Please use this identifier to cite or link to this item:
Title: A Chernoff-based approach to the estimation of transformation matrices for binary hypothesis testing
Authors: Lorenzo-García, F. D.
Ravelo-García, A. G. 
Navarro-Mesa, J. L. 
Martín-González, S. I. 
Quintana-Morales, P. J. 
Hernández-Pérez, E. 
UNESCO Clasification: 3325 Tecnología de las telecomunicaciones
Keywords: Recognition
Issue Date: 2006
Publisher: Institute of Electrical and Electronics Engineers (IEEE)
Journal: Proceedings of the ... IEEE International Conference on Acoustics, Speech, and Signal Processing 
Conference: 31st IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2006) 
Abstract: We present a new method for improving the classificacation score in the problem of binary hypothesis testing where the classes are modeled by a Gaussian mixture. We define a cost function which is based on the Chernoff distance and from it a transformation matrix is estimated that maximizes the separation between the classes. Once defined the cost function we derive an iterative method for which we give a simplified version where one mixture component per class is previously selected to participate in the estimation. The initialization of the method is studied and we give two possibilities for this. One is based on the Bhattacharyya distance and the other is based on the average divergence measure. The experiments are carried out over a database of speech with and without pathology and show that our approach represents an improvement in classification scores over other methods also based on matrix transformation.
ISBN: 1-­4244-­0469-­X/06
ISSN: 1520-6149
Source: Proceedings of the ... IEEE International Conference on Acoustics, Speech, and Signal Processing [ISSN 1520-6149], v. 5, p. 753-756, (2006)
Appears in Collections:Actas de congresos
Unknown (152,72 kB)
Show full item record


checked on Mar 7, 2021


checked on Aug 23, 2020

Page view(s)

checked on Mar 7, 2021


checked on Mar 7, 2021

Google ScholarTM




Export metadata

Items in accedaCRIS are protected by copyright, with all rights reserved, unless otherwise indicated.