Browse Prior Art Database

Fast Two-Stage Matcher for Handwritten Character Recognition

IP.com Disclosure Number: IPCOM000036125D
Original Publication Date: 1989-Sep-01
Included in the Prior Art Database: 2005-Jan-28
Document File: 4 page(s) / 85K

Publishing Venue

IBM

Related People

Tappert, CC: AUTHOR

Abstract

A two-stage matching system is disclosed to provide a fast and accurate recognizer for handwritten characters. The first stage consists of a fast, linear matcher to reduce the character choices. The second stage consists of a slower, elastic (non-linear) matcher to provide the required accuracy.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 55% of the total text.

Page 1 of 4

Fast Two-Stage Matcher for Handwritten Character Recognition

A two-stage matching system is disclosed to provide a fast and accurate recognizer for handwritten characters. The first stage consists of a fast, linear matcher to reduce the character choices. The second stage consists of a slower, elastic (non-linear) matcher to provide the required accuracy.

Computation speed is critical for on-line handwriting recognition systems because such systems usually run on small stand-alone computers. The previous handwriting recognition system * used elastic matching to provide sufficient character recognition accuracy. However, although computation speed was about 9 characters/second on the IBM System 370/Model 308l, it is not sufficient for smaller machines.

The basic changes were to match on a stroke-by-stroke basis and to add a fast, first-stage linear match.

(Image Omitted)

Pruning. Since linear matching is done on a stroke-by-stroke basis, those prototypes having a different number of strokes compared to the unknown were pruned, i.e., eliminated from consideration. Prototypes were also pruned as required by elastic matching, i.e., those prototypes in which the length of any stroke was not within the elasticity range of half to double the length of the corresponding stroke of the unknown.

Linear matching. A linear match was performed to further reduce the number of prototypes to be elastically matched against the unknown. This is essentially an additional method of pruning.

This linear match was performed on a stroke-by-stroke basis. Within each stroke, the points of the unknown were linearly matched against points of the prototype (using the point distance metric *) as follows. The first point of the unknown stroke was matched to the first of the prototype, and the last to the last. Each intermediate point of the un...