Blind source separation and analysis of multispectral astronomical images

Celestial sources are studied by means of multispectral analysis, for which a pixel value can be considered as a mixture of different sources. We show that the application of blind source separation methods can display interesting features, which can help astronomers to improve their description of celestial sources. Experiments were carried out on Hubble Space Telescope images of the galaxy 3C 120. Different separation methods were applied, either based on the source correlation, or on non-Gaussian probability density functions. Each obtained source set was quantified by the mutual information between the sources. Different algorithms give quite similar decompositions which mainly correspond to real phenomena.