Context dependent face selection in a visual and semantic perspective by