<s>
Optical	B-Application
music	I-Application
recognition	I-Application
(	O
OMR	O
)	O
is	O
a	O
field	O
of	O
research	O
that	O
investigates	O
how	O
to	O
computationally	O
read	O
musical	O
notation	O
in	O
documents	O
.	O
</s>
<s>
In	O
the	O
past	O
it	O
has	O
,	O
misleadingly	O
,	O
also	O
been	O
called	O
"	O
music	O
optical	B-Application
character	I-Application
recognition	I-Application
"	O
.	O
</s>
<s>
Optical	B-Application
music	I-Application
recognition	I-Application
of	O
printed	O
sheet	O
music	O
started	O
in	O
the	O
late	O
1960s	O
at	O
the	O
Massachusetts	O
Institute	O
of	O
Technology	O
when	O
the	O
first	O
image	B-Application
scanners	I-Application
became	O
affordable	O
for	O
research	O
institutes	O
.	O
</s>
<s>
The	O
first	O
commercial	O
OMR	O
application	O
,	O
MIDISCAN	B-Application
(	O
now	O
SmartScore	B-Application
)	O
,	O
was	O
released	O
in	O
1991	O
by	O
Musitek	B-Application
Corporation	I-Application
.	O
</s>
<s>
The	O
availability	O
of	O
smartphones	B-Application
with	O
good	O
cameras	O
and	O
sufficient	O
computational	O
power	O
,	O
paved	O
the	O
way	O
to	O
mobile	O
solutions	O
where	O
the	O
user	O
takes	O
a	O
picture	O
with	O
the	O
smartphone	B-Application
and	O
the	O
device	O
directly	O
processes	O
the	O
image	O
.	O
</s>
<s>
Optical	B-Application
music	I-Application
recognition	I-Application
relates	O
to	O
other	O
fields	O
of	O
research	O
,	O
including	O
computer	B-Application
vision	I-Application
,	O
document	O
analysis	O
,	O
and	O
music	B-Application
information	I-Application
retrieval	I-Application
.	O
</s>
<s>
Optical	B-Application
music	I-Application
recognition	I-Application
has	O
frequently	O
been	O
compared	O
to	O
Optical	B-Application
character	I-Application
recognition	I-Application
.	O
</s>
<s>
Notice	O
that	O
there	O
is	O
no	O
proper	O
equivalent	O
in	O
text	B-Application
recognition	I-Application
.	O
</s>
<s>
By	O
analogy	O
,	O
recovering	O
the	O
music	O
from	O
an	O
image	O
of	O
a	O
music	O
sheet	O
can	O
be	O
as	O
challenging	O
as	O
recovering	O
the	O
HTML	B-Language
source	O
code	O
from	O
the	O
screenshot	B-Application
of	O
a	O
website	O
.	O
</s>
<s>
Optical	B-Application
music	I-Application
recognition	I-Application
is	O
frequently	O
underestimated	O
due	O
to	O
the	O
seemingly	O
easy	O
nature	O
of	O
the	O
problem	O
:	O
If	O
provided	O
with	O
a	O
perfect	O
scan	O
of	O
typeset	O
music	O
,	O
the	O
visual	O
recognition	O
can	O
be	O
solved	O
with	O
a	O
sequence	O
of	O
fairly	O
simple	O
algorithms	O
,	O
such	O
as	O
projections	O
and	O
template	O
matching	O
.	O
</s>
<s>
surveyed	O
techniques	O
for	O
optical	B-Application
music	I-Application
recognition	I-Application
.	O
</s>
<s>
With	O
the	O
advent	O
of	O
deep	B-Algorithm
learning	I-Algorithm
,	O
many	O
computer	B-Application
vision	I-Application
problems	O
have	O
shifted	O
from	O
imperative	O
programming	O
with	O
hand-crafted	O
heuristics	O
and	O
feature	O
engineering	O
towards	O
machine	O
learning	O
.	O
</s>
<s>
In	O
optical	B-Application
music	I-Application
recognition	I-Application
,	O
the	O
staff	O
processing	O
stage	O
,	O
the	O
music	O
object	O
detection	O
stage	O
,	O
as	O
well	O
as	O
the	O
music	O
notation	O
reconstruction	O
stage	O
have	O
seen	O
successful	O
attempts	O
to	O
solve	O
them	O
with	O
deep	B-Algorithm
learning	I-Algorithm
.	O
</s>
<s>
Using	O
synthetic	O
data	O
helped	O
with	O
avoiding	O
copyright	O
issues	O
and	O
training	O
the	O
artificial	O
intelligence	O
algorithms	O
on	O
musical	O
cases	O
that	O
rarely	O
occur	O
in	O
actual	O
repertoire	O
,	O
ultimately	O
resulting	O
in	O
(	O
according	O
to	O
claims	O
by	O
the	O
company	O
)	O
more	O
accurate	O
music	B-Application
recognition	I-Application
.	O
</s>
<s>
SmartScore	B-Application
by	O
Musitek	O
.	O
</s>
<s>
Formerly	O
packaged	O
as	O
"	O
MIDISCAN	B-Application
"	O
.	O
</s>
<s>
(	O
SmartScore	B-Application
Lite	O
has	O
been	O
used	O
in	O
previous	O
versions	O
of	O
Finale	O
)	O
.	O
</s>
<s>
Released	O
in	O
May	O
2021	O
,	O
Maestria	O
is	O
an	O
example	O
of	O
new-generation	O
OMR	O
technology	O
based	O
on	O
deep-learning	B-Algorithm
.	O
</s>
