Method and computer program product for identifying output classes with multi-modal dispersion in feature space and incorporating multi-modal structure into a pattern recognition system

• US 20040096100A1
• Filed: 11/14/2002
• Published: 05/20/2004
• Est. Priority Date: 11/14/2002
• Status: Active Grant

• 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.
• Granted Patent: US 7,233,692 B2
• Time in Patent Office: Days
• US Class Current: 382/159
• CPC Class Codes: G06F 18/2431 Multiple classes

Abstract

A method and computer program product are disclosed for identifying output classes with multi-modal dispersion in feature space and incorporating multi-modal structure into a pattern recognition system architecture. 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 grouped into clusters according to the similarities between the pattern samples. Clusters that contain samples associated with a represented output class are identified via independent review. The classifier is then retrained to recognize the identified clusters as output pseudoclasses separate from the represented output class with which they are associated. The system architecture is reorganized to incorporate the output pseudoclasses. The output pseudoclasses are rejoined to their associated class after classification.

Claims

1. A method, operative within a pattern recognition system with a plurality of represented output classes, for identifying output classes with multi-modal dispersion in feature space and incorporating multi-modal structure into a system architecture, said method comprising:

rejecting a plurality of input pattern samples determined not to be associated with any of a plurality of represented output classes by a pattern recognition classifier;

grouping the rejected pattern samples into clusters according to the similarities between the pattern samples;

identifying clusters that contain pattern samples associated with a represented output class via independent review;

retraining the classifier to recognize the identified clusters as output pseudoclasses separate from the represented output class with which they are associated;

reorganizing the system architecture to incorporate the output pseudoclasses; and

rejoining the output pseudoclasses to their associated classes after classification.

2. A method as set forth in claim 1, wherein the step of reorganizing the system architecture includes applying a Kohonen clustering algorithm.

3. A method as set forth in claim 1, wherein the step of reorganizing the system architecture includes minimizing a cost function associated with a pattern recognition classifier as part of a clustering algorithm.

4. A method as set forth in claim 1, wherein the step of reorganizing the system architecture includes minimizing a function of the within group variance of the plurality of clusters as part of a clustering algorithm.

5. A method as set forth in claim 1, wherein the step of identifying clusters associated with a represented output class via independent review includes subjecting the clustered pattern samples to a computerized verification process.

6. A method as set forth in claim 1, wherein the step of identifying clusters associated with a represented output class via independent review includes outputting the clusters of rejected pattern samples to a human operator.

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

8. A method as set forth in claim 7, wherein at least one of the plurality of represented output classes represents a variety of postal indicia.

9. A method as set forth in claim 7, wherein at least one of the plurality of represented output classes represents an alphanumeric character.

10. A computer program product, operative within a data processing system, for use in a pattern recognition system with a plurality of represented output classes for identifying output classes with multi-modal dispersion in feature space and incorporating multi-modal structure into a system architecture, said method comprising:

a classification portion that rejects a plurality of input pattern samples determined not to be associated with any of a plurality of represented output classes by a pattern recognition classifier;

a clustering portion that groups the rejected pattern samples into clusters according to the similarities between the pattern samples;

an identification portion that identifies clusters that contain samples associated with a represented output class via independent review;

a training portion that retrains the classifier to recognize the identified clusters as output pseudoclasses separate from the represented output class with which they are associated;

a reorganization portion that reorganizes the system architecture to incorporate the output pseudoclasses; and

a post-processing portion that rejoins the output pseudoclasses to their associated class after classification.

11. A computer program product as set forth in claim 10, wherein the reorganization portion applies a Kohonen clustering algorithm in determining an optimal system architecture.

12. A computer program product as set forth in claim 10, wherein the reorganization portion determines an optimal system architecture by minimizing a cost function associated with a pattern recognition classifier as part of a clustering algorithm.

13. A computer program product as set forth in claim 10, wherein the reorganization portion determines an optimal clustering by minimizing a function of the within group variance of the plurality of clusters as part of a clustering algorithm.

14. A computer program product as set forth in claim 10, wherein the identification portion subjects the clusters of samples to a computerized verification process for independent review.

15. A computer program product as set forth in claim 10, wherein the identification portion outputs the clusters of samples to a human operator for independent review.

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

17. A computer program product as set forth in claim 16, wherein at least one of the represented output classes represents a variety of postal indicia.

18. A computer program product as set forth in claim 16, wherein at least one of the represented output classes represents an alphanumeric character.