Browse Prior Art Database

Arc Segmentation Method using Writing Direction

IP.com Disclosure Number: IPCOM000108742D
Original Publication Date: 1992-Jun-01
Included in the Prior Art Database: 2005-Mar-22
Document File: 8 page(s) / 279K

Publishing Venue

IBM

Related People

Kamon, Y: AUTHOR [+2]

Abstract

Disclosed is a segmentation process for on-line Japanese handwriting recognition. In conventional segmentation methods, as strokes are represented by lines, it is difficult to get an optimum approximation of the original strokes, especially in hiragana characters, which have long complex strokes. In this disclosure, a stroke is represented by a list of line and arc segments. As a result, a small number of segments can represent a complex stroke accurately.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 24% of the total text.

Arc Segmentation Method using Writing Direction

       Disclosed is a segmentation process for on-line Japanese
handwriting recognition.  In conventional segmentation methods, as
strokes are represented by lines, it is difficult to get an optimum
approximation of the original strokes, especially in hiragana
characters, which have long complex strokes.  In this disclosure, a
stroke is represented by a list of line and arc segments.  As a
result, a small number of segments can represent a complex stroke
accurately.

      An on-line handwriting recognition system gets its input from a
tablet.  For recognition purposes, the main input is a list of (x, y,
time) items, which indicate the time-series position on the tablet.
The segmentation process represents this as a list of basic segments.
This basic segment information is used in different stages of a
recognition system, such as stroke type filtering, stroke type
discrimination, and verification.

      In typical conventional methods, a line is the only type of
segment. On the other hand, arc segmentation gives additional
information on the type and curvature of arcs, and it also keeps the
number of segments small, since arcs are represented by one arc
segment instead of several lines. Using arcs as a basic segment type
is important especially for hiragana and alphanumeric characters,
which contain many cursive strokes.
1. Overview

      The segmentation process consists mainly of three phases:
preprocessing, local segmentation, and postprocessing (Figures
1a-1d).

      In preprocessing, line segments are first detected. A line
segment is an interpolated line that bridges two successive sampling
points (Figure 1a).  Noise elements are also eliminated in this step,
and the input strokes are divided into substrokes if any break points
are detected (Figure 1b).

      In local segmentation, input strokes are classified as lines or
arcs according to the relative directions of successive line segments
(Figure 1c).

      In postprocessing, short segments are merged or ignored
according to the neighboring segments, and all the segmentation steps
are completed (Figure 1d).
2. Preprocessing

      Preprocessing consists of reduction of noise, detection of
special strokes, and division of strokes into smaller parts.  Before
the application of preprocessing, every two successive sampling
points are treated as a set and make a line segment.  A line segment
is a basic element of the arc segmentation method.  In the following
explanation, LS stands for 'line segment.'
2.1 Noise Reduction

      Initial and final portions of a stroke that are shorter than a
certain threshold are removed.  If the length of an LS is very short
in comparison with the average length of those in the stroke, the LS
will be merged with the second shortest neighboring LS.
2.2 Detection of Special Strokes

      Dots and small circles are special stroke types. Special
strokes are...