<s>
Integral	B-Algorithm
Channel	I-Algorithm
Features	I-Algorithm
(	O
ICF	O
)	O
,	O
also	O
known	O
as	O
ChnFtrs	O
,	O
is	O
a	O
method	O
for	O
object	B-General_Concept
detection	I-General_Concept
in	O
computer	B-Application
vision	I-Application
.	O
</s>
<s>
It	O
uses	O
integral	B-Algorithm
images	I-Algorithm
to	O
extract	O
features	O
such	O
as	O
local	O
sums	O
,	O
histograms	O
and	O
Haar-like	B-Algorithm
features	I-Algorithm
from	O
multiple	O
registered	O
image	O
channels	O
.	O
</s>
<s>
in	O
their	O
work	O
for	O
pedestrian	B-General_Concept
detection	I-General_Concept
,	O
that	O
was	O
first	O
described	O
at	O
the	O
BMVC	O
in	O
2009	O
.	O
</s>
<s>
The	O
features	O
extracted	O
from	O
various	O
channels	O
are	O
called	O
integral	B-Algorithm
channel	I-Algorithm
features	I-Algorithm
.	O
</s>
<s>
Train	O
the	O
AdaBoost	B-Algorithm
classifier	O
.	O
</s>
<s>
used	O
boosting	O
technique	O
which	O
offers	O
faster	O
learning	O
but	O
training	O
could	O
be	O
done	O
with	O
any	O
of	O
the	O
other	O
available	O
methods	O
such	O
as	O
support	B-Algorithm
vector	I-Algorithm
machine	I-Algorithm
.	O
</s>
<s>
Typically	O
,	O
a	O
"	O
channel	O
"	O
refers	O
to	O
a	O
certain	O
component	O
that	O
defines	O
pixel	O
values	O
in	O
a	O
digital	B-Algorithm
image	I-Algorithm
.	O
</s>
<s>
A	O
color	B-Algorithm
image	I-Algorithm
,	O
for	O
example	O
is	O
an	O
aggregate	O
of	O
three	O
channels	O
(	O
red	O
,	O
green	O
and	O
blue	O
)	O
.	O
</s>
<s>
While	O
this	O
definition	O
of	O
a	O
"	O
channel	O
"	O
is	O
widely	O
accepted	O
across	O
various	O
domains	O
,	O
there	O
exists	O
a	O
broader	O
definition	O
in	O
computer	B-Application
vision	I-Application
,	O
which	O
allows	O
one	O
to	O
exploit	O
other	O
features	O
of	O
an	O
image	O
besides	O
the	O
color	O
information	O
.	O
</s>
<s>
A	O
few	O
examples	O
are	O
Gabor	O
filter	O
and	O
difference	B-Algorithm
of	I-Algorithm
Gaussians	I-Algorithm
(	O
DoG	O
)	O
.	O
</s>
<s>
The	O
most	O
popular	O
ones	O
are	O
canny	B-Algorithm
edge	I-Algorithm
detector	I-Algorithm
and	O
gradient	O
magnitude	O
.	O
</s>
<s>
Histogram	B-Algorithm
of	I-Algorithm
oriented	I-Algorithm
gradients	I-Algorithm
(	O
HOG	O
)	O
descriptor	O
is	O
a	O
popular	O
descriptor	O
that	O
was	O
developed	O
by	O
Dalal	O
and	O
Triggs	O
.	O
</s>
<s>
Note	O
that	O
it	O
is	O
computationally	O
less	O
expensive	O
to	O
extract	O
local	O
sums	O
from	O
an	O
integral	B-Algorithm
image	I-Algorithm
rather	O
than	O
from	O
image	O
pixels	O
.	O
</s>
<s>
In	O
fact	O
,	O
Dollár	O
,	O
et	O
al	O
,	O
used	O
integral	B-Algorithm
images	I-Algorithm
in	O
their	O
work	O
to	O
extract	O
features	O
.	O
</s>
<s>
The	O
channels	O
studied	O
were	O
histogram	B-Algorithm
of	I-Algorithm
oriented	I-Algorithm
gradients	I-Algorithm
(	O
HOG	O
)	O
,	O
gradient	O
histogram	O
channel	O
(	O
Hist	O
)	O
,	O
gradient	O
magnitude	O
(	O
Grad	O
)	O
,	O
color	O
channels	O
(	O
RGB	O
,	O
HSV	O
,	O
LUV	O
)	O
and	O
grayscale	O
channel	O
.	O
</s>
<s>
The	O
performance	O
was	O
evaluated	O
in	O
terms	O
of	O
pedestrian	B-General_Concept
detection	I-General_Concept
rates	O
at	O
the	O
reference	O
point	O
of	O
10	O
-	O
4	O
fppw	O
(	O
false	O
positive	O
per	O
window	O
)	O
.	O
</s>
<s>
About	O
30,000	O
first	O
order	O
features	O
were	O
used	O
to	O
train	O
AdaBoost	B-Algorithm
classifier	O
.	O
</s>
<s>
The	O
ChnFtrs	O
+	O
AdaBoost	B-Algorithm
detector	O
was	O
tested	O
on	O
full	O
images	O
from	O
INRIA	O
and	O
Caltech	O
datasets	O
.	O
</s>
<s>
The	O
ICF	O
method	O
(	O
ChnFtrs	O
)	O
has	O
been	O
widely	O
exploited	O
by	O
researchers	O
in	O
Computer	B-Application
Vision	I-Application
after	O
the	O
work	O
was	O
initially	O
published	O
by	O
Dollar	O
et	O
al	O
..	O
</s>
<s>
Work	O
by	O
Zhang	O
et	O
al	O
also	O
exploited	O
integral	B-Algorithm
channel	I-Algorithm
features	I-Algorithm
in	O
developing	O
Informed	O
Haar	O
detector	O
for	O
pedestrian	B-General_Concept
detection	I-General_Concept
.	O
</s>
<s>
It	O
is	O
also	O
important	O
to	O
note	O
that	O
they	O
used	O
informed	O
Haar-like	B-Algorithm
features	I-Algorithm
,	O
which	O
are	O
second	O
order	O
features	O
according	O
to	O
the	O
terminology	O
described	O
in	O
,	O
whereas	O
Dollár	O
et	O
al	O
.	O
</s>
