Independent component analysis: source assessment and separation, a Bayesian approach
Stephen John Roberts · IEE Proceedings - Vision Image and Signal Processing · 1998
The author presents a method of independent component analysis which assesses the most probable number of source sequences from a larger number of observed sequences and estimates the unknown source sequences and mixing matrix. The estimation of the number of true sources is regarded as a model-order estimation problem and is tackled under a Bayesian paradigm. The method is shown to give good results on both synthetic and real data.