Visual Attention-Driven Spatial Pooling for Image Memorability

In daily life, humans demonstrate astounding ability to remember images they see on magazines, commercials, TV, the web and so on, but automatic prediction of intrinsic memorability of images using computer vision and machine learning techniques was not investigated until a few years ago. However, despite these recent advances, none of the available approaches makes use of any attentional mechanism, a fundamental aspect of human vision, which selects relevant image regions for higher-level processing. Our goal in this paper is to explore the role of visual attention in understanding memorability of images. In particular, we present an attention-driven spatial pooling strategy for image memorability and show that the regions estimated by bottom-up and object-level saliency maps are more effective in predicting memorability than considering a fixed spatial pyramid structure as in the previous studies.