Browse Prior Art Database

Method for Creating Multi-templates in Online Handwritten Character Recognition

IP.com Disclosure Number: IPCOM000108444D
Original Publication Date: 1992-Jun-01
Included in the Prior Art Database: 2005-Mar-22
Document File: 3 page(s) / 124K

Publishing Venue

IBM

Related People

Kitamura, K: AUTHOR

Abstract

Disclosed is a method for automatically creating multi-templates from character samples for pattern matching in online handwritten character recognition. The created multi-templates support variations in both the number and order of strokes.

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

Method for Creating Multi-templates in Online Handwritten Character Recognition

       Disclosed is a method for automatically creating
multi-templates from character samples for pattern matching in online
handwritten character recognition. The created multi-templates
support variations in both the number and order of strokes.

      Pattern matching is an accurate and stable recognition method
that matches an input with templates in a dictionary. The templates
are generally created by (1) collecting character samples, (2)
clustering them according to the number and order of strokes, and (3)
averaging the character samples in each cluster to create a template.
Therefore, the templates can support variations in both the number
and order of strokes, which is essential for online Kanji character
recognition.

      The existing method supports only automatic clustering of
character samples that do not include variations in the number of
strokes.  If variations in stroke number are included in the
character samples, the clustering process requires human
intervention.  For example, the existing method does not support the
automatic clustering of character samples that have the same numbers
of strokes and whose strokes are connected at different positions, as
shown in the Kanji example in the figure.  This makes it difficult to
ensure the quality of the clustering, because the volume of the
character samples is enormous, especially in Kanji.

      The disclosed method allows automatic clustering of character
samples that may include variations in both the number and order of
strokes.  Consequently, templates can be automatically created from
the collected character samples without any human intervention.

      The disclosed method consists of the following three steps:
(1) Select an average-shaped sample by using the distributions of
stroke features.
(2) Collect the samples whose shapes are similar to the above sample
by matching stroke features.
(3) Average the collected samples to create a template.

      The following is an outline of the disclosed method's algorithm
for creating templates for a character from N samples, which may
include variations in both the number and order of strokes:
(1-a) Group N samples according to the number of strokes.
       (A template will be created for each group whose number of
samples is larger than a threshold value.)
(1-b) Take a group with M samples and G strokes.
      (This gives a total of MG strokes.)
      From these MG strokes, make G stroke groups of M strokes, each
of which is writ...