Image Memorability Prediction Based on Machine Learning

Abstract

With the rapid development and popularization of computer network and multimedia technology, we are continuously exposed to photographs and images. Some images are easy to be remembered, but others are ignored or quickly forgotten. Previous studies have shown that image memorability is an intrinsic property of an image, which is used to measure the degree to which an image is later remembered or forgotten. Many works have shown that visual factors that influence the image memorability, such as highly memorable visual content, image depth information. Prior research has shown that image depth information has a positive relationship with its memorability score. Based on AMNet, this paper makes few improvements and utilizes depth cues to predict the image memorability scores. In this paper, the depth cues of the images are combined with the original image features to predict the final memorability score for the given image by utilizing the late fusion methodology. Experiments conducted on public image memorability datasets evaluate the effectiveness of the proposed model.