<s>
The	O
Kadir	B-Algorithm
–	I-Algorithm
Brady	I-Algorithm
saliency	I-Algorithm
detector	I-Algorithm
extracts	O
features	O
of	O
objects	O
in	O
images	O
that	O
are	O
distinct	O
and	O
representative	O
.	O
</s>
<s>
It	O
attempts	O
to	O
be	O
invariant	O
to	O
affine	B-Algorithm
transformations	I-Algorithm
and	O
illumination	O
changes	O
.	O
</s>
<s>
As	O
a	O
result	O
,	O
the	O
Kadir	B-Algorithm
–	I-Algorithm
Brady	I-Algorithm
saliency	I-Algorithm
detector	I-Algorithm
is	O
more	O
capable	O
at	O
object	O
recognition	O
than	O
other	O
detectors	O
whose	O
main	O
focus	O
is	O
on	O
whole	O
image	O
correspondence	O
.	O
</s>
<s>
Many	O
computer	B-Application
vision	I-Application
and	O
image	B-Algorithm
processing	I-Algorithm
applications	O
work	O
directly	O
with	O
the	O
features	O
extracted	O
from	O
an	O
image	O
,	O
rather	O
than	O
the	O
raw	O
image	O
;	O
for	O
example	O
,	O
for	O
computing	O
image	O
correspondences	O
,	O
or	O
for	O
learning	O
object	O
categories	O
.	O
</s>
<s>
Instead	O
of	O
finding	O
a	O
corner	O
,	O
or	O
blob	O
,	O
or	O
any	O
specific	O
shape	O
of	O
region	O
,	O
the	O
Kadir	B-Algorithm
–	I-Algorithm
Brady	I-Algorithm
saliency	I-Algorithm
detector	I-Algorithm
looks	O
for	O
regions	O
which	O
are	O
locally	O
complex	O
,	O
and	O
globally	O
discriminative	O
.	O
</s>
<s>
To	O
summarize	O
:	O
Affine	O
invariant	O
saliency	O
detector	O
is	O
invariant	O
to	O
affine	B-Algorithm
transformation	I-Algorithm
and	O
able	O
to	O
detect	O
more	O
generate	O
salient	O
regions	O
.	O
</s>
<s>
In	O
the	O
field	O
of	O
computer	B-Application
vision	I-Application
different	O
feature	O
detectors	O
have	O
been	O
evaluated	O
by	O
several	O
tests	O
.	O
</s>
<s>
The	O
most	O
profound	O
evaluation	O
is	O
published	O
in	O
the	O
International	O
Journal	O
of	O
Computer	B-Application
Vision	I-Application
in	O
2006	O
.	O
</s>
<s>
The	O
following	O
subsection	O
discuss	O
the	O
performance	O
of	O
Kadir	B-Algorithm
–	I-Algorithm
Brady	I-Algorithm
saliency	I-Algorithm
detector	I-Algorithm
on	O
a	O
subset	O
of	O
a	O
test	O
in	O
the	O
paper	O
.	O
</s>
<s>
where	O
A	O
is	O
the	O
locally	O
linearized	O
affine	B-Algorithm
transformation	I-Algorithm
of	O
the	O
homography	O
between	O
the	O
two	O
images	O
,	O
</s>
<s>
The	O
performance	O
of	O
Kadir	B-Algorithm
–	I-Algorithm
Brady	I-Algorithm
saliency	I-Algorithm
detector	I-Algorithm
is	O
inferior	O
to	O
most	O
of	O
other	O
detectors	O
mainly	O
because	O
the	O
number	O
of	O
points	O
detected	O
is	O
usually	O
lower	O
than	O
other	O
detectors	O
.	O
</s>
<s>
N	O
regions	O
are	O
detected	O
on	O
each	O
image	O
of	O
the	O
M	O
images	O
in	O
the	O
dataset	B-General_Concept
.	O
</s>
<s>
Then	O
for	O
a	O
particular	O
reference	O
image	O
,	O
i	O
,	O
the	O
correspondence	O
score	O
is	O
given	O
by	O
the	O
proportion	O
of	O
corresponding	O
to	O
detected	O
regions	O
for	O
all	O
the	O
other	O
images	O
in	O
the	O
dataset	B-General_Concept
,	O
i.e.	O
</s>
<s>
The	O
Kadir	B-Algorithm
–	I-Algorithm
Brady	I-Algorithm
saliency	I-Algorithm
detector	I-Algorithm
gives	O
the	O
highest	O
score	O
across	O
three	O
test	O
classes	O
which	O
are	O
motorbike	O
,	O
car	O
and	O
face	O
.	O
</s>
<s>
In	O
order	O
to	O
test	O
insensitivity	O
to	O
image	O
perturbation	O
the	O
data	B-General_Concept
set	I-General_Concept
is	O
split	O
into	O
two	O
parts	O
:	O
the	O
first	O
contains	O
images	O
with	O
a	O
uniform	O
background	O
and	O
the	O
second	O
images	O
with	O
varying	O
degrees	O
of	O
background	O
clutter	O
.	O
</s>
