Saliency model-based face segmentation and tracking in head-and-shoulder video sequences