Re: Re: scribble update

Keith Packard writes:
>
> > > How did you generate more strokes, Debby?
>
> > Hmmm.... Debby doesn't know either. That makes three of us. And
> > I'll bet Keith doesn't know keither, because he only ported it to
> > Xlib, and added the Xtest hack.
>
> 404 here too; I looked at it and prayed it worked. That code has
> been carefully obscured by many patient hands.
>

Grumble...

I've spent my metro time the last couple of days, (about 2 hours so
far), writing some text with xscribble on my iPAQ. Accuracy is
definitely a problem. The training set looks a little too clean to
me. It has about 4 "different" versions of each character but the 4
are almost identical. I would love to give this thing a more robust
training set.

In poking through the code, I found the references to the "Li/Yeung
recognition algorithm". A little searching turns up the following
paper:

        http://www.cs.ust.hk/faculty/dyyeung/paper/pdf/yeung.pr97.pdf

from among Yeung's handwriting recognition papers:

        http://www.cs.ust.hk/faculty/dyyeung/paper/hr.html

Does the January 1997 publication date jive with the history of the
scribble code? This paper seems to match the code as the paper's title
mentions "Dominant Points in Strokes" and the code includes
"lialg_compute_dominant_points". But perhaps the implementor didn't
follow the paper exactly? (The paper definitely has two-stroke
characters while the training set we have does not). Neither the
bibliography nor Yeung's web page mentions any other paper by Li and
Yeung.

I've only skimmed the paper so I don't yet know if it has the details
we would need to reverse engineer the .cl files or not, but I'm
hopeful.

-Carl

PS. Another approach would be to substitute a different recognition
algorithm of course. Any suggestions? There seem to be dozens of
papers on this sort of thing, I would think there would be more code
out there...
Received on Wed Sep 06 2000 - 07:51:15 EDT