Has anyone seen a good way to extract the strokes corresponding to
each letter of a word that is written in Ink, particularly for cursive
writing? I know that InkAnalyzer will only break it down to the word
level-of-detail.

Presumably at some stage in the recognition, the system knows which
strokes (or parts thereof) go with which letter. Are there hooks into
the InkAnalyzer processing that would allow this to be extracted? Is
there an intermediate API between RealTimeStylus and InkAnalyzer that
would be able to do this? Has anyone seen a third party recognizer
that has the ability to extract strokes from individual letters?

I could theoretically achieve this by recognizing the word, then
breaking up the associated strokes at several points, and recognizing
the sections. By varying the breaking points until I maximize the
recognition confidence, I could get an approximate letter by letter
breakdown. However, this does not seem robust at all.

Any thoughts?

Eli

It doesn't work that way. The handwriting recognizer does not work on a cell
by cell basis. That's what led to the ridiculing of the Newton. The
handwriting recognizer analyzes the word in its entirety and matches it to a
dictionary or an expected pattern. It doesn't try to subdivide the writing
into letters. In other words (no pun intended), the "word" *is* the leaf
node. Doing it this way is similar in concept to recognizing letter by
letter (after all, recognizing a single letter is just the same as
recognizing a word that consists of only one letter such as "I" or "a")
except that ambiguity is greatly reduced and thus accuracy is greatly
improved.

The only way to get letter by letter is to input each cell separately.

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


"Eli Gibson" <eli.gibson******.com> wrote in message
news:f2a8a574-b52b-4345-a9ba-432f01b5dc2e@i7g2000prf.googlegroups.com...
> Has anyone seen a good way to extract the strokes corresponding to
> each letter of a word that is written in Ink, particularly for cursive
> writing? I know that InkAnalyzer will only break it down to the word
> level-of-detail.
>
> Presumably at some stage in the recognition, the system knows which
> strokes (or parts thereof) go with which letter. Are there hooks into
> the InkAnalyzer processing that would allow this to be extracted? Is
> there an intermediate API between RealTimeStylus and InkAnalyzer that
> would be able to do this? Has anyone seen a third party recognizer
> that has the ability to extract strokes from individual letters?
>
> I could theoretically achieve this by recognizing the word, then
> breaking up the associated strokes at several points, and recognizing
> the sections. By varying the breaking points until I maximize the
> recognition confidence, I could get an approximate letter by letter
> breakdown. However, this does not seem robust at all.
>
> Any thoughts?
>
> Eli