Affective image classification by jointly using low-level visual features and interpretable aesthetic features

Affective image classification has drawn increasing attentions in the multimedia community, in which effectively describing the affective semantics plays an essential role. Although many features have been attempted, the affective gap, however, still remains a major challenge. In this paper, we propose a novel affective image classification algorithm by applying the combined low-level visual features and psychology and art theory-based aesthetic features to a supervised hierarchical classifier, which is constructed based on the support vector machine (SVM) and AdaBoost. We apply the combined features We have evaluated this algorithm against three state-of-the-art approaches in three two-dimensional discrete emotion spaces defined respectively by three pairs of adjectives with opposite meanings, including warmcool, light-heavy and static-dynamic. Our pilot results suggest that the proposed algorithm can provide more accurate affective image classification.