Emotion Detection Using Hybrid Structural and Appearance Descriptors.

In computer vision the facial expression recognition descriptors extracted from raw images are categorized as structural or appearance descriptors. A lot of effort has been done in the literature for improving both type of descriptors for making them more robust; in most cases, both types of descriptors have been used separately. In this work we propose a hybrid model that uses both descriptors for emotion inferring. Our model is based in detecting Action Units and uses a probabilistic approach for emotion prediction based on an ensemble of Support Vector Machine classifiers. Fully detailed inner workings of the method are provided for experiment replication as well as detailed results to assess emotion inferring performance.