US 5,712,922 AGranted

Neural network optical character recognition system and method for classifying characters in a moving web

AssigneeEastman Kodak Company

InventorsKenneth H. Loewenthal, Steven M. Bryant

Life of this patent

Estimating the term…

Loading patent…

Abstract

A neural network based optical character recognition technique is presented for identifying characters in a moving web. Image acquisition means defines an imaging window through which the moving web passes such that the characters printed thereon can be imaged. Classification data is extracted and accumulated for each printed web character passing through the imaging window. A light source provides transmissive illumination of the web as it is being imaged. A neural network accelerator is coupled to the image acquisition means for intelligent processing of the accumulated classification data to produce therefrom printed character classification information indicative of each corresponding character imaged. A processor is coupled to the accelerator for converting the classification information into the appropriate ASCII character code. The technique is particularly useful for reading dot-matrix-type characters on a noisy, semi-transparent background at fast real-time rates. A neural network algorithm based recognition method is also described.

Claims36

Dependent claims

A neural network based optical character recognition system for extracting and classifying characters disposed on a moving web of material, said recognition system comprising:

image acquisition means defining an imaging window, said image acquisition means being positionable such that said characters pass through said defined imaging window, said image acquisition means extracting and accumulating image data on a web character passing through said imaging window;
a light source for providing illumination of said web character as said web character passes through said defined imaging window;
neural network accelerator means coupled to said image acquisition means for processing accumulated image data on said web character passing through said imaging window according to a predefined multilayer neural network, and for producing therefrom character classification information, said neural network accelerator means including:
dual-ported memory means receiving said accumulated image data from said image acquisition means, said dual-ported memory means having a first portion and a second portion, said first portion of said dual-ported memory means initially storing said accumulated image data, and n processing nodes connected in parallel to receive from said first portion and simultaneously process image data at different nodes for an input layer in said multilayer neural network, an output of each processing node for said input layer being stored in said second portion of said memory means, n being an integer.gtoreq.1, said first portion and second portion subsequently storing processing node outputs for alternate layers of said multilayer neural network, respectively;
and
classification means coupled to said neural network accelerator means for classifying said web character imaged by said image acquisition means using the corresponding character classification information produced by said neural network accelerator means.

3 more elements

A method for extracting and classifying characters disposed on a moving web of material using a multilayer neural network, said method comprising the steps of:

(a) passing said moving web through a predefined imaging window;
(b) illuminating said web within said imaging window simultaneous with said passing step (a);
(c) acquiring an image of said web within said imaging window simultaneous with said passing step (a);
(d) extracting and accumulating image data on characters disposed on said web imaged in said step (c);
and
(e) utilizing a multilayer neural network to process image data accumulated in said step (d), said multilayer neural network producing character classification information based upon accumulated image data, said utilizing step including:
for each layer of said multilayer neural network, processing data by n nodes thereof in parallel, n being an integer.ltoreq.1;
utilizing a memory storage having a first portion and a second portion for simultaneously providing input data and storing output data for each respective layer of said multilayer neural network;
initially storing in the first portion of said memory storage said image data accumulated in said step (d);
during processing of accumulated image data received from said first portion by n nodes of an input layer of said multilayer neural network, storing the output of said n nodes into said second portion of said memory storage;
and
for each network layer subsequent said input layer, alternately coupling one of said second portion and said first portion of the memory storage as an input to said n nodes, and storing the output of said n nodes into the other of said memory storage portions not containing the input data thereto.

9 more elements

Dependencies are inferred from the claim text.

Facts

Application: 08/340,270Filed 15 Nov 1994 · continuation in part
Priority: 14 Apr 1992
Granted: 27 Jan 1998
Term ends: Not estimatedThe source records don’t support an estimate.

Classification4 codes

G06V 30/1434
Image acquisition: special illumination such as grating, reflections or deflections, e.g. for characters with relief
G06V 30/194
Using simultaneous comparisons or correlations of the image signals with a plurality of references: references adjustable by an adaptive method,…
G06V 30/10
Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition: character recognition
G06V
Image and video recognition