Blind Source Separation of Multispectral Astronomical Images

Multispectral images lead to classify pixels, but often with the drawback that each pixel value is the result of a combination of different sources. We examined the ability of Blind Source Separation (BSS) methods to restore the independent sources. We tested different tools on HST images of the Seyfert galaxy 3C120: the Karhunen-Loéve expansion based on the diagonalization of the cross correlation matrix, algorithms which maximize contrast functions and programs which take into account the cross correlation between shift sources. With the last tools we obtained similar decompositions corresponding mainly to real phenomena. BSS can be considered as an interesting exploratory tool for astronomical data mining.