Browse Prior Art Database

Hybrid Optical Character Recognition System with Neural Network and Template Matching Recognition Methods

IP.com Disclosure Number: IPCOM000114122D
Original Publication Date: 1994-Nov-01
Included in the Prior Art Database: 2005-Mar-27
Document File: 2 page(s) / 49K

Publishing Venue

IBM

Related People

Mano, T: AUTHOR [+2]

Abstract

An Optical Character Recognition (OCR) system with Neural Network (NN) and Template Matching (TM) methods is described to improve both recognition accuracy and cumulative accuracy. (The cumulative accuracy is the rate of appearance of a correct category in a sequence of recognition candidates for an input character image.) Generally speaking, the recognition accuracy of NN OCR systems is higher than that of TM OCR systems. On the other hand, the cumulative accuracy of TM OCR systems is better than that of NN OCR system. It should be noted that the cumulative accuracy is very important to get better results on post-processing of an OCR system such as spell checkers. The method described here improves both recognition accuracy and cumulative accuracy by a smart combination of NN and TM methods.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 59% of the total text.

Hybrid Optical Character Recognition System with Neural Network and
Template Matching Recognition Methods

      An Optical Character Recognition (OCR) system with Neural
Network (NN) and Template Matching (TM) methods is described to
improve both recognition accuracy and cumulative accuracy.  (The
cumulative accuracy is the rate of appearance of a correct category
in a sequence of recognition candidates for an input character
image.)  Generally speaking, the recognition accuracy of NN OCR
systems is higher than that of TM OCR systems.  On the other hand,
the cumulative accuracy of TM OCR systems is better than that of NN
OCR system.  It should be noted that the cumulative accuracy is very
important to get better results on post-processing of an OCR system
such as spell checkers.  The method described here improves both
recognition accuracy and cumulative accuracy by a smart combination
of NN and TM methods.

      At first, the template matching method recognizes an input
character image and outputs a sequence of recognition candidates.  In
case the template matching method has enough reliability in the
recognized first candidate, the OCR system outputs the candidate
sequence without any change.

      If the template matching method does not have enough
reliability, the neural network method recognized the input character
image and it outputs another sequence of recognition candidates.
Then the OCR system merges two sequences of recognition candidates
and...