<s>
The	O
Kuwahara	B-Algorithm
filter	I-Algorithm
is	O
a	O
non-linear	O
smoothing	O
filter	O
used	O
in	O
image	B-Algorithm
processing	I-Algorithm
for	O
adaptive	O
noise	O
reduction	O
.	O
</s>
<s>
Most	O
filters	O
that	O
are	O
used	O
for	O
image	O
smoothing	O
are	O
linear	O
low-pass	B-Algorithm
filters	I-Algorithm
that	O
effectively	O
reduce	O
noise	O
but	O
also	O
blur	O
out	O
the	O
edges	O
.	O
</s>
<s>
However	O
the	O
Kuwahara	B-Algorithm
filter	I-Algorithm
is	O
able	O
to	O
apply	O
smoothing	O
on	O
the	O
image	O
while	O
preserving	O
the	O
edges	O
.	O
</s>
<s>
It	O
is	O
named	O
after	O
Michiyoshi	O
Kuwahara	O
,	O
Ph.D.	O
,	O
who	O
worked	O
at	O
Kyoto	O
and	O
Osaka	O
Sangyo	O
Universities	O
in	O
Japan	O
,	O
developing	O
early	O
medical	B-Application
imaging	I-Application
of	O
dynamic	O
heart	O
muscle	O
in	O
the	O
1970s	O
and	O
80s	O
.	O
</s>
<s>
The	O
arithmetic	O
mean	O
and	O
standard	B-General_Concept
deviation	I-General_Concept
of	O
the	O
four	O
regions	O
centered	O
around	O
a	O
pixel	O
(	O
x	O
,	O
y	O
)	O
are	O
calculated	O
and	O
used	O
to	O
determine	O
the	O
value	O
of	O
the	O
central	O
pixel	O
.	O
</s>
<s>
The	O
output	O
of	O
the	O
Kuwahara	B-Algorithm
filter	I-Algorithm
for	O
any	O
point	O
is	O
then	O
given	O
by	O
where	O
.	O
</s>
<s>
The	O
location	O
of	O
the	O
pixel	O
in	O
relation	O
to	O
an	O
edge	O
plays	O
a	O
great	O
role	O
in	O
determining	O
which	O
region	O
will	O
have	O
the	O
greater	O
standard	B-General_Concept
deviation	I-General_Concept
.	O
</s>
<s>
Similarly	O
to	O
the	O
median	B-Algorithm
filter	I-Algorithm
the	O
Kuwahara	B-Algorithm
filter	I-Algorithm
uses	O
a	O
sliding	O
window	O
approach	O
to	O
access	O
every	O
pixel	O
in	O
the	O
image	O
.	O
</s>
<s>
However	O
,	O
there	O
are	O
variations	O
of	O
the	O
Kuwahara	B-Algorithm
filter	I-Algorithm
that	O
use	O
rectangular	O
windows	O
.	O
</s>
<s>
The	O
main	O
problem	O
with	O
that	O
is	O
that	O
the	O
quadrants	O
will	O
have	O
different	O
standard	B-General_Concept
deviations	I-General_Concept
for	O
each	O
of	O
the	O
channels	O
.	O
</s>
<s>
For	O
example	O
,	O
the	O
upper	O
left	O
quadrant	O
may	O
have	O
the	O
lowest	O
standard	B-General_Concept
deviation	I-General_Concept
in	O
the	O
red	O
channel	O
,	O
but	O
the	O
lower	O
right	O
quadrant	O
may	O
have	O
the	O
lowest	O
standard	B-General_Concept
deviation	I-General_Concept
in	O
the	O
green	O
channel	O
.	O
</s>
<s>
To	O
overcome	O
this	O
problem	O
,	O
for	O
color	O
images	O
a	O
slightly	O
modified	O
Kuwahara	B-Algorithm
filter	I-Algorithm
must	O
be	O
used	O
.	O
</s>
<s>
The	O
filter	O
will	O
produce	O
an	O
output	O
for	O
each	O
channel	O
which	O
will	O
correspond	O
to	O
the	O
mean	O
of	O
that	O
channel	O
from	O
the	O
quadrant	O
that	O
had	O
the	O
lowest	O
standard	B-General_Concept
deviation	I-General_Concept
in	O
"	O
brightness	O
"	O
.	O
</s>
<s>
ImageMagick	B-Language
uses	O
a	O
similar	O
approach	O
,	O
but	O
using	O
the	O
Rec	O
.	O
</s>
<s>
Originally	O
the	O
Kuwahara	B-Algorithm
filter	I-Algorithm
was	O
proposed	O
for	O
use	O
in	O
processing	O
RI-angiocardiographic	O
images	O
of	O
the	O
cardiovascular	O
system	O
.	O
</s>
<s>
The	O
fact	O
that	O
any	O
edges	O
are	O
preserved	O
when	O
smoothing	O
makes	O
it	O
especially	O
useful	O
for	O
feature	B-Algorithm
extraction	I-Algorithm
and	O
segmentation	O
and	O
explains	O
why	O
it	O
is	O
used	O
in	O
medical	B-Application
imaging	I-Application
.	O
</s>
<s>
The	O
Kuwahara	B-Algorithm
filter	I-Algorithm
however	O
also	O
finds	O
many	O
applications	O
in	O
artistic	O
imaging	O
and	O
fine-art	O
photography	O
due	O
to	O
its	O
ability	O
to	O
remove	O
textures	O
and	O
sharpen	O
the	O
edges	O
of	O
photographs	O
.	O
</s>
<s>
These	O
applications	O
have	O
known	O
great	O
success	O
and	O
have	O
encouraged	O
similar	O
research	O
in	O
the	O
field	O
of	O
image	B-Algorithm
processing	I-Algorithm
for	O
the	O
arts	O
.	O
</s>
<s>
Although	O
the	O
vast	O
majority	O
of	O
applications	O
have	O
been	O
in	O
the	O
field	O
of	O
image	B-Algorithm
processing	I-Algorithm
there	O
have	O
been	O
cases	O
that	O
use	O
modifications	O
of	O
the	O
Kuwahara	B-Algorithm
filter	I-Algorithm
for	O
machine	O
learning	O
tasks	O
such	O
as	O
clustering	O
.	O
</s>
<s>
The	O
Kuwahara	B-Algorithm
filter	I-Algorithm
has	O
been	O
implemented	O
in	O
CVIPtools	B-Application
.	O
</s>
<s>
The	O
Kuwahara	B-Algorithm
filter	I-Algorithm
despite	O
its	O
capabilities	O
in	O
edge	O
preservation	O
has	O
certain	O
drawbacks	O
.	O
</s>
<s>
At	O
a	O
first	O
glance	O
it	O
is	O
noticeable	O
that	O
the	O
Kuwahara	B-Algorithm
filter	I-Algorithm
does	O
not	O
take	O
into	O
account	O
the	O
case	O
where	O
two	O
regions	O
have	O
equal	O
standard	B-General_Concept
deviations	I-General_Concept
.	O
</s>
<s>
This	O
is	O
not	O
often	O
the	O
case	O
in	O
real	O
images	O
since	O
it	O
is	O
rather	O
hard	O
to	O
find	O
two	O
regions	O
with	O
exactly	O
the	O
same	O
standard	B-General_Concept
deviation	I-General_Concept
due	O
to	O
the	O
noise	O
that	O
is	O
always	O
present	O
.	O
</s>
<s>
In	O
cases	O
where	O
two	O
regions	O
have	O
similar	O
standard	B-General_Concept
deviations	I-General_Concept
the	O
value	O
of	O
the	O
center	O
pixel	O
could	O
be	O
decided	O
at	O
random	O
by	O
the	O
noise	O
in	O
these	O
regions	O
.	O
</s>
<s>
However	O
,	O
it	O
is	O
not	O
unusual	O
for	O
regions	O
of	O
very	O
different	O
means	O
to	O
have	O
the	O
same	O
standard	B-General_Concept
deviation	I-General_Concept
.	O
</s>
<s>
This	O
makes	O
the	O
Kuwahara	B-Algorithm
filter	I-Algorithm
susceptible	O
to	O
noise	O
.	O
</s>
<s>
Different	O
ways	O
have	O
been	O
proposed	O
for	O
dealing	O
with	O
this	O
issue	O
,	O
one	O
of	O
which	O
is	O
to	O
set	O
the	O
value	O
of	O
the	O
center	O
pixel	O
to	O
in	O
cases	O
where	O
the	O
standard	B-General_Concept
deviation	I-General_Concept
of	O
two	O
regions	O
do	O
not	O
differ	O
more	O
than	O
a	O
certain	O
value	O
.	O
</s>
<s>
The	O
Kuwahara	B-Algorithm
filter	I-Algorithm
is	O
also	O
known	O
to	O
create	O
block	O
artifacts	O
in	O
the	O
images	O
especially	O
in	O
regions	O
of	O
the	O
image	O
that	O
are	O
highly	O
textured	O
.	O
</s>
<s>
The	O
success	O
of	O
the	O
Kuwahara	B-Algorithm
filter	I-Algorithm
has	O
spurred	O
an	O
increase	O
the	O
development	O
of	O
edge-enhancing	O
smoothing	O
filters	O
.	O
</s>
<s>
Several	O
variations	O
have	O
been	O
proposed	O
for	O
similar	O
use	O
most	O
of	O
which	O
attempt	O
to	O
deal	O
with	O
the	O
drawbacks	O
of	O
the	O
original	O
Kuwahara	B-Algorithm
filter	I-Algorithm
.	O
</s>
<s>
The	O
"	O
Generalized	O
Kuwahara	B-Algorithm
filter	I-Algorithm
"	O
proposed	O
by	O
P	O
.	O
Bakker	O
considers	O
several	O
windows	O
that	O
contain	O
a	O
fixed	O
pixel	O
.	O
</s>
<s>
Similarly	O
to	O
the	O
Kuwahara	B-Algorithm
filter	I-Algorithm
it	O
assumes	O
a	O
window	O
of	O
size	O
but	O
instead	O
of	O
searching	O
amongst	O
four	O
subregions	O
of	O
size	O
for	O
the	O
one	O
with	O
minimum	O
variance	O
it	O
searches	O
amongst	O
all	O
possible	O
subregions	O
.	O
</s>
<s>
The	O
"	O
Adaptative	O
Kuwahara	B-Algorithm
filter	I-Algorithm
"	O
,	O
proposed	O
by	O
K	O
.	O
Bartyzel	O
,	O
is	O
a	O
combination	O
of	O
the	O
anisotropic	O
Kuwahara	B-Algorithm
filter	I-Algorithm
and	O
the	O
adaptative	O
median	B-Algorithm
filter	I-Algorithm
.	O
</s>
<s>
In	O
comparison	O
with	O
the	O
standard	O
Kuwahara	B-Algorithm
filter	I-Algorithm
,	O
both	O
the	O
objects	O
and	O
the	O
edges	O
retain	O
a	O
better	O
quality	O
.	O
</s>
<s>
As	O
opposed	O
to	O
the	O
standard	O
Kuwahara	B-Algorithm
filter	I-Algorithm
,	O
the	O
window	O
size	O
is	O
changing	O
,	O
depending	O
on	O
the	O
local	O
properties	O
of	O
the	O
image	O
.	O
</s>
