Browse Prior Art Database

Improved Quantization of Direction Angle and Angle Change Parameters in Online Handwriting Recognition

IP.com Disclosure Number: IPCOM000121244D
Original Publication Date: 1991-Aug-01
Included in the Prior Art Database: 2005-Apr-03
Document File: 1 page(s) / 40K

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). The accuracy of these recognizers depends on the distance metric used in matching. The memory required to store prototypes is proportional to the length of the writing of a character (in number of points) and the storage required for the parameters of a point.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 80% of the total text.

Improved Quantization of Direction Angle and Angle Change Parameters
in Online Handwriting Recognition

      Many online handwriting recognition systems use elastic
curve matching to match an unknown character against prototype
(template) characters (1).  The accuracy of these recognizers depends
on the distance metric used in matching. The memory required to store
prototypes is proportional to the length of the writing of a
character (in number of points) and the storage required for the
parameters of a point.

      Disclosed here is an improved quantization of the
direction-angle and angle-change parameters used in the distance
metric.  The metric distance, described in detail in [2], between
point i in the unknown and point j in prototype k is where the first
two terms are the Euclidean distance, the third term is a constant
(empirically determined) times the direction-angle difference, and
the fourth term is a constant (empirically determined) times the
difference of the change in direction angle.

      In order to reduce the storage requirements for prototypes the
direction-angle and angle-change parameters were quantized to fit
into 8-bit integers in the range 0-255.  Storage for the
direction-angle and angle-change parameters was halved, from 16-bit
numbers to 8-bit numbers, without loss in recognition accuracy.  In
order to maintain accuracy, it was necessary to retain the 16-bit
representation for the x and y offsets from the center of gravity.

 ...