Presentation 2003/10/17
Classifier Combination with Confidence Transformation
Cheng-Lin Liu,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) This paper investigates the effects of confidence transformation in combining multiple classifiers using various combination rules. The combination methods are tested in handwritten digit recognition by combining varying classifier sets. The classifier outputs are transformed to confidence measures by combining three scaling functions and three confidence types. The combination rules include fixed rules (sum-rule, product-rule, max-rule, median-rule. etc.) and trained rules (linear discriminants and weighted combination with various parameter estimation techniques). From the experimental results, we can draw some conclusions as follows. (1) Confidence transformation is beneficial to the performance of classifier combination, either by fixed rules or by trained rules. (2) Trained combination rules mostly outperform fixed rules, particularly when the classifier set contain weak classifiers. (3) In combination using linear discriminant, the support vector machine with linear kernel (linear SVM) performs best. (4) The weighted combination with optimized weights performs as well as the linear SVM.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Classifier combination / confidence transformation / fixed rules / linear discriminant / linear SVM / weighted combination
Paper # PRMU2003-130,NC2003-61
Date of Issue

Conference Information
Committee PRMU
Conference Date 2003/10/17(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Pattern Recognition and Media Understanding (PRMU)
Language ENG
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Classifier Combination with Confidence Transformation
Sub Title (in English)
Keyword(1) Classifier combination
Keyword(2) confidence transformation
Keyword(3) fixed rules
Keyword(4) linear discriminant
Keyword(5) linear SVM
Keyword(6) weighted combination
1st Author's Name Cheng-Lin Liu
1st Author's Affiliation Central Research Laboratory, Hitachi, Ltd.()
Date 2003/10/17
Paper # PRMU2003-130,NC2003-61
Volume (vol) vol.103
Number (no) 390
Page pp.pp.-
#Pages 6
Date of Issue