Method and computer program product for identifying and incorporating new output classes in a pattern recognition system during system operation

• US 7,164,791 B2
• Filed: 07/30/2002
• Issued: 01/16/2007
• Est. Priority Date: 07/30/2002
• Status: Expired due to Fees

• Current Assignee: Lockheed Martin Corporation (Martin Marietta Corporation)
• Original Assignee: Lockheed Martin Corporation (Martin Marietta Corporation)
• Inventors: Cecala, Joseph F., Paradis, Rosemary D., Reitz, Elliott D. II
• Primary Examiner(s): Mancuso; Joseph
• Assistant Examiner(s): SCHAFFER, JONATHAN C
• Application Number:
  US10/208,447
• Publication Number:
  US 20040022441A1
• Time in Patent Office: 1,631 Days
• Field of Search: 382/225, 382/159
• US Class Current: 382/159
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting

Abstract

The present invention recites a method and computer program product for identifying one or more new pattern classes and incorporating the classes into a pattern recognition classifier as output classes. A plurality of input patterns determined not to be associated with any of a set of at least one represented output class by a pattern recognition classifier are rejected. The rejected pattern samples are then grouped into clusters according to the similarities between the pattern samples. Clusters that represent new pattern classes are identified via independent review. The classifier is then retrained to include the new pattern classes as output classes.

Claims

1. A method for identifying one or more new pattern classes and incorporating the classes into a pattern recognition classifier as output classes, comprising:

providing a plurality of input patterns having no known class affiliation to the pattern recognition classifier;

rejecting a subset of the plurality of input patterns that are determined not to be associated with any of a set of at least one represented output class by a pattern recognition classifier;

grouping the rejected pattern samples into clusters according to their proximity in a feature space associated with the pattern recognition classifier;

identifying clusters that represent new output classes via independent review; and

retraining the classifier to include the new pattern classes as output classes.

2. A method as set forth in claim 1, wherein the step of identifying clusters representing new output classes includes subjecting the clustered pattern samples to a computerized verification process.

3. A method as set forth in claim 1, wherein the step of identifying clusters representing new output classes includes outputting them to a human operator.

4. A method as set forth in claim 1, wherein the plurality of input patterns include scanned images.

5. A method as set forth in claim 4, wherein at least one of the represented output classes represents a variety of postal indicia.

6. A method as set forth in claim 4, wherein at least one of the represented output classes represents an alphanumeric character.

7. A computer program product, operative in a data processing system and implemented on a computer readable medium, for identifying one or more new pattern classes and incorporating the pattern classes into a pattern recognition classifier as output classes, comprising:

a classification portion that receives a plurality of input patterns having no known class affiliation and rejects a subset of the plurality of input patterns that are determined not to be associated with any of a set of at least one represented output class;

a clustering portion that groups the rejected pattern samples into clusters according to their proximity in a feature space associated with the pattern recognition classifier;

a class identification portion that identifies clusters that represent new output classes via an independent review; and

a training portion that retrains the classifier to include the new pattern classes as output classes.

8. A computer program product as set forth in claim 7, wherein the class identification portion submits the grouped pattern samples to a computerized verification process.

9. A computer program product as set forth in claim 7, wherein the class identification portion outputs the grouped pattern samples to a human operator.

10. A computer program product as set forth in claim 7, wherein the plurality of input patterns include scanned images.

11. A computer program product as set forth in claim 10, wherein at least one of the represented output classes represents a variety of postal indicia.

12. A computer program product as set forth in claim 10, wherein at least one of the represented output classes represents an alphanumeric character.

13. A method for identifying new pattern classes and incorporating the classes into a pattern recognition classifier as output classes:

classifying a plurality of input patterns at the pattern recognition classifier such that each of the plurality of input patterns is assigned to one of an output class from a plurality of output classes associated with the pattern recognition classifier and set of rejected input patterns;

executing a clustering algorithm to assign the set of rejected input patterns to at least one cluster according to their proximity in a multidimensional feature space associated with the pattern recognition classifier;

identifying at least one cluster that represents a new output class from the at least one cluster produced by the clustering algorithm; and

retraining the pattern recognition classifier to include at least one new output class represented by the identified at least one cluster.

14. The method of claim 13, wherein executing a clustering algorithm comprises executing a k-means clustering algorithm.

15. The method of claim 13, wherein the at least one new output class comprises at least one output class that represents a common noise pattern, such that retraining the pattern recognition classifier to recognize the common noise pattern allows the pattern recognition classifier to recognize and discard input patterns comprised of the common noise pattern.

16. The method of claim 13, wherein identifying at least one cluster representing a new output classes includes subjecting the clustered pattern samples to a computerized verification process.

17. The method of claim 13, wherein the identifying clusters representing new pattern classes includes outputting them to a human operator.

18. The method of claim 13, further comprising:

determining if the set of rejected input patterns is sufficiently large to perform a clustering algorithm; and

executing the clustering algorithm if the set of rejected input patterns is sufficiently large.

19. The method of claim 13, wherein at least one of the output classes associated with the pattern recognition classifier represents a variety of postal indicia.

20. The method of claim 13, wherein classifying the plurality of input patterns at the pattern recognition classifier comprises classifying the plurality of input samples at a Bayesian Mahalanobis distance classification system.