Here are three exemplary mixture decompositions performed using MILCA

The first one is a “cocktail party” problem with 5 audio sources (4 speakers, one music sample

and a white noise signal). We mixed these sources by a 5x5 non-negative random mixing matrix

and assumed that all 5 “microphone recordings” are available. The results of blind source separation

by MILCA are channels 1-5.

Next is an image restoration and de-noising problem. Again, 3 sources and a white noise signal

are mixed by a 4x4 random non-negative matrix.

A synthetic hyperspectral image - three species ("red", "green" and "blue") with distinct spectra distributed 

on a 100x100 grid. The observations are given by a three-way 100x100x1000 matrix  where the third 

dimension represents spectral resolution. The results of blind separation are spectra of components   

and their spatial distributions ("maps" of mixing coefficients or abundances).