Numerical experiments on blind separation of astrophysical maps by independent factor analysis

The celestial microwave radiation is generated by various astrophysical sources. Measuring this radiation does not enable us to immediately extract information about the sources, since the individual signals are superimposed to one another all over the measurement bandwidth. Each source process radiates in accordance with a typical frequency-emission law, which, unfortunately, is often unknown. Some blind technique should thus be used to separate the individual components from the total radiation. In this paper, we apply a recently introduced approach, called Independent Factor Analysis (IFA), which models the sources as mixtures of Gaussians. After checking the appropriateness of this model to our case, we briefly describe the IFA approach and show some results from simulated but realistic data.