Classification of affective semantics in images based on discrete and dimensional models of emotions

The classification of affective semantics in images is a very challenging research direction that gains more and more attention in the research community. However, as an emerging topic, contributions remain relatively rare, and a lot of issues need to be addressed particularly concerning the three following fundamentals problems: emotion representation, image features used to represent emotions and classification schemes designed to handle the distinctive characteristics of emotions. Thus, we present in this paper two classification approaches based on the dimensional and discrete emotion models. Traditional and emotional image features are used as input of classifiers relying on neural networks and on the evidence theory whose interesting properties allow to handle the ambiguous and subjective nature of emotions as it has been brought to the fore by our experimental results.