Method and computer program product for generating training data for a new class in a pattern recognition classifier

• US 7,113,636 B2
• Filed: 08/30/2002
• Issued: 09/26/2006
• Est. Priority Date: 08/30/2002
• Status: Expired due to Fees

• Current Assignee: Lockheed Martin Corporation (Martin Marietta Corporation)
• Original Assignee: Lockheed Martin Corporation (Martin Marietta Corporation)
• Inventors: Ii, David L., Reitz, Elliott D. II, Tillotson, Dennis A.
• Primary Examiner(s): Mancuso, Joseph
• Assistant Examiner(s): Conover, Damon M
• Application Number:
  US10/232,133
• Publication Number:
  US 20040042652A1
• Time in Patent Office: 1,488 Days
• Field of Search: 382/157, 382/159, 382/225
• US Class Current: 382/159
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting

Abstract

A method and computer program product are disclosed for generating training data over a plurality of feature variables for a new output class in a pattern recognition classifier representing a plurality of existing classes with previously calculated statistical parameters. Deviation measures are generated for each feature variable for each of the plurality of existing classes from the calculated statistical parameters. The deviation measures for each feature variable are averaged across the plurality of existing classes. Feature data is extracted from an ideal pattern, representing the new class, for each of the feature variables. Statistical parameters are approximated for each feature variable for the new class from the extracted feature data and the averaged deviation measures.

Claims

1. A method of generating training data over a plurality of feature variables for a new output class in a pattern recognition classifier representing a plurality of existing classes with previously calculated statistical parameters, comprising:

generating deviation measures for each feature variable for each of the plurality of existing classes from the calculated statistical parameters;

averaging the deviation measures for each feature variable across the plurality of existing classes;

extracting feature data from an ideal pattern, representing the new class, for each of the feature variables; and

approximating statistical parameters for each feature variable for the new class from the extracted feature data and the averaged deviation measures.

2. A method as set forth in claim 1, wherein the method includes the additional step of refining the approximated statistical parameters in response to new data.

3. A method as set forth in claim 1, wherein the ideal pattern is a scanned image.

4. A method as set forth in claim 3, wherein at least one of the plurality of existing classes represent an alphanumeric character.

5. A method as set forth in claim 3, wherein at least one of the plurality of existing classes represent a form of postal indicia.

6. A method as set forth in claim 1, wherein the generated deviation measures include the difference between the mean and the minimum value for each feature variable.

7. A method as set forth in claim 1, wherein the generated deviation measures include the difference between the mean and the maximum value for each feature variable.

8. A method as set forth in claim 1, wherein the approximated statistical parameters include the mean of the new class.

9. A method as set forth in claim 1, wherein the approximated statistical parameters include the minimum value of the new class.

10. A method as set forth in claim 1, wherein the approximated statistical parameters include the maximum value of the new class.

11. A computer program product, implemented in a computer readable medium and operative in a data processing system, for generating training data over a plurality of feature variables for a new output class in a pattern recognition classifier representing a plurality of existing classes with previously calculated statistical parameters, comprising:

an averaging portion that generates deviation measures for each feature variable for each of the plurality of existing classes from the calculated statistical parameters and averages the deviation measures for each feature variable across the plurality of existing classes;

a feature extraction portion that extracts feature data from an ideal pattern, representing the new class, for each of the feature variables; and

a data generation portion that approximates statistical parameters for each feature variable for the new class from the extracted feature data and the averaged deviation measures.

12. A computer program product as set forth in claim 11, where the computer program product includes a refinement portion that refines the approximated statistical parameters in response to new data.

13. A computer program product as set forth in claim 11, wherein the ideal pattern is a scanned image.

14. A computer program product as set forth in claim 13, wherein at least one of the plurality of existing classes represent an alphanumeric character.

15. A computer program product as set forth in claim 13, wherein at least one of the plurality of existing classes represent a form of postal indicia.

16. A computer program product as set forth in claim 11, wherein the generated deviation measures include the interquartile range for each of the feature variables.

17. A computer program product as set forth in claim 11, wherein the approximated statistical parameters include the median of the new class.

18. A computer program product as set forth in claim 11, wherein the approximated statistical parameters include the interquartile range of the new class.

19. A computer program product as set forth in claim 11, wherein the approximated statistical parameters include the standard deviation of the new class.

20. A computer program product, implemented in a computer readable medium and operative in a data processing system, for generating training data over a plurality of feature variables for a new output class in a pattern recognition classifier representing a plurality of existing classes with previously calculated statistical parameters, comprising:

an averaging portion that generates deviation measures, including a standard deviation, for each feature variable for each of the plurality of existing classes from the calculated statistical parameters and averages the deviation measures for each feature variable across the plurality of existing classes;

a feature extraction portion that extracts feature data from an ideal pattern, representing the new class, for each of the feature variables; and

a data generation portion that approximates statistical parameters for each feature variable for the new class from the extracted feature data and the averaged deviation measures.