Extraction of stellar spectra from dense fields in hyperspectral muse data cubes using non-negative matrix factorization

In this article, we present a method to extract stellar spectra from dense field images of the MUSE instrument. MUSE is an instrument under construction which will provide hyperspectral astrophysical data cubes. Due to the PSF (Point Spread Function) effects, stars are not point-like objects in MUSE images, but spread with a certain radius so that we cannot distinguish stars that are too close in the images. Hence, there is a need for Source Separation methods, to extract stars spectra from the data cubes and permit their identification by astrophysicists. We first present our mixing model and explain how we adapt it to easily apply a source separation method, using available assumptions about MUSE PSF. We then propose a Non-negative Matrix Factorization method, coupled with least square estimation with non negativity constraints. We take advantage of some prior information about the data. Our approach may thus be considered as "semi blind".