Towards a dynamic computational model of visual attention