Browse Prior Art Database

Resolving Handwriting Recognition Confusion by Elastic Feature Extraction

IP.com Disclosure Number: IPCOM000121016D
Original Publication Date: 1991-Jul-01
Included in the Prior Art Database: 2005-Apr-02
Document File: 1 page(s) / 57K

Publishing Venue

IBM

Related People

Leibman, GJ: AUTHOR

Abstract

Currently, an effective handwriting recognition method is elastic matching, which compares and obtains scores between unknown and prototype character strokes [1,2]. In the case of an unknown which is close to two prototypes differing only in a small (but human-recognizable) feature, the difference in scores contributed by this feature is small in comparison with its importance, since all other points of the stroke are equally weighed. Character pairs, such as (o,O), (A,H), (U,V), etc., are often indistinguishable as a result, since writers tend to articulate unimportant parts of characters less carefully. Disclosed is a method which identifies these distinguishing features in a modified elastic matching method.

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

Resolving Handwriting Recognition Confusion by Elastic Feature Extraction

      Currently, an effective handwriting recognition method is
elastic matching, which compares and obtains scores between unknown
and prototype character strokes [1,2].  In the case of an unknown
which is close to two prototypes differing only in a small (but
human-recognizable) feature, the difference in scores contributed by
this feature is small in comparison with its importance, since all
other points of the stroke are equally weighed.  Character pairs,
such as (o,O), (A,H), (U,V), etc., are often indistinguishable as a
result, since writers tend to articulate unimportant parts of
characters less carefully.  Disclosed is a method which identifies
these distinguishing features in a modified elastic matching method.
It is to be used following a recognition attempt where an unknown is
close to two prototypes, improving recognition in cases as described
above.

      This invention first performs an elastic match between the two
contending prototypes, with the added step of saving with each point
the distance between it and its counterpart. A heuristic then is
applied to the resulting sets of points with their distance
information: if the distance associated with a point is above a
threshold, the point is marked (as being of interest).

      The second step of this invention performs a modified elastic
match between the unknown and each of the contending prototypes.  As
is normal el...