<s>
Scene	B-Algorithm
text	I-Algorithm
is	O
text	O
that	O
appears	O
in	O
an	O
image	O
captured	O
by	O
a	O
camera	O
in	O
an	O
outdoor	O
environment	O
.	O
</s>
<s>
The	O
detection	O
and	O
recognition	O
of	O
scene	B-Algorithm
text	I-Algorithm
from	O
camera	O
captured	O
images	O
are	O
computer	B-Application
vision	I-Application
tasks	I-Application
which	O
became	O
important	O
after	O
smart	O
phones	O
with	O
good	O
cameras	O
became	O
ubiquitous	O
.	O
</s>
<s>
The	O
recognition	O
of	O
scene	B-Algorithm
text	I-Algorithm
is	O
further	O
complicated	O
sometimes	O
by	O
non-uniform	O
illumination	O
and	O
focus	O
.	O
</s>
<s>
To	O
improve	O
scene	B-Algorithm
text	I-Algorithm
recognition	O
,	O
the	O
International	O
Conference	O
on	O
Document	O
Analysis	O
and	O
Recognition	O
(	O
ICDAR	O
)	O
conducts	O
a	O
robust	O
reading	O
competition	O
once	O
in	O
two	O
years	O
.	O
</s>
<s>
Text	B-Application
detection	I-Application
is	O
the	O
process	O
of	O
detecting	O
the	O
text	O
present	O
in	O
the	O
image	O
,	O
followed	O
by	O
surrounding	O
it	O
with	O
a	O
rectangular	O
bounding	O
box	O
.	O
</s>
<s>
Text	B-Application
detection	I-Application
can	O
be	O
carried	O
out	O
using	O
image	O
based	O
techniques	O
or	O
frequency	O
based	O
techniques	O
.	O
</s>
<s>
In	O
image	O
based	O
techniques	O
,	O
an	O
image	O
is	O
segmented	B-Algorithm
into	O
multiple	O
segments	O
.	O
</s>
<s>
Machine	O
learning	O
approaches	O
such	O
as	O
support	B-Algorithm
vector	I-Algorithm
machine	I-Algorithm
and	O
convolutional	B-Architecture
neural	I-Architecture
networks	I-Architecture
are	O
used	O
to	O
classify	O
the	O
components	O
into	O
text	O
and	O
non-text	O
.	O
</s>
<s>
In	O
frequency	O
based	O
techniques	O
,	O
discrete	B-Algorithm
Fourier	I-Algorithm
transform	I-Algorithm
(	O
DFT	O
)	O
or	O
discrete	O
wavelet	O
transform	O
(	O
DWT	O
)	O
are	O
used	O
to	O
extract	O
the	O
high	O
frequency	O
coefficients	O
.	O
</s>
<s>
Images	O
are	O
not	O
segmented	B-Algorithm
in	O
most	O
of	O
these	O
methods	O
.	O
</s>
<s>
Hence	O
,	O
the	O
top-down	O
approach	O
is	O
sometimes	O
referred	O
as	O
segmentation	B-Algorithm
free	O
recognition	O
.	O
</s>
<s>
In	O
the	O
bottom-up	O
approaches	O
,	O
the	O
image	O
is	O
segmented	B-Algorithm
into	O
multiple	O
components	O
and	O
the	O
segmented	B-Algorithm
image	O
is	O
passed	O
through	O
a	O
recognition	O
engine	O
.	O
</s>
<s>
Either	O
an	O
off	O
the	O
shelf	O
Optical	B-Application
character	I-Application
recognition	I-Application
(	O
OCR	O
)	O
engine	O
or	O
a	O
custom-trained	O
one	O
is	O
used	O
to	O
recognise	O
the	O
text	O
.	O
</s>
