Text this: Viewers perceive shape in pictures according to per-fixation perspective