A Computational Foundation for Affective Analysis of Social Images

作者:Wang Xiaohui*; Qin Jingyan
来源:12th International Conference on Fuzzy Systems and Knowledge Discovery (FSKD), 2015-08-15 to 2015-08-17.

摘要

Social image is an image in social networks, which has two characters: the image itself and its social environment. We build a computational foundation for affective analysis of social images from these two aspects. From the aspect of the image itself, inspired by the appraisal theories of emotions in psychology, we adopt the aesthetic effects as the middle layer to investigate the mental processes underlying the elicitation of emotions. From the aspect of its social environment, we explore the correlations between social attributes and visual features of images. Experiments on 1.2 million Flickr images show that distributions of aesthetic effects are different for various emotional categories, which displays some tendencies. And social related phenomena are social images specific. For example, color features between friends' images are similar, especially in a time period; and images in a group tend to have the similar features, which implies some kinds of interest aggregation. Based on the observations and conclusions, we also develop two useful prototype applications demonstrating the practical applicability of our findings.