
Illu
st
ratio
n of
GLC
M
con
s
tru
c
tio
ns: (a) m
a
trix
of im
age I, (b) conve
rted
m
a
trix, (c) c
o
-occ
urre
nce
matrix
ob
tain
ed
b
y
choo
sing
d
=
1,
θ
= 0 a
n
d
G
=
8
A
s
f
i
gu
r
e
above, Figur
e
3
-
a i
s
a m
a
tr
ix
r
e
presen
tatio
n of
an
im
ag
e I
w
ith th
e assu
m
p
tio
n
h
a
v
e
gr
ay
lev
e
l 0-255
,
wh
ile Fi
g
u
re3-b d
e
scrib
e
s co
nv
erted
m
a
trix
b
y
G
= 8
,
with
ru
les:
i
n
ten
s
ity
fro
m
h
*
2
to
(
h
+1
)*(
2
-1)
cha
nge
d t
o
(
h
+1
),
w
h
er
e
h
= 0,1,2
,
…,
6
,
7.
2.
3
Naï
v
e-B
aye
s c
l
assifier
Naïv
e-Bayes classifier wo
rk
s u
s
ing
t
h
e pri
n
cip
l
e o
f
Naï
v
e-Bayes,
t
h
e co
nd
itio
n
a
l pro
b
a
b
ility
to
ev
en
ts m
u
tu
ally in
d
e
p
e
nd
en
t. For ex
am
p
l
e g
i
v
e
n
th
e task o
f
classificatio
n wh
ich con
s
ists o
f
t
w
o classes,
whe
r
e each cl
ass is recogni
zed or
determined by consi
d
eri
ng three f
eatures, the
Naïve-Bayes classifier
d
e
term
in
es an
ev
en
t or ite
m
i
n
to
th
e p
a
rticu
l
ar class b
y
first calcu
latin
g
th
e p
r
o
b
a
b
ility t
h
at an
ite
m
g
o
e
s in
to
each a give
n class, in this suppose that c
1
and c
2
are class, th
en
th
e p
r
obab
ilities d
e
n
o
t
ed
b
y
Pr (c
1
|
x
)
and Pr
(c
2
|
x
)
.
Su
pp
ose
i
t
e
m
x has t
h
e val
u
e o
f
a part
i
c
ul
ar feat
u
r
e, i
e
f
1
, f
2
and f
3
. The
n
t
h
e cal
cu
l
a
t
i
on i
s
based
on t
h
e
assum
p
tion
of
inde
pende
n
ce
of each feat
ure
of a
n
item
x, Pr (c
i
|
x
) is
calcul
a
ted
by the
form
ula
:
Pr(c
1
|x
) =
Pr
(c
1
|f
1
)* Pr
(c
2
|f
2
)
*
Pr(c
1
|f
3
) a
n
d
Pr(c
2
|x
) =
Pr
(c
2
|f
1
)* Pr
(c
2
|f
2
)
*
Pr(c
2
|f
3
)
(3
)
If Pr
(c
1
|x
) >
P
r
(c
2
|
x
) t
h
e
n
the ite
m
x would be
classified a
s
a m
e
m
b
er of
the class c
1
an
d vi
ce versa
.
If there a
r
e more tha
n
two c
l
asses,
then the ite
m will
be
placed into a class with
the greatest proba
b
ility.
Naï
v
e B
a
y
e
s cl
assi
fi
er has t
w
o p
r
i
m
ary
out
put
s, i
e
post
e
ri
o
r
m
a
t
r
i
x
and
ve
ct
or p
r
edi
c
t
i
o
n.
M
a
t
r
i
x
cont
ai
ns t
h
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:20
88-
870
8
I
J
ECE
Vo
l. 6
,
N
o
. 5
,
O
c
tob
e
r
20
16
:
216
7–2
175
2
170
p
r
ob
ab
ility th
at so
m
e
ite
m
s
t
o
b
e
classi
fied
to
b
e
lon
g
to each
class, an
d t
h
e
p
r
ed
ictio
n
vecto
r
is a
v
ect
o
r
th
at
expresse
d t
h
e c
l
ass chose
n
for that item
s
.
3.
PROP
OSE
D
METHO
D
In
th
is st
u
d
y
t
h
e classification
of woo
d
b
a
sed
on
its qu
ality b
y
first ap
p
l
yin
g
edg
e
d
e
tectio
n
on
th
e
ori
g
inal
wood im
age. Edge
detecti
on operators use
d
in ou
r propose
d
m
e
thod
a
r
e
R
obe
rts, Sobel, Prewitt,
C
a
nny
an
d L
o
G (
w
i
t
h
som
e
param
e
t
e
r val
u
e)
. Ext
r
act
i
o
n
of t
e
xt
ure f
e
a
t
ures
per
f
o
r
m
e
d usi
ng
GLC
M
and
Naïv
e Bayes is ad
op
ted
fo
r cl
assifyin
g
th
e differen
t
qu
aliti
es o
f
woo
d
. Th
e p
r
op
osed
meth
od
is su
mmarized
on
t
h
e
di
ag
ram
i
n
Fi
gu
re
4
bel
o
w:
Fi
gu
re
4.
Sc
he
m
e
of t
h
e
pr
op
ose
d
m
e
t
hod
In t
h
i
s
m
e
t
hod
, we use
fo
u
r
t
e
xt
u
r
al
feat
ur
e
s
, (i
.e. C
o
nt
ras
t
, C
o
rrel
a
t
i
o
n,
Ener
gy
an
d H
o
m
ogenei
t
y
)
deri
ved
fr
om
the c
o
-
o
ccu
rre
n
ce m
a
t
r
i
x
wi
t
h
vari
ou
s val
u
e
s
of
GLC
M
p
a
ram
e
t
e
rs. We
ha
ve f
o
ur
ki
n
d
s
of
wood
quality (as e
xpe
rt
database, see Fi
gure
4),
wher
e
each type is c
o
m
posed of
20 im
ages, whi
c
h a
r
e
di
vi
de
d i
n
t
o
t
w
o
set
s
(i
.e
. t
e
st
set
an
d t
r
ai
n
i
ng set
,
eac
h c
ont
ai
ni
ng
1
0
i
m
ages).
GLC
M
t
e
xt
ure
feat
ures
f
r
om
all images in the training set
are stor
e
d
as a
feature
databas
e
, that is then
use
d
t
o
pe
rf
or
m
t
h
e cl
assi
fi
cat
i
on o
f
th
e i
m
ag
es in
th
e test set. The wood
qu
ality classifica
tio
n
resu
lts ob
tain
ed
b
y
Naiv
e Bayes classifier will b
e
com
p
ared
or
v
e
ri
fi
ed
by
t
h
e e
xpe
rt
cl
assi
fi
cat
i
on res
u
l
t
s
t
h
a
t
are i
n
t
h
e ex
p
e
rt
dat
a
base
. A
nd as
fi
nal
out
put
of
the propose
d
c
l
assification method is the accuration rate
, with
pre
c
ision are
opti
ona
l outputs. Ac
curation rate
co
m
p
u
t
e b
y
d
i
v
i
d
i
ng
th
e
nu
mb
er
o
f
all correct p
r
ed
icted
item
s
in
test set
with
th
e
n
u
m
b
e
r of ite
m
s
in
t
e
st set.
Pr
ecision
co
mp
u
t
e
d
e
p
e
nd
on
th
e qu
er
y
(
i
n ou
r
p
a
p
e
r, th
e po
ssib
l
e
qu
ery
is lab
e
l of
w
o
od
t
y
pe, i
.
e
.
T
F
, F
,
M
and GG).
4.
WOO
D
I
M
AGE
Th
is sectio
n
will b
e
d
i
v
i
d
e
d
in
to
two
p
a
rts. Th
e first part sh
ows d
i
fferen
t
k
i
nd
s of woo
d
qu
ality
b
a
sed
o
n
v
i
su
al tex
t
u
r
e feat
u
r
es pres
en
t in
wood
im
ag
es th
at will
b
e
test
ed
.
Th
e seco
nd
p
a
rt
d
i
scu
sses abou
t
t
h
e res
u
l
t
s
of
v
a
ri
o
u
s e
dge
det
ect
i
on m
e
t
hods
o
n
wo
o
d
i
m
age.
4.
1
Woo
d
imag
e
acquisit
i
o
n
As t
h
e resul
t
s
of
wo
od i
m
ag
e acqui
si
t
i
on, we have
fo
ur t
y
pes of
wo
od
whi
c
h are
di
st
ing
u
i
s
hed
by
th
eir q
u
a
lity. T
h
ese i
m
ag
es ar
e p
r
ov
id
ed
b
y
LE2
I
lab
o
rato
ry, Un
iv
ersité d
e
Bo
u
r
go
gn
e,
Fran
ce. Th
ere
are 8
0
im
ages fro
m
8
0
sa
m
p
les of wood, which are
grouped into
four types of quality
and each group consists of 20
sam
p
l
e
s. Fi
gure 5 sho
w
s an e
x
am
ple of
wo
o
d
im
ages with
fou
r
diffe
rent q
u
alities. These im
ages are grayscale
and i
n
png
format, wherein appears the transv
erse cross
-
sect
i
ons, whi
c
h r
e
veal
t
h
e annu
al
ri
ngs and t
h
e
ray
s
of
woo
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE I
S
SN
:
208
8-8
7
0
8
Woo
d
Cla
ssifica
tio
n Ba
sed on Edg
e
Detection
s
a
n
d
Textu
r
e Fea
t
u
r
es
S
e
lectio
n
(Achma
d
Fah
r
u
r
o
z
i)
2
171
Figure
5. Im
age of f
o
u
r
q
u
alities of w
o
o
d
: (a) very
g
o
o
d
,
(b
) g
o
o
d
, (c)
m
e
d
i
u
m
and (d
) p
o
o
r
4.
2
Edg
e
detection imag
es
The
pu
rp
ose
of
t
h
e ed
ge
det
ect
i
on ap
pl
i
e
d t
o
t
h
e w
o
od
im
ag
e is to
furth
e
r streng
th
en
th
e ap
p
e
aran
ces
of t
h
e
w
o
o
d
t
e
xt
u
r
e feat
u
r
es
cont
ai
ne
d t
h
e
r
ei
n, especi
al
l
y
t
o
obt
ai
n t
h
e fi
ber c
o
m
posi
t
i
on a
nd t
h
e l
i
n
es t
h
at
co
nstru
c
t annual rin
g
s
th
at are co
n
s
id
ered
as p
a
ram
e
ters t
o
d
e
term
in
e th
e q
u
a
lity o
f
wo
od
. Fi
g
u
re 6
sh
ow
s
edge
detection results of se
veral
m
e
thods (i.e. Robe
rts,
S
obel, P
r
e
w
itt, Canny and LoG) a
pplied to
ori
g
inal
w
ood
im
ag
e in
Figu
r
e
5-
b.
Fig
u
re
6
.
Edg
e
d
e
tection
resu
l
t
s o
f
sev
e
ral
o
p
erato
r
s: (a) Rob
e
rts, (b
)
So
b
e
l
,
(c) Prewitt, (d) Cann
y an
d (e)
LoG (with
p
a
ra
m
e
ter
σ
= 0.66
an
d size
5
)
We ca
n
see in
Figure
6 that t
h
e
res
u
lts of
Sobel
and
Prewi
tt o
p
e
rators
h
a
v
e
sim
ilar ch
aracteristics as
m
e
nt
i
oned i
n
[
14]
. Im
age of s
econ
d
o
r
der ed
ge det
ect
i
o
n m
e
t
h
o
d
s ha
ve ch
aract
eri
s
t
i
c
s
m
u
ch
di
ffe
re
nt
t
h
an t
h
e
r
e
su
lts of
th
e f
i
r
s
t o
r
d
e
r. Canny d
e
tecto
r
g
e
ner
a
tes th
e ed
g
e
s in
b
i
n
a
r
y
fo
rmat, th
at can
c
a
p
t
ur
e th
e str
o
n
g
an
d
weak e
d
ges,
but these e
d
ges
are ve
ry
t
h
i
n
com
p
ared t
o
t
h
e ot
hers
o
p
e
r
at
or
s res
u
l
t
s
.
The
out
put
of
LO
G
o
p
e
rator seem
s lik
e th
e
orig
inal i
m
ag
e with
rein
fo
rcem
en
t on
th
e v
i
ew
o
f
fi
b
e
rs.
5.
E X
P
ERIME
N
TAL RESULT
Th
is sectio
n
presen
ts th
e an
alysis o
f
GLCM tex
t
ure feat
ures
, effect of
change GLCM param
e
ters,
and cl
assi
fi
cat
i
on re
sul
t
s
u
s
i
ng
o
u
r
pr
op
os
ed m
e
t
hod.
A
s
expl
ai
ned i
n
t
h
e fi
rst
pa
rt
,
t
h
at
t
h
i
s
st
ud
y
was
conducted to c
l
assify wood
based on
its qua
lity with a focus
on
utilizing the i
m
age of
the test wood. B
ecause
pre
v
i
o
us re
sear
ch rel
a
t
e
d t
o
t
h
e
m
echani
cal
p
r
o
p
ert
i
e
s o
f
wo
od i
s
do
ne
by
usi
n
g ul
t
r
as
o
n
i
c
waves
,
x
-ray
s
and
ot
he
r l
a
b
o
r
at
or
y
equi
pm
ent
(
not
use
i
m
age o
f
wo
o
d
)
,
t
h
e
n
t
h
e c
o
m
p
ari
s
on
ag
ai
nst
t
h
e
res
u
l
t
s
i
n
t
h
i
s
pa
per
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:20
88-
870
8
I
J
ECE
Vo
l. 6
,
N
o
. 5
,
O
c
tob
e
r
20
16
:
216
7–2
175
2
172
aren'
t
do
ne.
R
e
search
o
n
t
h
e
im
age of
w
o
od
i
s
ge
ne
ral
l
y
per
f
o
r
m
e
d t
o
det
ect
w
o
o
d
defect
s
[
1
]
,
[
7
]
or t
o
classif
y
w
ood
b
a
sed
on
sp
eci
es [9
]-
[1
3
]
. Ther
efo
r
e, we pr
esen
t exp
e
r
i
m
e
n
t
al r
e
su
lts of
cl
assif
i
catio
n
o
f
w
ood
dat
a
base
t
h
at
we
use i
n
t
h
i
s
pape
r
were
i
m
pl
em
ent
e
d i
n
s
o
m
e
cl
assi
fi
cati
on m
e
t
hod
b
a
s
e
d
on
w
o
od
s
p
eci
es.
5.
1
Wo
od Cl
as
si
fi
cati
on of
Pr
op
osed Me
th
od
As explained
earlier that GLCM as fea
t
ures extract
or has three parameters, i.e. spatia
l distance
,
di
rect
i
on and
gray
l
i
m
i
t
s
. The sel
ecti
on o
f
di
fferent
va
lues of these para
m
e
ters
can affect the res
u
lts of
statistical that calculate
from co-occurence
matr
icex, th
at
maybe
influence class
i
fica
tio
n results. Some edge
det
ect
i
on
m
e
t
hods al
so co
nsi
d
er t
h
e ori
e
nt
a
t
i
on t
o
wards
t
h
e edges, s
o
i
n
t
h
i
s
paper, c
h
ange
of t
h
ese
t
h
ree
param
e
ters are considered for
texture features
extraction
that use as classification f
eatures. Table 1
presents the
differences of
classific
a
tion result (as measured
by accuration ra
te, precision, and running time) that
di
st
i
ngui
shed
b
y
t
h
e use
of
ed
ge det
ect
i
on m
e
t
hods a
n
d
t
h
e
sel
ect
i
on of s
p
at
i
a
l
di
st
ance. Accu
rat
i
on rat
e
and
preci
si
on are
present
e
d i
n
p
e
rcent
,
and r
u
nni
n
g
t
i
m
e
i
n
second.
In t
h
i
s
paper, we
p
r
esent
t
h
e average of
preci
si
on o
f
al
l
possi
bl
e q
u
eri
e
s (i
.e. TF, F,
M
and G
G
).
Tabl
e 1. C
l
assifi
cat
i
on R
e
sul
t
s vs Spat
i
a
l
Di
st
an
ce Val
u
e B
a
sed o
n
Vari
o
u
s
Edge
Det
ect
i
o
n M
e
t
hods
Oper
ator E
v
aluation
Spatial Distance
1 2
3
4
5
Rober
t
s
Accur
a
tion Rate
85
85
85
82.
5
85
Pr
ecision 87.
412
6
88.
311
7
88.
311
7
86.
607
1
88.
311
7
Runnin
g
T
i
m
e
13.
699
7
13.
715
6
13.
708
0
13.
846
2
13.
695
3
Sobel
Accur
a
tion Rate
82.
5
87.
5
82.
5
85
82.
5
Pr
ecision 86.
417
7
89.
393
9
86.
417
7
88.
690
5
87.
087
9
Runnin
g
T
i
m
e
44.
890
2
44.
944
2
44.
889
0
44.
817
0
44.
879
0
Prewitt
Accur
a
tion Rate
82.
5
87.
5
82.
5
85
85
Pr
ecision 86.
417
7
89.
393
9
86.
417
7
88.
690
5
88.
690
5
Runnin
g
T
i
m
e
44.
897
1
44.
806
7
44.
890
0
44.
842
2
44.
832
1
Canny
Accur
a
tion Rate
80
75
52.
5
52.
5
47.
5
Pr
ecision 80.
524
0
77.
638
0
50.
000
0
54.
155
8
46.
969
7
Runnin
g
T
i
m
e
23.
556
5
21.
998
9
22.
165
8
22.
182
0
22.
060
3
Lo
G
Accur
a
tion Rate
77.
5
82.
5
77.
5
80
80
Pr
ecision 77.
478
4
83.
018
6
77.
061
7
81.
095
6
82.
359
3
Runnin
g
T
i
m
e
18.
274
9
18.
229
5
18.
164
4
18.
195
9
18.
109
3
In Table 1 it can be seen that cha
nges i
n
t
h
e
val
u
e of s
p
at
i
a
l di
st
ance param
e
ter doesn'
t
gi
ve a si
gni
fi
cant
effect on the
classification results, wh
ich
is characterize
d
by a constant
percentage
of accuration
rate and
preci
si
on, except
for C
a
nny
operat
o
r
.
W
h
i
l
e
t
h
e
im
pact
o
f
changes t
o
t
h
e
G
against
classification results are
present
e
d i
n
Ta
bl
e 2.
Tabl
e 2. C
l
assifi
cat
i
on R
e
sul
t
vs Gray
Li
m
i
t
s
Val
u
e B
a
sed o
n
Vari
o
u
s E
d
g
e
Det
ect
i
on M
e
t
hods
Operator
Hal
yang
Dia
m
ati
Gra
y
li
m
its
8 16
32
64
128
Rober
t
s
Accur
a
tion Rate
85
82.
5
82.
5
82.
5
82.
5
Pr
ecision 87.
412
6
86.
039
0
86.
417
7
86.
417
7
86.
417
7
Runnin
g
T
i
m
e
13.
699
7
14.
773
6
14.
980
4
15.
028
0
15.
436
7
Sobel
Accur
a
tion Rate
82.
5
82.
5
82.
5
85
80
Pr
ecision 86.
417
7
86.
417
7
86.
417
7
87.
791
4
79.
711
5
Runnin
g
T
i
m
e
44.
890
2
45.
957
2
45.
914
6
45.
872
0
46.
283
Prewitt
Accur
a
tion Rate
82.
5
82.
5
82.
5
82.
5
82.
5
Pr
ecision 86.
417
7
86.
417
7
86.
417
7
86.
417
7
86.
417
7
Runnin
g
T
i
m
e
44.
897
1
45.
835
3
45.
781
8
45.
880
8
46.
340
3
Lo
G
Accur
a
tion Rate
77.
5
77.
5
77.
5
80
80
Pr
ecision 77.
478
3
77.
478
4
77.
478
4
80.
833
3
80.
833
3
Runnin
g
T
i
m
e
18.
274
9
18.
376
8
18.
197
4
18.
339
0
18.
641
4
B
a
sed on t
h
e resul
t
s
obt
ai
ned i
n
Tabl
e 1 and Tabl
e 2,
we can sum
m
ari
ze t
h
at
R
obert
s operat
o
r
generally provi
des the best classifi
ca
tion results, both in term
s
of accura
tion rate, precision,
running time and
the stability of level of
accu
racy. Change the param
e
ters
of spatia
l dis
t
ance and gra
y
limits not provide a
considerable influence on the
accurati
on rate
and precision,
except on Canny
operator. T
hus, the best value for
a param
e
t
e
r G
t
h
at
can be select
ed i
s
8, as
m
e
nt
i
oned i
n
[19]
wi
t
h
t
h
e i
n
t
e
nt
i
on of red
u
c
i
ng t
h
e co
m
put
at
i
on,
because the size of co-occurre
nce
matrix is s
m
aller than the cu
rre
nt G value greater than
8.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE I
S
SN
:
208
8-8
7
0
8
Woo
d
Cla
ssifica
tio
n Ba
sed on Edg
e
Detection
s
a
n
d
Textu
r
e Fea
t
u
r
es
S
e
lectio
n
(Achma
d
Fah
r
u
r
o
z
i)
2
173
Furt
her
a
n
al
y
s
i
s
carri
e
d
out
o
n
t
h
e a
ngl
e
o
r
i
e
nt
at
i
o
n
p
a
ram
e
t
e
r and
R
obe
rt
s
ope
rat
o
r
,
due
o
n
l
y
R
obe
rt
s o
p
erat
or
usi
n
g a
di
ago
n
al
di
rect
i
o
n
t
o
t
h
e det
e
rm
inat
i
on
o
f
t
h
e e
dge a
n
d gi
ve t
h
e be
st
cl
assi
fi
cat
i
o
n
resu
lt con
s
tan
tly. Th
en
research
d
i
d
b
y
usin
g
th
e features GLCM wit
h
th
e
p
a
ram
e
t
e
rs ob
tain
ed
fro
m
a
di
ag
onal
di
rect
i
on,
i
e
45
an
d
13
5
de
grees
.
R
e
sul
t
s
o
f
cl
as
si
fi
cat
i
on
by
u
s
i
n
g
t
h
e i
m
age o
f
R
o
be
rt
s
op
erat
o
r
sho
w
n i
n
Ta
bl
e 3.
Tabl
e
3. T
h
e
E
ffect
of
C
h
a
ngi
ng
F
eatures
Se
lection on R
oberts Im
age
Dir
ection
Spatial Distance
1 2
0,
45,
90 and 135
85
85
45 and 13
5
87.
5
85
It can be seen
from
Table 3 that
t
h
e cl
assi
fi
cat
i
on res
u
l
t
s
us
i
ng t
h
e feat
ure
s
from
t
h
e di
rect
i
on 4
5
an
d
135 degrees with
the
spatial distance
1
provides
a better accuration rate
than
using all
the features that come
fr
om
t
h
e fo
ur
d
i
rect
i
ons
o
f
GL
C
M
on
R
o
bert
s i
m
age.
5.
2
Wo
od Cl
as
si
fi
cati
on of
O
t
he
r
Me
th
od
C
l
assi
fi
cat
i
on m
e
t
hod u
s
ed
h
e
re i
s
t
h
e cl
assi
fi
cat
i
on of
wo
od
base
d on s
p
eci
es, pr
op
ose
d
by
M
oha
n
S.
i
n
[
12]
. Thi
s
m
e
t
hod
use
s
C
a
nny
ope
rat
o
r
at
p
r
e-
p
r
oces
si
ng st
age
a
n
d uses
t
h
ree
t
e
xt
ure feat
u
r
es of
GLC
M
i
e
C
ont
rast
, E
n
t
r
o
p
y
,
an
d St
an
dar
d
De
vi
at
i
o
n
.
The n
u
m
b
er of fe
at
ure
s
use
d
i
s
30
, beca
us
e i
t
uses bl
oc
k im
age
from
the origi
n
al im
age. Cla
ssifier is used
Pears
on c
o
rr
elation factor. T
o
determ
ine
the class of an image in
test set (herein
a
fter refe
rre
d to test sam
p
le), calcula
ted
th
e co
rrelation
facto
r
b
e
t
w
een
t
h
e test sa
m
p
le
with
each im
age in
training set.
Suppose t
h
e
highest c
o
rrelation
factor is
with a
n
im
age include
d i
n
the
class K,
then test sa
m
p
le classified as wood
of cla
ss K. Th
is m
e
th
od
lo
ok
s like k
-
NN classi
fier with
correlatio
n
distance a
n
d
k
= 1
.
Th
e d
a
tab
a
se an
d th
e t
e
st set d
e
scri
bed
o
n
t
h
ird
part in
th
is
p
a
p
e
r im
p
l
e
m
en
ted
to
th
is
m
e
t
hod t
h
en c
o
m
p
ared t
h
e
re
sul
t
s
wi
t
h
o
u
r
pr
o
pose
d
m
e
t
hod
. C
l
assi
fi
cat
i
on
res
u
l
t
s
o
f
t
h
i
s
m
e
t
hod
pre
s
ent
e
d
in
Tab
l
e 4, where ite
m
s
in
th
e train
i
ng
set be n
u
m
b
e
red
fro
m
0
1
to
40
an
d
item
s
in
th
e test set b
e
n
u
m
b
ered
fr
om
41 t
o
8
0
.
Tab
l
e 4
.
Im
p
l
emen
tatio
n
Resu
lts
of D
a
ta
of Wood
-Based
Q
u
ality on
M
o
h
a
n
'
s Classificatio
n
Method
Sa
m
p
le
Test
Highest
Correlation
Closest
Neighboor
Prediction
Class
Actual
Class
Sa
m
p
le
Test
Highest
correlation
Closest
Neighboor
Prediction
Class
Actual
Class
41
0.
9999
04
T
F
T
F
61
0.
9998
26
M
M
42
0.
9984
37
GG
T
F
62
0.
9997
01
T
F
M
43
0.
9999
04
T
F
T
F
63
0.
9998
22
M
M
44
0.
9982
37
GG
T
F
64
0.
9994
16
F
M
45
0.
9999
04
T
F
T
F
65
0.
9999
24
M
M
46
0.
9996
05
T
F
T
F
66
0.
9998
23
M
M
47
0.
9995
06
T
F
T
F
67
0.
9995
15
F
M
48
0.
9995
02
T
F
T
F
68
0.
9996
28
M
M
49
0.
9997
03
T
F
T
F
69
0.
9999
24
M
M
50
0.
9974
38
GG
T
F
70
0.
9999
23
M
M
51
0.
9998
19
F
F
71
0.
9995
37
GG
GG
52
0.
9992
15
F
F
72
0.
9998
38
GG
GG
53
0.
9998
13
F
F
73
0.
9994
37
GG
GG
54
0.
9998
19
F
F
74
0.
9998
38
GG
GG
55
0.
9999
19
F
F
75
0.
9988
15
F
GG
56
0.
9996
15
F
F
76
0.
9993
31
GG
GG
57
0.
9998
13
F
F
77
0.
9991
15
F
GG
58
0.
9998
19
F
F
78
0.
9989
31
GG
GG
59
0.
9998
13
F
F
79
0.
9998
37
GG
GG
60
0.
9954
39
GG
F
80
0.
9995
38
GG
GG
Accur
a
tion Rate =
77.
5
Runnin
g
T
i
m
e
= 2
7
.
0946
Pr
ecision = 80.
85
Accur
a
tion r
a
te an
d pr
ecision ar
e in p
e
r
cent,
r
unning tim
e
ar
e in second
In Table
4 a
bove, t
h
e closest
sam
p
le in ques
tion is a
n
obje
ct or im
age in the traini
ng set
that ha
s the
h
i
gh
est correlatio
n
v
a
lu
e ag
ai
n
s
t to
test samp
le co
m
p
ared to
o
t
h
e
r
obj
ects in
th
e trai
n
i
ng set. It can
b
e
seen
that the accura
cy and precision of cl
assification m
e
thod
by M
oha
n S is qui
t
e low, and
no
single type of
wood
that has
a
preci
sion up t
o
100%.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:20
88-
870
8
I
J
ECE
Vo
l. 6
,
N
o
. 5
,
O
c
tob
e
r
20
16
:
216
7–2
175
2
174
6.
CO
NCL
USI
O
N
B
a
sed o
n
t
h
e cl
assi
fi
cat
i
on re
sul
t
s
obt
ai
ne
d,
edge
det
ect
i
on
m
e
t
hod o
f
t
h
e
fi
rst
or
de
r gi
ve
s t
h
e im
age
that characte
r
izes the sprea
d
of
fibe
r which
is good
for eac
h wood
group. Canny are less suitable for use i
n
the case of
wood classification base
d on quality, because
it captures the
weak e
dge
t
h
ough, so obsc
ure the
charact
e
r
i
s
t
i
c
that
can
di
st
i
n
g
u
i
s
h am
on
g t
y
pes o
f
wo
o
d
. P
r
e-
pr
ocessi
ng
usi
n
g e
dge
det
ect
i
on m
e
t
hods
i
n
o
u
r
pr
o
pose
d
m
e
t
h
od
, especi
al
l
y
t
h
e fi
rst
orde
r, gi
ve a fai
r
l
y
goo
d cl
assi
f
i
cat
i
on resul
t
s
.
R
obert
s o
p
er
at
or i
s
relatively
m
o
re
efficient beca
use of
t
h
e n
u
m
b
er
of
feat
ures
use
d
can
be
re
duce
d
t
o
hal
f
a
nd
p
r
o
v
i
d
e t
h
e
bet
t
e
r
accuration
rate on
spatial dist
ance of
1, a
n
d gene
rally its
perform
a
nce is com
p
arable
to othe
rs detect
or that
use m
o
re
feat
ures
.
Wo
o
d
c
l
assi
fi
cat
i
on
m
e
t
hod
base
d
o
n
s
p
eci
es
b
y
M
oha
n
S.
l
e
ss s
u
i
t
a
bl
e i
f
i
t
i
s
i
m
p
l
e
m
en
ted
to
th
e
d
a
ta in
t
h
is stud
y th
at is wood
-b
as
ed
q
u
a
lity. It is alleg
e
d
th
e
u
s
e
of Cann
y op
erat
o
r
and
because the
da
ta characte
r
istics of
w
ood-bas
ed s
p
asies
not t
h
e sam
e
as th
e
data c
h
aracteri
s
tics of wood-base
d
q
u
a
lity th
at u
s
ed
in
th
is study. Th
e selection
of featur
es t
h
rou
g
h
th
e
d
e
term
in
atio
n
o
f
th
e p
a
ram
e
ters th
at
construct
GLC
M
also very i
m
portant beca
use quite a
ffec
t
s the accuracy
and
precisi
on
from
the classification
process
,
es
pecially that use R
obe
rt
s ope
rat
o
r
i
n
pre
-
p
r
ocess
i
ng st
age.
REFERE
NC
ES
[1]
R. W. Atmaja
, et al.
, “The Detection of Straight and Slant
Wood Fiber Th
rough Slop Ang
l
e Fiber Featur
e,”
TELKOMNIKA Indonesian Journ
a
l of
Electrical
Engineering,
vol/issue: 14(2)
, pp
. 318-322, 2015.
[2]
W. Lin and J. Wu, “Nondestructive Testing of
Wood Defects Based on Stre
ss Wa
ve Techno
log
y
,”
TE
LKOMNIKA
Indonesian Jour
nal of Elec
trical Engineering,
vol/issue: 11(11)
, p
p
. 6802-6807
, 2
013.
[3]
T. R
.
Mard
ikan
t
o
, e
t
a
l
.
, “
Sifa
t Mekanis
Ka
yu
,”
PT Penerbit IPB
Press, 2011.
[4]
L. Karlin
asari, “Non-destructiv
e ultr
asonic testin
g method for determining
bending strength properties of Gmelina
wood (Gmelina
arborea)
,”
Journ
a
l of Tropical
Forest Science,
vo
l. 20
, pp
. 99-104
, 2008
.
[5]
J
.
Baan,
et al,
“Prediction of Mechanical Propert
ies-Modulus of
Ru
pture and Modulus
of Elasticity
-
of Five Tropical
S
p
ecies
b
y
Nond
es
truct
i
ve M
e
tho
d
s
,
”
M
a
de
ra
s,
Ci
e
n
ci
a
y Te
c
n
ol
og
i
a
, vo
l/issue: 1
7
(2), pp
. 239-25
2, 2015
.
[6]
V.
Bucur,
“
Nond
estructive Chara
c
terization
and I
m
aging of Wood
,” Spring
er-Verlag Ber
lin
Heidelberg, 2003
.
[7]
H. I. A
.
Wibo
wo and F. Mer
i
audeau, “Rule-
Based Wood Knot Defect
Image Classificatio
n,”
Proc.
Mast
ers
Erasmus Mundus in Vision
&
Ro
botics M
eeting
Day
, pp
. 87-91
,
2010.
[8]
P. Mohanaiah
,
et al.
, “Image Textur
e Featur
e Extraction Using GLCM App
r
oach,”
International Journal o
f
Scien
tifi
c and
R
e
search Publ
icat
i
ons,
vol/issue: 3(
5), 2013
.
[9]
P
r
as
eti
y
o,
et al.
, “A Co
mparativ
e Stud
y
of Featu
r
e Extr
acti
on Methods for Wood Textur
e Classification,”
Proc. o
f
the 6th In
ternational Conferen
ce on Signal Im
age
Technolog
y and Internet
Based Systems (
S
ITIS)
, pp. 23-29,
2010.
[10]
J.
Y.
Tou,
et al.
, “Rotational inv
a
rian
t wood species r
ec
ognition
through wood species verif
i
cation,”
Proc. o
f
t
h
e
1st Asian Conference on Intel
lig
ent Information
and Dat
abase Systems, (
D
S ‘09
)
, IEEE
Xplore P
r
ess, Dong Hoi
,
pp. 115-120
, 20
09.
[11]
Brem
ananth R.
,
et al.
, “Wood Species Recog
n
ition S
y
st
em
,”
International Journal of
Computer, Electrical,
Automation
,
Co
ntrol and In
formation
Engineerin
g
, vol/issue: 3(4)
, 2009
.
[12]
S. Mohan,
et
al.
, “An Intellig
ent
Recogn
ition
S
y
stem
fo
r Ident
i
fi
cat
ion of Wood
Species,”
Journal of Computer
Scien
c
e
, vol/issue: 10(7)
, pp
. 123
1-1237, 2014
.
[13]
S. Mohan,
et al.
, “Wood Species Identification
S
y
stem,”
Inter
national
Journal of Engin
eerin
g and Computer
Scien
c
e
, vol/issue: 3(5)
, pp
. 5996
-6001, 2014
[14]
M. Juneja and S. Shandu, “Perfor
m
an
ce Evalu
a
tio
n of Edge Detection Techni
ques
for Images in Sp
atial Domain,”
International Jo
urnal of Com
puter Theory amd
Engineering
,
vol/issue: 1(5), pp. 6
41–621, 2009
.
[15]
R. C. Gonzalez
and R.
E. Woods, “D
igital Image
Processing Third
Edition,”
New
York, Pearson P
r
entice H
a
ll, 200
8.
[16]
R. Maini and H. Agwaral, “
S
tud
y
and Com
a
pri
s
on of Various
Image Edge
Det
ect
ion Techn
i
qu
e,”
International
Journal of Image Processing (
I
JIIP)
,
vol/issue: 3
(
1), 2009
.
[17]
M. Venkatar
a
mana,
et al
.,
“A Review of Recent Text
u
r
e Classification: Meth
ods,”
IOSR Jou
r
nal of Computer
Engineering (
I
OSR-JCE)
, vo
l/issue: 14(1)
, pp
. 54
-60, 2013
.
[18]
R. M. Haralick,
et al
.
,
“
T
extua
l
F
eatures
of I
m
ages
Clas
s
i
fic
a
tion
,
”
IEEE Transaction on Systems, Man and
Cybernatics
, vol/issue: 3(6)
, pp
.
610-621, 1973
.
[19]
A. Fahrurozi,
et al.,
“Wood Texture Features Extraction b
y
Using GL
CM Co
mbin
ed With Various Edge Detectio
n
Methods”,
In The 2016 Interna
tional Congress on Theoretical
and Applied M
a
thematics, Physics
&
Chemistry
(
T
he Science 20
16)
, April 2016.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE I
S
SN
:
208
8-8
7
0
8
Woo
d
Cla
ssifica
tio
n Ba
sed on Edg
e
Detection
s
a
n
d
Textu
r
e Fea
t
u
r
es
S
e
lectio
n
(Achma
d
Fah
r
u
r
o
z
i)
2
175
BIOGRAP
HI
ES
OF AUTH
ORS
Achm
ad Fahrurozi – has receiv
e
d B.S. degre
e
i
n
Mathem
atics f
r
om
Universit
y
of Indonesia in
July
2009
, M.S. in Mathematics
from University
of Indon
esia in Januar
y
2012, and the Ph.D
.
degree from Gunadarma Univer
sity
– Indonesia, in
2016. His areas of Intere
st also include
image
processing, coding
theor
y
&
stat
i
s
tics.
Em
ail:
a
c
h
m
ad.fahruroz
i
12
@gm
a
il.com
Prof. Sarifuddin
Madenda - Currently
Dir
ecto
r
of High School of Computer Scien
ce and
Management Jakarta (STMIK Jaka
rta STI&K).
He received
the B.
S
.
degr
ee fro
m
Univers
i
t
y
o
f
Indonesia, in 19
89, the M.S. degree from Institut
e National des
Sciences
Appliquées de
Ly
on
(INSA de Ly
on)
– French, in 19
92, and
the Ph.D
. degree from
University
of B
u
rgund
y
-
French,
in 1995. From 1995 to 1996
, he
was a Research
Asso
ciate at LIESIB
Laborator
y
-
University
of
Burgund
y
.
In 1997, he becam
e a Lec
t
urer i
n
Com
puter Sc
ienc
e Departm
e
nt, Gunadarm
a
University
– In
donesia. From 2002 to 2007, he
was a research
er at Academic Resear
ch
Consortium
on digita
l im
aging
,
video,
audio
and
m
u
ltim
edia (Co
R
IMedia),
Cana
da. His res
earc
h
inter
e
s
t
s
are im
a
g
e proces
s
i
ng: i
m
a
ge com
p
res
s
i
on, color im
agin
g, im
age dat
a
bas
e
and s
earch
ing,
m
e
dical im
age
anal
ysis, im
ple
m
enting of im
age processing al
gorithm
s
on
FPGA for real time
image an
aly
s
is.
Em
ail: sarif@sta
ff.gunadarma.ac.id
Dr. Ern
a
stuti -
has received
B.S. deg
r
ee in
Mathematics f
r
om University of Indon
esia
in
Decem
ber 1985, and the M.S. in
Com
puter Scie
nce from
Universit
y
of Indon
esia, in Jul
y
1994
,
and PhD degrees in Computer Science from Guna
darma University
, Indonesia,
in
April 2008. Sh
e
is
curren
t
l
y
an
as
s
o
cia
t
e
prof
es
s
o
r in th
e f
acul
t
y
of
com
puter s
c
ien
ce
and Inform
atio
n
Engineering, G
unadarma Univ
ersity
. Her
curr
ent
r
e
s
ear
ch in
t
e
res
t
s
in
clude
g
r
aph th
eor
y
and
combinatorial o
p
timization, gr
a
ph-theoretic interconnection
n
e
twor
ks, parallel and distributed
computing,
and
design and
an
alysis of
algor
ithms. Email: er
n
a
s_tuti@
y
a
hoo.co.id
Djati Keram
i
- rece
ived the Doct
or degree in in
fo
rm
atics from
Institut
e
Nationa
l Pol
y
te
chnique de
Toulouse, France, in 1985. He is working at Math
ematics Department Univ
ersity
of Indonesia as
a profes
s
o
r.
His
curren
t
r
e
s
ear
c
h
int
e
res
t
s
in
clu
d
e m
a
them
at
ic
al
m
odeling
in bi
oinform
a
tics
an
d
im
age proc
essin
g
algor
ithm
s
. E
m
ail: d
j
at
ikr@sc
i.ui
.a
c.id
Evaluation Warning : The document was created with Spire.PDF for Python.