Neural network approach to blind separation and enhancement of images

In this contribution we propose a new solution for the problem of blind separation of sources (for one dimensional signals and images) in the case that not only the waveform of sources is unknown, but also their number. For this purpose multi-layer neural networks with associated adaptive learning algorithms are developed. The primary source signals can have any non-Gaussian distribution, i.e. they can be sub-Gaussian and/or super-Gaussian. Computer experiments are presented which demonstrate the validity and high performance of the proposed approach.