Facial Emotions and Emotion Intensity Levels Classification and Classification Evaluation

Abstract

In this paper we analyze the problem of human facial emotion and emotion intensity levels recognition and resulting classification accuracy evaluation. Final testing set classification accuracy value is usually taken as a quantifier of method quality. However, this value is often strongly affected by the testing set parameters such as number, age and gender of subjects or intensity of their emotions etc. In this work we propose a different classifier evaluation methodology that uses the human visual system as a reference point. We employed active appearance models and support vector machines for facial emotion classification. Our SVM classifier gave slightly more consistent labels to emotion categories for images than human subjects, while humans were more consistent at identifying emotion intensity level than SVM.