Modular classification architecture for a pattern recognition application

• US 20040042651A1
• Filed: 08/30/2002
• Published: 03/04/2004
• Est. Priority Date: 08/30/2002
• Status: Active Grant

• Current Assignee: LOCKHEED MARTIN CORPORATI0N
• Original Assignee: Lockheed Martin Corporation (Martin Marietta Corporation)
• Inventors: Ii, David L., Reitz, Elliott D. II, Tillotson, Dennis A.
• Granted Patent: US 7,181,062 B2
• Time in Patent Office: Days
• US Class Current: 382/159
• CPC Class Codes: G06F 18/254 of classification results, e.g. of results related to same input data
• CPC Class Codes: G06V 30/242 Division of the character sequences into groups prior to recognition; Selection of dictionaries

Abstract

A method and computer program product are disclosed for use in classifying an input pattern into one of a plurality of output classes. A plurality of modular classifiers each represent a set of at least one associated output class. The modular classifiers are capable of being trained separately from the system. The classifiers select one of the associated output classes as a classification result and compute an associated confidence value. This confidence value possesses common significance across the plurality of modular classifiers.

A processing stage processes the confidence values from the plurality of modular classifiers. As a result of this processing, the processing stage selects an associated classification result.

Claims

1. A computer program product operative in a data processing system for use in classifying an input pattern into one of a plurality of output classes, the computer program product comprising:

a plurality of modular classifiers, each representing a set of at least one associated output class, capable of being trained separately from the system and operative to select one of the set of at least one associated output class as a classification result and compute an associated confidence value possessing common significance across the classifiers; and

a processing stage that processes the confidence values from the plurality of modular classifiers and selects an associated classification result.

2. A computer program product as set forth in claim 1, wherein each of the plurality of modular classifiers selects a classification result via a first classification technique, which may vary among the plurality of classifiers, and produce a confidence value via a second classification technique, which is common to the plurality of classifiers.

3. A computer program product as set forth in claim 2, wherein the first classification technique is a Bayesian classification technique for at least one of the plurality of modular classifiers.

4. A computer program product as set forth in claim 2, wherein at least one of the plurality of modular classifiers represents only one associated output class and the first classification technique includes selecting the one associated output class.

5. A computer program product as set forth in claim 2, wherein the second classification technique utilizes a radial basis function.

6. A computer program product as set forth in claim 1, wherein the input pattern is a scanned image and each of the plurality of output classes represents a variety of postal indicia.

7. A computer program product as set forth in claim 1, wherein the input pattern is a scanned image and each of the plurality of output classes represents an alphanumeric character.

8. A computer program product as set forth in claim 1, wherein the processing stage compares the confidence values from each of the plurality of modular classifiers and selects the classification result associated with the largest confidence value.

9. A computer program product as set forth in claim 1, wherein the processing stage compares each confidence value to a predetermined threshold and selects the classification result associated with the first confidence value to exceed the threshold.

10. A method for classifying an input pattern into one of a plurality of output classes via a modular classification architecture comprising:

selecting, at each of a plurality of modular classifiers having at least one associated output class, an output class most likely to be associated with the input pattern, via a classification technique associated with each classifier;

calculating, via a single classification technique common to all of the modular classifiers, at least one confidence value associated with a modular classifier and reflective of the likelihood that an output class selected at the modular classifier is associated with the input pattern;

processing the confidence values associated with at least one modular classifiers; and

selecting an associated classification result.

11. A method as set forth in claim 10, wherein the classification technique associated with at least one of the plurality of modular classifiers is a Bayesian classification technique.

12. A method as set forth in claim 10, wherein at least one of the plurality of modular classifiers represents only one associated output class and the classification technique associated with the at least one modular classifier with only one associated class includes selecting the one associated output class.

13. A method as set forth in claim 10, wherein the classification technique common to all of the modular classifiers utilizes a radial basis function.

14. A method as set forth in claim 10, wherein the input pattern is a scanned image and each of the plurality of output classes represents a variety of postal indicia.

15. A method as set forth in claim 10, wherein the input pattern is a scanned image and each of the plurality of output classes represents an alphanumeric character.

16. A method as set forth in claim 10, wherein the step of processing the confidence values associated with at least one modular classifier includes determining the largest of the calculated confidence values and selecting the output class selected at the classifier associated with the largest confidence value.

17. A method as set forth in claim 10, wherein the step of calculating at least one confidence value includes calculating confidence values for the output classes selected at the plurality of modular classifiers in a predetermined order.

18. A method as set forth in claim 17, wherein the step of processing the confidence values from each confidence value includes comparing each confidence value to a predetermined threshold and selecting the output class selected at the modular classifier associated with the first confidence value to exceed the threshold.

19. A method for adding an output class to an existing pattern recognition system, comprising:

training a modular classifier, capable of selecting a classification result via a first classification technique and calculating an associated confidence value via a second classification technique, on sample patterns from at least one associated output class; and

incorporating the trained modular classifier into an existing pattern recognition system with at least one additional modular classifier capable of calculating a confidence value by the second classification technique, such that the confidence value of the trained modular classifier can be compared to the confidence value of the at least one additional modular classifier to determine which classification result should be accepted by the system.

20. A method as set forth in claim 19, wherein the second classification technique utilizes a radial basis function.

21. A computer program product for adding an output class to an existing pattern recognition system, comprising:

a training portion that trains a modular classifier on sample patterns from at least one associated output class, the modular classifier being capable of selecting a classification result via a first classification technique and calculating an associated confidence value via a second classification technique; and

a control portion that incorporates the trained modular classifier into an existing pattern recognition system with at least one additional modular classifier capable of calculating a confidence value by the second classification technique, such that the confidence value of the trained modular classifier can be compared to the confidence value of the at least one additional modular classifier to determine which classification result should be accepted by the system.

22. A computer program product as set forth in claim 21, wherein the second classification technique utilizes a radial basis function.