Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Optimal Point-to-Point Distance Metric for Elastic Matching In Online Character Recognition

IP.com Disclosure Number: IPCOM000120381D
Original Publication Date: 1991-Apr-01
Included in the Prior Art Database: 2005-Apr-02
Document File: 1 page(s) / 44K

IBM

Related People

Chefalas, TE: AUTHOR [+3]

Abstract

Linear and elastic template matching are common techniques for online character recognition (1). In matching an unknown against a prototype, a distance metric is used to compare an arbitrary point i in the unknown with a point j in the kth prototype. Typical point-to-point distance metrics are Euclidean distance and direction-angle difference. In earlier recognition systems we used the point distance (Image Omitted) where x and y are the normalized (by the center of gravity of the character) coordinates, the height from the baseline, and the direction angle.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 72% of the total text.

Optimal Point-to-Point Distance Metric for Elastic Matching In Online
Character Recognition

Linear and elastic template matching are common
techniques for online character recognition (1).  In matching an
unknown against a prototype, a distance metric is used to compare an
arbitrary point i in the unknown with a point j in the kth prototype.
Typical point-to-point distance metrics are Euclidean distance and
direction-angle difference.  In earlier recognition systems we used
the point distance

(Image Omitted)

where x and y are the normalized (by the center of gravity of the
character) coordinates, the height from the baseline, and the
direction angle.

Disclosed is the following improved point distance

(Image Omitted)

where the first two terms are the Euclidean distance and the third a
constant (empirically determined) times the direction-angle
difference.  This metric differs from the earlier one in several
ways.  First, it was found the Euclidean metric was superior to the
absolute-value metric. Second, the height-from-baseline measure is no
longer used because it usually acts like noise and decreases overall
recognition accuracy, even though it improves accuracy for certain
discriminations, like P-p and Y-y that differ in position relative to
the baseline.  Thus, it is not useful for shape discrimination but
rather for non- shape-related d...