<s>
Scale	B-Algorithm
co-occurrence	I-Algorithm
matrix	I-Algorithm
(	O
SCM	O
)	O
is	O
a	O
method	O
for	O
image	O
feature	B-Algorithm
extraction	I-Algorithm
within	O
scale	B-Algorithm
space	I-Algorithm
after	O
wavelet	B-Algorithm
transformation	I-Algorithm
,	O
proposed	O
by	O
Wu	O
Jun	O
and	O
Zhao	O
Zhongming	O
(	O
Institute	O
of	O
Remote	O
Sensing	O
Application	O
,	O
China	O
)	O
.	O
</s>
<s>
In	O
practice	O
,	O
we	O
first	O
do	O
discrete	O
wavelet	B-Algorithm
transformation	I-Algorithm
for	O
one	O
gray	O
image	O
and	O
get	O
sub	O
images	O
with	O
different	O
scales	O
.	O
</s>
<s>
Last	O
we	O
use	O
selected	O
functions	O
(	O
such	O
as	O
Harris	O
statistical	O
approach	O
)	O
to	O
calculate	O
measurements	O
with	O
SCM	O
and	O
do	O
feature	B-Algorithm
extraction	I-Algorithm
and	O
classification	O
.	O
</s>
<s>
One	O
basis	O
of	O
the	O
method	O
is	O
the	O
fact	O
:	O
way	O
texture	O
information	O
changes	O
from	O
one	O
scale	O
to	O
another	O
can	O
represent	O
that	O
texture	O
in	O
some	O
extent	O
thus	O
it	O
can	O
be	O
used	O
as	O
a	O
criterion	O
for	O
feature	B-Algorithm
extraction	I-Algorithm
.	O
</s>
<s>
The	O
matrix	O
captures	O
the	O
relation	O
of	O
features	O
between	O
different	O
scales	O
rather	O
than	O
the	O
features	O
within	O
a	O
single	O
scale	B-Algorithm
space	I-Algorithm
,	O
which	O
can	O
represent	O
the	O
scale	O
property	O
of	O
texture	O
better	O
.	O
</s>
<s>
Traditional	O
texture	O
analysis	O
can	O
be	O
divided	O
into	O
four	O
major	O
issues	O
:	O
feature	B-Algorithm
extraction	I-Algorithm
,	O
texture	O
discrimination	O
,	O
texture	O
classification	O
and	O
shape	O
from	O
texture(to reconstruct 3D surface geometry from texture information )	O
.	O
</s>
<s>
For	O
tradition	O
feature	B-Algorithm
extraction	I-Algorithm
,	O
approaches	O
are	O
usually	O
categorized	O
into	O
structural	O
,	O
statistical	O
,	O
model	O
based	O
and	O
transform	O
.	O
</s>
<s>
Wavelet	B-Algorithm
transformation	I-Algorithm
is	O
a	O
popular	O
method	O
in	O
numerical	O
analysis	O
and	O
functional	O
analysis	O
,	O
which	O
captures	O
both	O
frequency	O
and	O
location	O
information	O
.	O
</s>
<s>
The	O
discrete	O
wavelet	O
frames	O
is	O
nearly	O
identical	O
to	O
the	O
standard	O
wavelet	B-Algorithm
transform	I-Algorithm
,	O
except	O
that	O
one	O
upsamples	O
the	O
filters	O
,	O
rather	O
than	O
downsamples	O
the	O
image	O
.	O
</s>
<s>
Given	O
an	O
image	O
,	O
the	O
DWF	O
decomposes	O
its	O
channel	O
using	O
the	O
same	O
method	O
as	O
the	O
wavelet	B-Algorithm
transform	I-Algorithm
,	O
but	O
without	O
the	O
subsampling	O
process	O
.	O
</s>
<s>
The	O
decomposition	O
is	O
then	O
continued	O
in	O
the	O
LL	O
channels	O
only	O
as	O
in	O
the	O
wavelet	B-Algorithm
transform	I-Algorithm
,	O
but	O
since	O
the	O
image	O
is	O
not	O
subsampled	O
,	O
the	O
filter	O
has	O
to	O
be	O
upsampled	O
by	O
inserting	O
zeros	O
in	O
between	O
its	O
coefficients	O
.	O
</s>
