Beyond bottom-up: Incorporating task-dependent influences into a computational model of spatial attention