I remember doing some blind source separation for audio when i was at school, that allows you to discriminate multiple voices in noise environment. I wonder what would be the set of output signals in the case of images ... maybe the image illuminated by different sources?