Recovering Dynamic Information from Static, Handwritten Word Images

Bridging the Gap between On-Line and Off-Line Handwriting Recognition


Handwriting recognition deals with the automatic recognition of
handwritten symbols by means of computers. The field of handwriting
recognition is divided into two distinctive areas: off-line recognition
and on-line recognition. Postal automation is a typical example of
off-line recognition, where handwritten addresses are processed by a
handwriting recognizer and mechanically routed to the appropriate
sorting bins. A typical example of on-line recognition are Personal
Digital Assistants (PDAs), small handheld computers accepting
pen-based input and combining agenda, address book and telecommunication
facilities.

On-line recognition methods exploit dynamic information such as the position
of the pen at a given time. To date, on-line and off-line recognition
are not compatible due to the absence of dynamic information in
static, handwritten words. This book presents an approach to bridge the
gap between on-line and off-line handwriting recognition.
The original pen movements that generated a handwritten word image are
recovered by means of a graph-theoretical, global optimization method.
This method is based on solving the Traveling Salesman Problem, a well-known
problem in computer science.