Neural network for character recognition and verification

• US 5,742,702 A
• Filed: 08/09/1996
• Issued: 04/21/1998
• Est. Priority Date: 10/01/1992
• Status: Expired due to Fees

• Current Assignee: Sony Corporation (Sony Group Corp.), Sony Electronics Inc. (Sony Group Corp.)
• Original Assignee: Sony Corporation (Sony Group Corp.), Sony Electronics Inc. (Sony Group Corp.)
• Inventors: Oki, Toru
• Primary Examiner(s): Razavi, Michael T.
• Assistant Examiner(s): CHANG, JON CARLTON
• Application Number:
  US08/694,810
• Time in Patent Office: 620 Days
• Field of Search: 382/217, 382/309, 382/156, 382/158, 382/161, 382/310, 395/21, 395/27
• US Class Current: 382/156
• CPC Class Codes: G06V 10/98 Detection or correction of errors, e.g. by rescanning the pattern or by human intervention; Evaluation of the quality of the acquired patterns
• CPC Class Codes: G06V 30/2504 Coarse or fine approaches, e.g. resolution of ambiguities or multiscale approaches

Abstract

A neural network is used to recognize characters from a character set. Based upon the character recognized, a smaller neural network is used for verification of the character recognized. The smaller neural network is trained to recognize only a single character of the set and provides a "yes" or "no" type verification of correct identification of the character.

Claims

1. A method for verifying a plurality of groups of symbols from a symbol set of N possible symbols, comprising the steps of:

(a) providing a plurality of N verification neural networks, each associated with and trained to verify one of N possible symbols wherein each of said N verification neural networks accepts as an input a representation of a symbol associated with each respective N verification neural network and has two output nodes indicating whether the input representation of a symbol is either verified or not verified;

(b) selecting a group of verification neural networks from said plurality of N verification neural networks associated with a group of symbols;

(c) applying representations of one group of symbols as inputs to said group of verification neural networks;

(d) processing said inputs in said verification neural networks; and

(e) repeating steps (c) and (d) for each of the plurality of groups of symbols other than said one group, wherein step (e) terminates when any one of said verification neural networks indicates a verification fail.

2. A method for processing a plurality of groups of symbols from a symbol set of N possible symbols, comprising the steps of:

(a) receiving a representation of a group of symbols as inputs to a recognition neural network;

(b) processing said inputs in said recognition neural network one at a time in order to produce outputs representing a group of recognized symbols;

(c) providing a plurality of N verification neural networks, each associated with and trained to verify one of said N possible symbols;

(d) selecting a group from said N verification neural networks associated with each member of said group of recognized symbols, said group from said N verification neural networks having an output indicative of either a verification pass or a verification fail;

(e) applying said representation of said group of symbols as inputs to each respective said verification neural network;

(f) processing said inputs in said verification neural networks;

(g) applying a representation of a group of symbols, which is different from the group applied in step (e), as inputs to said verification neural networks;

(h) processing said inputs in said verification neural networks; and

(i) repeating steps (g) and (h) for each remaining group of symbols, terminating when any one of said verification neural networks indicates a verification fail.

3. A method for processing a plurality of groups of symbols from a symbol set of N possible symbols, comprising the steps of:

(a) receiving an output of a video camera;

(b) digitizing said output of the video camera to produce a plurality of arrays P(K) of digital pixel values each representing a member of one group of said plurality of groups of symbols;

(c) applying said representation of said group of symbols as inputs to a recognition neural network;

(d) processing said inputs in said recognition neural network one at a time in order to produce outputs representing a group of recognized symbols;

(e) selecting a group of verification neural networks from said N verification neural networks associated with each member of said recognized group of symbols, said verification neural networks having an output indicative of either a verification pass or fail;

(f) applying representations of said group of symbols as inputs to said group of verification neural networks;

(g) processing said inputs in said selected group of verification neural networks; and

(h) repeating steps (a), (b), (f) and (g) for each of the plurality of groups of symbols other than said one group, terminating when any one of said verification neural networks indicates a verification fail.

4. A method for processing a serial number on a surface of a compact disc, comprising the steps of:

(a) receiving a representation of a group of symbols as inputs to a recognition neural network;

(b) processing said inputs in said recognition neural network one at a time in order to produce outputs representing a group of recognized symbols;

(c) providing a plurality of N verification neural networks, each associated with and trained to verify one of said N possible symbols;

(d) selecting a group from said N verification neural networks associated with each member of said group of recognized symbols, said group from said N verification neural networks having an output indicative of either a verification pass or a verification fail;

(e) applying said representation of said group of symbols as inputs to each respective said verification neural network;

(f) processing said inputs in said verification neural networks;

(g) applying a representation of a group of symbols, which is different from the group applied in step (e), as inputs to said verification neural networks;

(h) processing said inputs in said verification neural networks; and

(i) repeating steps (g) and (h) for each remaining group of symbols, terminating when any one of said verification neural networks indicates a verification fail.