Handwriting recognition has always been tough problem. It takes years to come to a
commercially valuable result in this area - years of research, years of development
and years of turning theory into software that would do at least a small piece of an
evident thing - turning human handwritten digital curves into letters and words.
We've been working on that since 1996.
Our handwriting recognition attitude
We believe that there is an ultimate solution of the general real-time handwriting
recognition problem, which can be stated as "To convert a set of digital
handwritten curves into a set of letters and/or sensible words in arbitrary
language". Last 5 years of research advanced us to a partial solution of this
common problem - we've got a technology, that allows to convert a set of digital
handwritten curves into a set of letters for any certain language. The technology
proved to be effective by delivering recognition for alphabets of different
graphical structure - Latin, Extended Latin, Cyrillic, Greek and Hebrew.
Research commitment
A number of neighbour tasks have been rose and solved to deliver current technology.
The list of neighbour task includes: problem of separating neighbour/overlapped
handwritten characters, composition and recognition of multistroke handwritten
characters, separation and segmentation of multi-lined handwritten text,
normalization of hand-written character slope, normalization and pre-processing of
digital handwritten trajectories, implementation of various mathematical functions
in integer form, fast integer calculation of various functions, discrete
representation of irrational curves, automatic detection of uppercase characters,
automatic detection of spaces and many others. In a move to provide the solution for
a connected (cursive) handwriting we're continuing our work. Recent research efforts
have been focused on following problems: fragmentation and composition of
hand-written strokes in connected (cursive) handwriting, integration of linguistic
and morphology databases with graphic recognition modules, unified
graphic-linguistic recognition engine for connected (cursive) handwritings.
Multilanguage coverage
We consider multilanguage support as an essential part of handwriting recognition
problem. So far we succeeded to keep our handwriting solutions applicable to almost
any human language. Connected (cursive) handwriting is likely to require linguistic
knowledge to be involved into a recognition process. However early versions of our
connected (cursive) handwriting recognition core have indicated specific classes of
connected (cursive) handwriting that can be recognized without linguistic analysis.
The issue of these classes "width" and the issue of their commercial
applicability still remain the subject of our research. However there is no clear
indication, that connected (cursive) handwriting obligatory requires linguistic
information. We keep researching...
Technology demonstration
One of the most recent versions of PenReader recognition kernel is available on-line
running along with a Java applet under Linux (SuSE) on this web site - check it out.
Technology stage and applications
Currently available technology is implemented in C++ cross platform core, compilable
for a following platforms: Wintel, Windows CE, Symbian and Palm OS. Commercial
products are available for Wintel and Windows CE platforms. Following version are
being developed to be shipped in year 2001: GPRS oriented Java powered client-server
PenReader, PenReader for Symbian OS 6.0 and PenReader for Palm OS.
