<s>
Handwriting	B-Application
recognition	I-Application
(	O
HWR	O
)	O
,	O
also	O
known	O
as	O
handwritten	B-Application
text	I-Application
recognition	I-Application
(	O
HTR	O
)	O
,	O
is	O
the	O
ability	O
of	O
a	O
computer	O
to	O
receive	O
and	O
interpret	O
intelligible	O
handwritten	O
input	O
from	O
sources	O
such	O
as	O
paper	O
documents	O
,	O
photographs	O
,	O
touch-screens	B-Device
and	O
other	O
devices	O
.	O
</s>
<s>
The	O
image	O
of	O
the	O
written	O
text	O
may	O
be	O
sensed	O
"	O
off	O
line	O
"	O
from	O
a	O
piece	O
of	O
paper	O
by	O
optical	O
scanning	O
(	O
optical	B-Application
character	I-Application
recognition	I-Application
)	O
or	O
intelligent	B-General_Concept
word	I-General_Concept
recognition	I-General_Concept
.	O
</s>
<s>
A	O
handwriting	B-Application
recognition	I-Application
system	I-Application
handles	O
formatting	O
,	O
performs	O
correct	O
segmentation	B-Algorithm
into	O
characters	O
,	O
and	O
finds	O
the	O
most	O
plausible	O
words	O
.	O
</s>
<s>
Offline	O
handwriting	B-Application
recognition	I-Application
involves	O
the	O
automatic	O
conversion	O
of	O
text	O
in	O
an	O
image	O
into	O
letter	O
codes	O
that	O
are	O
usable	O
within	O
computer	O
and	O
text-processing	O
applications	O
.	O
</s>
<s>
Offline	O
handwriting	B-Application
recognition	I-Application
is	O
comparatively	O
difficult	O
,	O
as	O
different	O
people	O
have	O
different	O
handwriting	O
styles	O
.	O
</s>
<s>
And	O
,	O
as	O
of	O
today	O
,	O
OCR	O
engines	O
are	O
primarily	O
focused	O
on	O
machine	O
printed	O
text	O
and	O
ICR	B-Application
for	O
hand	O
"	O
printed	O
"	O
(	O
written	O
in	O
capital	O
letters	O
)	O
text	O
.	O
</s>
<s>
Offline	O
character	B-Application
recognition	I-Application
often	O
involves	O
scanning	O
a	O
form	O
or	O
document	O
.	O
</s>
<s>
Feature	B-Algorithm
extraction	I-Algorithm
works	O
in	O
a	O
similar	O
fashion	O
to	O
neural	O
network	O
recognizers	O
.	O
</s>
<s>
Where	O
traditional	O
techniques	O
focus	O
on	O
segmenting	B-Algorithm
individual	O
characters	O
for	O
recognition	O
,	O
modern	O
techniques	O
focus	O
on	O
recognizing	O
all	O
the	O
characters	O
in	O
a	O
segmented	O
line	O
of	O
text	O
.	O
</s>
<s>
State-of-the-art	O
methods	O
use	O
convolutional	B-Architecture
networks	I-Architecture
to	O
extract	O
visual	O
features	O
over	O
several	O
overlapping	O
windows	O
of	O
a	O
text	O
line	O
image	O
which	O
a	O
recurrent	B-Algorithm
neural	I-Algorithm
network	I-Algorithm
uses	O
to	O
produce	O
character	O
probabilities	O
.	O
</s>
<s>
Online	O
handwriting	B-Application
recognition	I-Application
involves	O
the	O
automatic	O
conversion	O
of	O
text	O
as	O
it	O
is	O
written	O
on	O
a	O
special	O
digitizer	O
or	O
PDA	B-Application
,	O
where	O
a	O
sensor	O
picks	O
up	O
the	O
pen-tip	O
movements	O
as	O
well	O
as	O
pen-up/pen	O
-down	O
switching	O
.	O
</s>
<s>
This	O
kind	O
of	O
data	O
is	O
known	O
as	O
digital	O
ink	B-Application
and	O
can	O
be	O
regarded	O
as	O
a	O
digital	O
representation	O
of	O
handwriting	O
.	O
</s>
<s>
The	O
elements	O
of	O
an	O
online	O
handwriting	B-Application
recognition	I-Application
interface	O
typically	O
include	O
:	O
</s>
<s>
a	O
pen	O
or	O
stylus	B-Device
for	O
the	O
user	O
to	O
write	O
with	O
.	O
</s>
<s>
a	O
software	O
application	O
which	O
interprets	O
the	O
movements	O
of	O
the	O
stylus	B-Device
across	O
the	O
writing	O
surface	O
,	O
translating	O
the	O
resulting	O
strokes	O
into	O
digital	O
text	O
.	O
</s>
<s>
The	O
process	O
of	O
online	O
handwriting	B-Application
recognition	I-Application
can	O
be	O
broken	O
down	O
into	O
a	O
few	O
general	O
steps	O
:	O
</s>
<s>
The	O
second	O
step	O
is	O
feature	B-Algorithm
extraction	I-Algorithm
.	O
</s>
<s>
Commercial	O
products	O
incorporating	O
handwriting	B-Application
recognition	I-Application
as	O
a	O
replacement	O
for	O
keyboard	B-Device
input	O
were	O
introduced	O
in	O
the	O
early	O
1980s	O
.	O
</s>
<s>
With	O
the	O
advent	O
of	O
the	O
large	O
consumer	O
market	O
for	O
personal	O
computers	O
,	O
several	O
commercial	O
products	O
were	O
introduced	O
to	O
replace	O
the	O
keyboard	B-Device
and	O
mouse	O
on	O
a	O
personal	O
computer	O
with	O
a	O
single	O
pointing/handwriting	O
system	O
,	O
such	O
as	O
those	O
from	O
Pencept	O
,	O
CIC	O
and	O
others	O
.	O
</s>
<s>
The	O
first	O
commercially	O
available	O
tablet-type	O
portable	O
computer	O
was	O
the	O
GRiDPad	O
from	O
GRiD	O
Systems	O
,	O
released	O
in	O
September	O
1989	O
.	O
</s>
<s>
Its	O
operating	O
system	O
was	O
based	O
on	O
MS-DOS	B-Application
.	O
</s>
<s>
In	O
the	O
early	O
1990s	O
,	O
hardware	O
makers	O
including	O
NCR	O
,	O
IBM	O
and	O
EO	B-Application
released	O
tablet	B-Device
computers	I-Device
running	O
the	O
PenPoint	B-Application
operating	O
system	O
developed	O
by	O
GO	O
Corp	O
.	O
PenPoint	B-Application
used	O
handwriting	B-Application
recognition	I-Application
and	O
gestures	O
throughout	O
and	O
provided	O
the	O
facilities	O
to	O
third-party	O
software	O
.	O
</s>
<s>
IBM	O
's	O
tablet	B-Device
computer	I-Device
was	O
the	O
first	O
to	O
use	O
the	O
ThinkPad	B-Operating_System
name	O
and	O
used	O
IBM	O
's	O
handwriting	B-Application
recognition	I-Application
.	O
</s>
<s>
This	O
recognition	O
system	O
was	O
later	O
ported	O
to	O
Microsoft	B-Device
Windows	I-Device
for	I-Device
Pen	I-Device
Computing	I-Device
,	O
and	O
IBM	O
's	O
Pen	O
for	O
OS/2	O
.	O
</s>
<s>
Advancements	O
in	O
electronics	O
allowed	O
the	O
computing	O
power	O
necessary	O
for	O
handwriting	B-Application
recognition	I-Application
to	O
fit	O
into	O
a	O
smaller	O
form	O
factor	O
than	O
tablet	B-Device
computers	I-Device
,	O
and	O
handwriting	B-Application
recognition	I-Application
is	O
often	O
used	O
as	O
an	O
input	O
method	O
for	O
hand-held	O
PDAs	B-Application
.	O
</s>
<s>
The	O
first	O
PDA	B-Application
to	O
provide	O
written	O
input	O
was	O
the	O
Apple	B-Device
Newton	I-Device
,	O
which	O
exposed	O
the	O
public	O
to	O
the	O
advantage	O
of	O
a	O
streamlined	O
user	O
interface	O
.	O
</s>
<s>
By	O
the	O
time	O
of	O
the	O
release	O
of	O
the	O
Newton	B-Device
OS	I-Device
2.0	O
,	O
wherein	O
the	O
handwriting	B-Application
recognition	I-Application
was	O
greatly	O
improved	O
,	O
including	O
unique	O
features	O
still	O
not	O
found	O
in	O
current	O
recognition	O
systems	O
such	O
as	O
modeless	O
error	O
correction	O
,	O
the	O
largely	O
negative	O
first	O
impression	O
had	O
been	O
made	O
.	O
</s>
<s>
After	O
discontinuation	O
of	O
Apple	B-Device
Newton	I-Device
,	O
the	O
feature	O
was	O
incorporated	O
in	O
Mac	O
OS	O
X	O
10.2	O
and	O
later	O
as	O
Inkwell	B-Application
.	O
</s>
<s>
Palm	O
later	O
launched	O
a	O
successful	O
series	O
of	O
PDAs	B-Application
based	O
on	O
the	O
Graffiti	B-Application
recognition	O
system	O
.	O
</s>
<s>
Graffiti	B-Application
improved	O
usability	O
by	O
defining	O
a	O
set	O
of	O
"	O
unistrokes	O
"	O
,	O
or	O
one-stroke	O
forms	O
,	O
for	O
each	O
character	O
.	O
</s>
<s>
The	O
Graffiti	B-Application
handwriting	B-Application
recognition	I-Application
was	O
found	O
to	O
infringe	O
on	O
a	O
patent	O
held	O
by	O
Xerox	O
,	O
and	O
Palm	O
replaced	O
Graffiti	B-Application
with	O
a	O
licensed	O
version	O
of	O
the	O
CIC	O
handwriting	B-Application
recognition	I-Application
which	O
,	O
while	O
also	O
supporting	O
unistroke	O
forms	O
,	O
pre-dated	O
the	O
Xerox	O
patent	O
.	O
</s>
<s>
A	O
Tablet	B-Device
PC	I-Device
is	O
a	O
notebook	O
computer	O
with	O
a	O
digitizer	B-Device
tablet	I-Device
and	O
a	O
stylus	B-Device
,	O
which	O
allows	O
a	O
user	O
to	O
handwrite	O
text	O
on	O
the	O
unit	O
's	O
screen	O
.	O
</s>
<s>
Windows	B-Application
Vista	I-Application
and	O
Windows	B-Device
7	I-Device
include	O
personalization	O
features	O
that	O
learn	O
a	O
user	O
's	O
writing	O
patterns	O
or	O
vocabulary	O
for	O
English	O
,	O
Japanese	O
,	O
Chinese	O
Traditional	O
,	O
Chinese	O
Simplified	O
and	O
Korean	O
.	O
</s>
<s>
This	O
system	O
is	O
distinct	O
from	O
the	O
less	O
advanced	O
handwriting	B-Application
recognition	I-Application
system	I-Application
employed	O
in	O
its	O
Windows	B-Operating_System
Mobile	I-Operating_System
OS	O
for	O
PDAs	B-Application
.	O
</s>
<s>
Although	O
handwriting	B-Application
recognition	I-Application
is	O
an	O
input	O
form	O
that	O
the	O
public	O
has	O
become	O
accustomed	O
to	O
,	O
it	O
has	O
not	O
achieved	O
widespread	O
use	O
in	O
either	O
desktop	O
computers	O
or	O
laptops	O
.	O
</s>
<s>
It	O
is	O
still	O
generally	O
accepted	O
that	O
keyboard	B-Device
input	O
is	O
both	O
faster	O
and	O
more	O
reliable	O
.	O
</s>
<s>
,	O
many	O
PDAs	B-Application
offer	O
handwriting	O
input	O
,	O
sometimes	O
even	O
accepting	O
natural	O
cursive	O
handwriting	O
,	O
but	O
accuracy	O
is	O
still	O
a	O
problem	O
,	O
and	O
some	O
people	O
still	O
find	O
even	O
a	O
simple	O
on-screen	B-Device
keyboard	I-Device
more	O
efficient	O
.	O
</s>
<s>
Early	O
software	O
could	O
understand	O
print	O
handwriting	O
where	O
the	O
characters	O
were	O
separated	O
;	O
however	O
,	O
cursive	O
handwriting	O
with	O
connected	O
characters	O
presented	O
Sayre	B-General_Concept
's	I-General_Concept
Paradox	I-General_Concept
,	O
a	O
difficulty	O
involving	O
character	O
segmentation	B-Algorithm
.	O
</s>
<s>
In	O
the	O
early	O
1990s	O
,	O
two	O
companies	O
ParaGraph	O
International	O
and	O
Lexicus	O
came	O
up	O
with	O
systems	O
that	O
could	O
understand	O
cursive	O
handwriting	B-Application
recognition	I-Application
.	O
</s>
<s>
The	O
ParaGraph	O
CalliGrapher	O
system	O
was	O
deployed	O
in	O
the	O
Apple	B-Device
Newton	I-Device
systems	O
,	O
and	O
Lexicus	O
Longhand	O
system	O
was	O
made	O
available	O
commercially	O
for	O
the	O
PenPoint	B-Application
and	O
Windows	O
operating	O
system	O
.	O
</s>
<s>
Lexicus	O
was	O
acquired	O
by	O
Motorola	O
in	O
1993	O
and	O
went	O
on	O
to	O
develop	O
Chinese	O
handwriting	B-Application
recognition	I-Application
and	O
predictive	B-General_Concept
text	I-General_Concept
systems	O
for	O
Motorola	O
.	O
</s>
<s>
ParaGraph	O
was	O
acquired	O
in	O
1997	O
by	O
SGI	O
and	O
its	O
handwriting	B-Application
recognition	I-Application
team	O
formed	O
a	O
P&I	O
division	O
,	O
later	O
acquired	O
from	O
SGI	O
by	O
Vadem	O
.	O
</s>
<s>
Microsoft	O
has	O
acquired	O
CalliGrapher	O
handwriting	B-Application
recognition	I-Application
and	O
other	O
digital	O
ink	B-Application
technologies	O
developed	O
by	O
P&I	O
from	O
Vadem	O
in	O
1999	O
.	O
</s>
<s>
Wolfram	O
Mathematica	O
(	O
8.0	O
or	O
later	O
)	O
also	O
provides	O
a	O
handwriting	O
or	O
text	B-Application
recognition	I-Application
function	O
TextRecognize	O
.	O
</s>
<s>
Handwriting	B-Application
recognition	I-Application
has	O
an	O
active	O
community	O
of	O
academics	O
studying	O
it	O
.	O
</s>
<s>
The	O
biggest	O
conferences	O
for	O
handwriting	B-Application
recognition	I-Application
are	O
the	O
International	O
Conference	O
on	O
Frontiers	O
in	O
Handwriting	B-Application
Recognition	I-Application
(	O
ICFHR	O
)	O
,	O
held	O
in	O
even-numbered	O
years	O
,	O
and	O
the	O
International	O
Conference	O
on	O
Document	O
Analysis	O
and	O
Recognition	O
(	O
ICDAR	O
)	O
,	O
held	O
in	O
odd-numbered	O
years	O
.	O
</s>
<s>
Since	O
2009	O
,	O
the	O
recurrent	B-Algorithm
neural	I-Algorithm
networks	I-Algorithm
and	O
deep	O
feedforward	B-Algorithm
neural	I-Algorithm
networks	I-Algorithm
developed	O
in	O
the	O
research	O
group	O
of	O
Jürgen	O
Schmidhuber	O
at	O
the	O
Swiss	O
AI	O
Lab	O
IDSIA	O
have	O
won	O
several	O
international	O
handwriting	O
competitions	O
.	O
</s>
<s>
In	O
particular	O
,	O
the	O
bi-directional	O
and	O
multi-dimensional	O
Long	B-Algorithm
short-term	I-Algorithm
memory	I-Algorithm
(	O
LSTM	B-Algorithm
)	O
of	O
Alex	O
Graves	O
et	O
al	O
.	O
</s>
<s>
won	O
three	O
competitions	O
in	O
connected	O
handwriting	B-Application
recognition	I-Application
at	O
the	O
2009	O
International	O
Conference	O
on	O
Document	O
Analysis	O
and	O
Recognition	O
(	O
ICDAR	O
)	O
,	O
without	O
any	O
prior	O
knowledge	O
about	O
the	O
three	O
different	O
languages	O
(	O
French	O
,	O
Arabic	O
,	O
Persian	O
)	O
to	O
be	O
learned	O
.	O
</s>
<s>
Recent	O
GPU-based	O
deep	B-Algorithm
learning	I-Algorithm
methods	O
for	O
feedforward	B-Algorithm
networks	O
by	O
Dan	O
Ciresan	O
and	O
colleagues	O
at	O
IDSIA	O
won	O
the	O
ICDAR	O
2011	O
offline	O
Chinese	O
handwriting	B-Application
recognition	I-Application
contest	O
;	O
their	O
neural	O
networks	O
also	O
were	O
the	O
first	O
artificial	O
pattern	O
recognizers	O
to	O
achieve	O
human-competitive	O
performance	O
on	O
the	O
famous	O
MNIST	B-General_Concept
handwritten	O
digits	O
problem	O
of	O
Yann	O
LeCun	O
and	O
colleagues	O
at	O
NYU	O
.	O
</s>
