Browse Prior Art Database

Character Recognition System

IP.com Disclosure Number: IPCOM000093226D
Original Publication Date: 1967-Jul-01
Included in the Prior Art Database: 2005-Mar-06
Document File: 2 page(s) / 33K

Publishing Venue

IBM

Related People

Nagy, G: AUTHOR [+2]

Abstract

This system corrects references used by a categorizer in a character recognition system to provide for deviations in font and quality of specimens being read. Each character on a series of documents is scanned and converted into a binary, i.e., feature, vector by scanner 10 and feature extractor 12 respectively. A minimum distance of hyperplane type categorizer 14 assigns a tentative identity to each feature vector on the basis of stored initial weights, i.e., references, from storage 16.

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 100% of the total text.

Page 1 of 2

Character Recognition System

This system corrects references used by a categorizer in a character recognition system to provide for deviations in font and quality of specimens being read. Each character on a series of documents is scanned and converted into a binary, i.e., feature, vector by scanner 10 and feature extractor 12 respectively. A minimum distance of hyperplane type categorizer 14 assigns a tentative identity to each feature vector on the basis of stored initial weights, i.e., references, from storage 16.

The feature vector are stored in push-down store 18. When store 18 is full, the earliest features are spilled through the overflow. The reject rate and the confidence level in the decisions are monitored constantly. When either reaches a preset level, the feature bits in store 18 are averaged under each character label to obtain the probabilities of occurance of the bits. This information suffices to generate new references from reference generator 20.

The feature vectors in store 18 are now recirculated through categorizer 14. The latter reassigns identities on the basis of the new references. The process is continued until sufficient confidence is placed in the decisions. Font change monitor 22 observes which particular subset of the initial references is active in the decision procedure. When a change exceeding a preset threshold occurs, store 18 is cleared.

1

Page 2 of 2

2

[This page contains 3 pictures or other non-text objects]