<s>
Contextual	B-Algorithm
image	I-Algorithm
classification	I-Algorithm
,	O
a	O
topic	O
of	O
pattern	O
recognition	O
in	O
computer	B-Application
vision	I-Application
,	O
is	O
an	O
approach	O
of	O
classification	B-General_Concept
based	O
on	O
contextual	O
information	O
in	O
images	O
.	O
</s>
<s>
Similar	O
as	O
processing	B-Language
language	I-Language
,	O
a	O
single	O
word	O
may	O
have	O
multiple	O
meanings	O
unless	O
the	O
context	O
is	O
provided	O
,	O
and	O
the	O
patterns	O
within	O
the	O
sentences	O
are	O
the	O
only	O
informative	O
segments	O
we	O
care	O
about	O
.	O
</s>
<s>
During	O
the	O
procedure	O
of	O
segmentation	B-Algorithm
,	O
the	O
methods	O
which	O
do	O
not	O
use	O
the	O
contextual	O
information	O
are	O
sensitive	O
to	O
noise	O
and	O
variations	O
,	O
thus	O
the	O
result	O
of	O
segmentation	B-Algorithm
will	O
contain	O
a	O
great	O
deal	O
of	O
misclassified	O
regions	O
,	O
and	O
often	O
these	O
regions	O
are	O
small	O
(	O
e.g.	O
,	O
one	O
pixel	O
)	O
.	O
</s>
<s>
The	O
small	O
regions	O
also	O
can	O
be	O
formed	O
by	O
correct	O
regions	O
rather	O
than	O
noise	O
,	O
and	O
in	O
this	O
case	O
the	O
method	O
is	O
actually	O
making	O
the	O
classification	B-General_Concept
worse	O
.	O
</s>
<s>
This	O
approach	O
is	O
widely	O
used	O
in	O
remote	B-Application
sensing	I-Application
applications	O
.	O
</s>
<s>
This	O
is	O
a	O
two-stage	O
classification	B-General_Concept
process	O
:	O
</s>
<s>
And	O
provide	O
these	O
regions	O
to	O
classifier	B-General_Concept
.	O
</s>
<s>
The	O
classifier	B-General_Concept
uses	O
the	O
grey	O
level	O
and	O
pixel	O
neighbourhood	O
(	O
contextual	O
information	O
)	O
to	O
assign	O
labels	O
to	O
pixels	O
.	O
</s>
<s>
Contextual	O
classification	B-General_Concept
of	O
image	O
data	O
is	O
based	O
on	O
the	O
Bayes	O
minimum	O
error	O
classifier	B-General_Concept
(	O
also	O
known	O
as	O
a	O
naive	B-General_Concept
Bayes	I-General_Concept
classifier	I-General_Concept
)	O
.	O
</s>
<s>
Apply	O
the	O
minimum	O
error	O
classification	B-General_Concept
on	O
a	O
pixel	O
,	O
if	O
the	O
probability	O
of	O
a	O
class	O
being	O
presenting	O
the	O
pixel	O
is	O
the	O
highest	O
among	O
all	O
,	O
then	O
assign	O
as	O
its	O
class	O
.	O
</s>
<s>
The	O
contextual	O
classification	B-General_Concept
rule	O
is	O
described	O
as	O
below	O
,	O
it	O
uses	O
the	O
feature	O
vector	O
rather	O
than	O
.	O
</s>
<s>
For	O
the	O
classifier	B-General_Concept
uses	O
a	O
vector	O
corresponding	O
to	O
each	O
pixel	O
,	O
and	O
the	O
vector	O
is	O
generated	O
from	O
the	O
pixel	O
's	O
neighbourhood	O
.	O
</s>
<s>
The	O
basic	O
steps	O
of	O
contextual	B-Algorithm
image	I-Algorithm
classification	I-Algorithm
:	O
</s>
<s>
Get	O
the	O
image	O
classification	B-General_Concept
result	O
.	O
</s>
<s>
The	O
template	B-Algorithm
matching	I-Algorithm
is	O
a	O
"	O
brute	O
force	O
"	O
implementation	O
of	O
this	O
approach	O
.	O
</s>
<s>
This	O
method	O
is	O
a	O
top	O
down	O
method	O
and	O
often	O
called	O
table	B-Data_Structure
look-up	I-Data_Structure
or	O
dictionary	B-Application
look-up	I-Application
.	O
</s>
<s>
The	O
Hilbert	B-Algorithm
curve	I-Algorithm
runs	O
in	O
a	O
unique	O
pattern	O
through	O
the	O
whole	O
image	O
,	O
it	O
traverses	O
every	O
pixel	O
without	O
visiting	O
any	O
of	O
them	O
twice	O
and	O
keeps	O
a	O
continuous	O
curve	O
.	O
</s>
