Browse Prior Art Database

Semi-Automatic Method for Starter Set Creation in Online Handwriting Recognition

IP.com Disclosure Number: IPCOM000111007D
Original Publication Date: 1994-Feb-01
Included in the Prior Art Database: 2005-Mar-26
Document File: 2 page(s) / 78K

Publishing Venue

IBM

Related People

Chefalas, TE: AUTHOR [+3]

Abstract

Many online, handwriting recognition systems use elastic curve matching to match an unknown character against prototype (template) characters [1,2]. Such systems usually represent the ways of writing a character by a set of prototypes. A user prototype set is obtained by training on an individual user of the system. A general set of prototypes that covers the common ways of writing the characters, called a starter set, enables a new user to immediately use the recognition system. The walk-up recognition accuracy obtained by a new user of the system is a function of the coverage and quality (representativeness) of the starter set. The speed of recognition is inversely proportional to the size of the prototype set.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Semi-Automatic Method for Starter Set Creation in Online Handwriting
Recognition

      Many online, handwriting recognition systems use elastic curve
matching to match an unknown character against prototype (template)
characters [1,2].  Such systems usually represent the ways of writing
a character by a set of prototypes.  A user prototype set is obtained
by training on an individual user of the system.  A general set of
prototypes that covers the common ways of writing the characters,
called a starter set, enables a new user to immediately use the
recognition system.  The walk-up recognition accuracy obtained by a
new user of the system is a function of the coverage and quality
(representativeness) of the starter set.  The speed of recognition is
inversely proportional to the size of the prototype set.

      Disclosed herein is a semi-automatic method of creating a
starter set of prototypes that maximizes the trade-off between
coverage and size while maintaining prototype quality.  The main idea
of this method is to combine (average) user prototypes from a large
number of writers.  Although the concept is simple, the details of
the method are complex and far from obvious.

      The method consists of four steps.  First, a large number (over
25 was found sufficient) of user (individual) prototypes sets is
obtained by training on individual writers (with no starter prototype
set).  Prototypes are created by averaging similarly-shaped,
same-labeled characters obtained during training.  In order to
capture the greatest coverage, that is, the variety of ways of
writing each character, the users should be chosen from diverse
educational, regional, and socioeconomic backgrounds.

      Second, all singletons are removed from the user starter sets.
A "singleton" is an average of a single training character.  As such,
a singleton is a rare occurrence and usually represents a badly
written character, a writing of a character not corresponding to the
label, or a result of a system problem, such as segmentation.
Therefore, r...