Structured document processing with lexical classes as context

• US 5,642,435 A
• Filed: 01/25/1995
• Issued: 06/24/1997
• Est. Priority Date: 01/25/1995
• Status: Expired due to Term

• Current Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Loris, Keith
• Primary Examiner(s): Boudreau, Leo
• Assistant Examiner(s): MEHTA, BHAVESH M
• Application Number:
  US08/378,151
• Time in Patent Office: 881 Days
• Field of Search: 382/229, 382/224, 382/225, 382/309, 382/310, 382/226, 382/227, 382/228
• US Class Current: 382/229
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/1444 Selective acquisition, locating or processing of specific regions, e.g. highlighted text, fiducial marks or predetermined fields
• CPC Class Codes: G06V 30/274 Syntactic or semantic context, e.g. balancing

Abstract

Patterns of intelligence detected in a digital image representation are correlated with pattern classes according to a predetermined set of rules for improved pattern identification. Identified patterns are subsequently mapped to classes for continued use of classification.

Claims

1. A character recognition method including the steps of:

receiving a digital image representing a hard copy structured document;

isolating characters in text portions of the digital image;

for a given string of isolated characters, specifying for each character possible alternative identifications of each character;

for the given string, forming a matrix of possible character identifications based on the specified alternative character identifications;

comparing each string in the matrix to a set of rules established for string classification to determine a lexical class for each string, wherein the lexical class establishes the rules to identify character patterns within each string based upon predefined meanings;

from the comparison step, generating a modified string, including characters confirmed by context processing specific to a determined lexical class;

from the comparison step, generating at least one lexical class identifier for association with the modified string.

2. The method as described in claim 1, where said digital image is received from a light intensity scanning device, which is operative to scan a hard copy document, to derive density-representing digital signals from sensed reflected light.

3. The method as described in claim 1, wherein the step of isolating characters in text portions of the digital image additionally identifies a relative location of the string in said hard copy structured document whereby occurrence of a string with a lexical class identifier is uniquely identified for further processing.

4. The method as described in claim 1, wherein the lexical class comprises a class selected from a set of pattern classes including:

Arabic numbers;

arithmetic expressions;

currency amounts;

Roman numbers;

phone numbers;

Social Security numbers;

strings that include punctuation; and

compound strings.