Learning Visual Explanations for DCNN-Based Image Classifiers Using an Attention Mechanism