A Model for Calculating Saliency from Both Input Image and Memory.