EP0621556A2
Expired / Lapsed
EP
⏱️ 0.0 years remaining
A process for combining the results of several classifiers
Assignee: Eastman Kodak Co
•
Filing: 1994-04-19
•
Granted: —
•
Expires: 2014-04-19
Abstract
Abstract A process for successfully combining the results of several classifiers provides a method for calculating confidences of each classification decision for every classifier involved. Confidences are then combined according to the Dempster-Shafer Theory of Evidence. Initially, basic probability assignments for each of the classifiers are calculated and used to calculate confidences for each classifier. The confidences for all of the classifiers are then combined. The combined confidences are then used to determine a class for the data input to the classifiers.
Inventors
- 👤 Galina C/O Eastman Kodak Company Rogova
Bibliographic Details
Application Number
—
Jurisdiction
EP
Data Source
google_patents
Last Synchronized
2026-11-09 15:18
CPC Classifications
G; G06; G06F; G06F18/00; G06F18/20; G06F18/25; G06F18/254
Claims Analysis
Total Claims: 10
Claim 1
Independent Claim
55 words
A method for combining results of several classifiers comprises the steps of: calculating basic probability assignments for each of the classifiers; using the basic probability assignments to calculate confidences for each classifier; combining the confidences for all of the classifiers; and using the combined confidences to determine a class for data input to the classifiers.
Claim 2
Dependent Claim
37 words
A method as claimed in claim 1 wherein the steps of calculating basic probability assignments and using the basic probability assignments to calculate confidences for each classifier comprise the step of applying a suitable theory of evidence.
Claim 3
Dependent Claim
66 words
A method as claimed in claim 1 wherein the step of using the basic probability assignments to calculate confidences for each classifier comprises the steps of: using a distance measure between a classifier output vector and a mean vector for a subset of training data corresponding to each class; and calculating evidences for all classification decisions for each classifier, using the distances as basic probability assignments.
Claim 4
Dependent Claim
19 words
A method as claimed in claim 3 wherein the distance measure comprises one of two almost equivalent distance measures.
Claim 5
Dependent Claim
34 words
A method as claimed in claim 4 wherein the first distance measure comprises a cosine between a classifier output vector and a mean vector for a subset of training data corresponding to each class.
Claim 6
Independent Claim
59 words
A system for combining the results of several classifiers comprising: means for calculating basic probability assignments for each of the classifiers; means for calculating confidences for each classifier from the basic probability assignments; means for combining the confidences for all of the classifiers; and means for determining a class for data input to the classifiers from the combined confidences.
Claim 7
Dependent Claim
26 words
A system as claimed in claim 6 wherein the means for calculating basic probability assignments and confidences for each classifier comprise a suitable theory of evidence.
Claim 8
Dependent Claim
58 words
A system as claimed in claim 7 wherein the means for calculating confidences for each classifier comprises: a distance measure between a classifier output vector and a mean vector for a subset of training data corresponding to each class; and means for calculating evidences for all classification decisions for each classifier, using the distances as basic probability assignments.
Claim 9
Dependent Claim
19 words
A system as claimed in claim 8 wherein the distance measure comprises one of two almost equivalent distance measures.
Claim 10
Dependent Claim
37 words
A system as claimed in claim 9 wherein the second distance measure comprises a function of Euclidean distance between a classifier output vector and a mean vector for a subset of training data corresponding to each class.
References Cited (Backward Citations)
No citation records indexed.
USPTO Annuity & Maintenance Windows
Maintenance schedule is calculated for granted US utility patents.