Characterizing internal models of the visual environment