Blind Source Separation Of Natural Signals Based On Approximate Complexity Minimization

An approach to blind source separation is presented based on minimizing complexity. The diiculty of measuring complexity of signals is dealt with by assuming that the signals are Gaussian, time-correlated stochas-tic processes. In a special case, this approach coincides with previously proposed methods where the separating solution is obtained from the eigendecomposition of a cross-correlation matrix.