Bayesian Source Separation of Linear and Linear-quadratic Mixtures Using Truncated Priors

Authors:
Leonardo Tomazeli Duarte;Christian Jutten;Saïd Moussaoui
Affiliations:
Laboratory of Signal Processing for Communications (DSPCom), University of Campinas (UNICAMP), Campinas, Brazil;GIPSA-lab (UMR CNRS 5216), Institut Polytechnique de Grenoble, Grenoble Cedex, France 38402;IRCCyN (UMR CNRS 6597), Ecole Centrale Nantes, Nantes Cedex 03, France 44321
Venue:
Journal of Signal Processing Systems
Year:
2011

Citing 9
Cited 0

Fundamentals of statistical signal processing: estimation theory

Fundamentals of statistical signal processing: estimation theory
Independent component analysis, a new concept?

Signal Processing - Special issue on higher order statistics
Separating a Real-Life Nonlinear Image Mixture

The Journal of Machine Learning Research
Fast correction of bleed-through distortion in grayscale documents by a blind source separation technique

International Journal on Document Analysis and Recognition
Recurrent networks for separating extractable-target nonlinear mixtures. Part I: Non-blind configurations

Signal Processing
Blind separation of linear-quadratic mixtures of real sources using a recurrent structure

IWANN '03 Proceedings of the 7th International Work-Conference on Artificial and Natural Neural Networks: Part II: Artificial Neural Nets Problem Solving Methods
Source separation in post-nonlinear mixtures

IEEE Transactions on Signal Processing
Separation of Non-Negative Mixture of Non-Negative Sources Using a Bayesian Approach and MCMC Sampling

IEEE Transactions on Signal Processing
Show-through cancellation in scans of duplex printed documents

IEEE Transactions on Image Processing

Quantified Score

Hi-index	0.00

Visualization

Abstract

In this work, we propose a Bayesian source separation method of linear-quadratic (LQ) and linear mixtures. Since our method relies on truncated prior distributions, it is particularly useful when the bounds of the sources and of the mixing coefficients are known in advance; this is the case, for instance, in non-negative matrix factorization. To implement our idea, we consider a Gibbs' sampler equipped with latent variables, which are set to simplify the sampling steps. Experiments with synthetic data point out that the new proposal performs well in situations where classical ICA-based solutions fail to separate the sources. Moreover, in order to illustrate the application of our method to actual data, we consider the problem of separating scanned images.