A Blind Source Separation Algorithm Based on a Unifying Model
Xiaofei Shi, Renjie Liu, Xiaoming Liu, Li Li · 2006
This paper presents a blind source separation algorithm, which can separate the mixture of super- and sub-Gaussian sources. A weighed tri-Gaussian model is proposed to estimate super- and sub-Gaussian probability density. The model can represent a broader range of sub-Gaussian densities as compared to some sub-Gaussian estimating models. In the framework of natural gradient, we derive the parameterized nonlinear score functions. Model parameters are calculated through online learning. Applying to the mixture of images, experiment shows that the proposed algorithm can efficient separate the mixture of super- and sub-Gaussian sources and has better performance