A PDF-Matched Modification to Stone's Measure of Predictability for Blind Source Separation

Authors:
Mahdi Khosravy;Mohammad Reza Alsharif;Katsumi Yamashita
Affiliations:
Department of Information Engineering, Faculty of Engineering, University of the Ryukyus, Okinawa, Japan 903-0213;Department of Information Engineering, Faculty of Engineering, University of the Ryukyus, Okinawa, Japan 903-0213;Graduate School of Engineering, Osaka Prefecture University, Osaka, Japan
Venue:
ISNN '09 Proceedings of the 6th International Symposium on Neural Networks on Advances in Neural Networks
Year:
2009

Citing 6
Cited 0

An information-maximization approach to blind separation and blind deconvolution

Neural Computation
Adaptive Blind Signal and Image Processing: Learning Algorithms and Applications

Adaptive Blind Signal and Image Processing: Learning Algorithms and Applications
Independent Component Analysis: A Tutorial Introduction

Independent Component Analysis: A Tutorial Introduction
A Note on Stone's Conjecture of Blind Signal Separation

Neural Computation
Blind Source Separation Using Temporal Predictability

Neural Computation
Nonlinear blind source separation using kernels

IEEE Transactions on Neural Networks

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper presents a PDF-matched modification to Stone's measure of predictability. The modified measure of predictability is a measure of non-gaussianity too. It is an extent of signal predictability by two different prediction terms. One prediction term is based on a normal gaussian PDF assumption for signal. In contrast, the other one is based on a unit variance supergaussian PDF assumption for signal. By contrastive deployment of the above prediction terms, the modified measure of predictability enables BSS to follow a high kurtosis PDF assumption for signals. As an advantage, not only signals with maximized predictability are recovered, but also with increased non-gaussianity too. Deploying the modified measure of predictability concludes more independent recovered signals. The dominance of BSS based on the modified measure to the previous one has been demonstrated by many tests performed over mixtures of realistic audio signals (music and speech) and over mixtures of gray-scale images.