Self-face recognition is characterized by “bilateral gain” and by faster, more accurate performance which persists when faces are inverted