From
Handwriting
recognition is a mature field today with many applications such as processing
of mail pieces, bank checks, census forms, medical forms, and more. We will
present two of these applications: the first is the postal application where
the handwriting technology has found the maximum success and the second is
reading data on handwritten medical forms which is the most topical.
Postal
automation represents a fertile area for the application of image processing
and pattern recognition techniques. The US Postal service processes over 675
million pieces of letter mail a day and about 10-15% of these are handwritten.
This makes handwritten address interpretation an attractive economic
proposition requiring solutions to many challenging pattern recognition tasks.
We
will describe the postal automation process, the current read rates, the savings
realized by the US Postal Service, and the challenges remaining. We will focus
on 3 of the many different handwriting recognition technologies developed in
our labs which are now part of the postal automation technologies in
Active
character recognition methods wherein the features used, the length of the
feature vectors, and the amount of computational esources
used are all dynamically determined based on the context. A
system for rapid verification of unconstrained handwritten phrases using perceptual
holistic features of the handwritten phrase images. A sequential method
for combining word recognizers wherein the dividual recognizers
in the cascade are dynamically chosen based on a method of predicting the
performance of a recognizer given the lexicon and the quality of the image.
We will conclude with a brief overview of some topical applications in the emergency medical services domain and extension of the core handwriting recognition technology to other languages, such as Hindi and Arabic.