Browse Prior Art Database

Segmentation Function Enhancement to Cursive Script Recognition System

IP.com Disclosure Number: IPCOM000043663D
Original Publication Date: 1984-Sep-01
Included in the Prior Art Database: 2005-Feb-05
Document File: 3 page(s) / 82K

Publishing Venue

IBM

Related People

Tappert, CC: AUTHOR

Abstract

The reference [*] describes a recognition system for cursive writing based on elastic matching of the unknown word against a set of prototypes established by each individual writer. The input to the system is point data produced by the dynamic trace of a stylus on an electronic tablet. Processing is performed on a word-by-word basis after the writing is separated into words. Using prototypes for each letter, the matching procedure allows any letter to follow any letter and finds the letter sequence which best fits the unknown word. A major advantage of this procedure is that it combines letter segmentation and recognition in one operation by, in essence, evaluation recognition at all possible segmentations, thus avoiding the usual segmentation-then-recognition philosophy.

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 52% of the total text.

Page 1 of 3

Segmentation Function Enhancement to Cursive Script Recognition System

The reference [*] describes a recognition system for cursive writing based on elastic matching of the unknown word against a set of prototypes established by each individual writer. The input to the system is point data produced by the dynamic trace of a stylus on an electronic tablet. Processing is performed on a word-by-word basis after the writing is separated into words. Using prototypes for each letter, the matching procedure allows any letter to follow any letter and finds the letter sequence which best fits the unknown word. A major advantage of this procedure is that it combines letter segmentation and recognition in one operation by, in essence, evaluation recognition at all possible segmentations, thus avoiding the usual segmentation-then-recognition philosophy. While this system gave excellent results, there are some inherent problems of segmentation at unreasonable points, particularly with certain letter sequences. For example, if segmentation is permitted at a corner, bo could easily be decoded as 1o because the decoder is allowed to follow an 1 prototype with one for o with a high initial ligature.

Also, other segmentations are made by this system at points which are clearly not reasonable segmentation points -- such as within a loop or at a cusp. Accordingly, the above procedure was modified by limiting segmentation points. Transition from one letter prototype to another (i.e., segmentation) is prohibited within loops, at cusps and corners, points not headed in the appropriate angular direction (essentially up and to the right), and points near ends of strokes except the last point of a stroke. The main intent of the presently described technique is to inhibit segmentation at those points which are definitely not segmentation points with the aim of both avoiding segmentation errors and speeding computation. Processing is performed on a word at a time. The segmentation function (see following paragraph) is determined and is then used to constrain transitions between letter prototypes in the elastic matching process. To determine the segmentation function, a binary function specifying which points are permitted as segmentation points, each stroke of a word is processed according to the flowchart of Fig. 1. First, the segmentation function is initialized to 1 (true) for all points.

Second, segmentation is prohibited, i.e., the segmentation function is set to 0 (false), at all points in loops. A loop is said to occur when the line segment defined by a pair of successive points intersects the line segment defined by another occurring pair of succes...