A Computational Perspective on Visual Attention