Doing some research into recognition accuracy of the MS engine, we've
run into what seems to be a bit of a limitation. One of our writers
wrote a long narrative of about 1800 strokes. If we feed all of these
strokes to the recognizer and then reco, it fails with a ProcessFailed
RecoStatus. By trial and error, we've determined it breaks down at
around 1570 strokes. This seems fairly arbitrary, so I'd guess maybe
it's a # of points or processing time limit or something.

Is there anywhere this sort of limitation is documented? I think we can
work around it if we know what we're looking for.

If you are using RecognizerContext, then the best way to get around the
problem (which I personally have not experienced, I will have to try it when
I get my Tablet PC back from the shop) is to use the Divider class.

Basically the scenario is that you would pre-parse the ink using Divider
which would give you back the ink parsed into logical groupings of
paragraphs, lines, and words. Then you can break up your recognition task
into several smaller tasks. It's worth mentioning though that there is an
API available now that is vastly superior to both Divider and
RecognizerContext called Ink Analysis. If you're developing a new
application, I recommend using that instead.

--
Josh Einstein (Tablet PC MVP)
Einstein Technologies
Tablet Enhancements for Outlook - Try it free: www.tabletoutlook.com



"Chris DiPierro" <cdipierr@mi-corporation.com> wrote in message
news:O0Wwl0t0HHA.5980@TK2MSFTNGP04.phx.gbl...
> Doing some research into recognition accuracy of the MS engine, we've run
> into what seems to be a bit of a limitation. One of our writers wrote a
> long narrative of about 1800 strokes. If we feed all of these strokes to
> the recognizer and then reco, it fails with a ProcessFailed RecoStatus. By
> trial and error, we've determined it breaks down at around 1570 strokes.
> This seems fairly arbitrary, so I'd guess maybe it's a # of points or
> processing time limit or something.
>
> Is there anywhere this sort of limitation is documented? I think we can
> work around it if we know what we're looking for.
>

Hello Chris,

The solution suggested by Josh is very good, you may consider it for
addressing your problem. In general, it is a good idea to provide the
recognizers maximum a few words (and only one line) to increase the accuracy
of the handwriting recognition (where practical to do so).

The Japanese, Chinese Simplified, Chinese Traditional and Korean handwriting
recognizers are limited to 1024 strokes and 32767 points per stroke. The
other handwriting recognizers (English, French, etc.) process currently up to
16000 strokes and 32767 points per stroke. This numbers are subject to change
in subsequent versions of the OS.

Thank you,
Mitica Manu