Toward a computational model of visual attention