Investigating Emotion Style in Human Faces Using Clustering Methods

For the past decade, performance-driven animation has been a reality in games and movies. While capturing and transferring emotions from human beings to avatars is a reasonably solved problem, it is accepted that humans express themselves in different ways, with personal styles, even when performing the same action. This article proposes a method to extract the style of human beings’ facial movement when expressing emotions in posed images. We hypothesize that personal facial styles may be detected by clustering methods based on the similarity of individuals’ facial expressions. We use the K-Means and Gaussian mixture model clustering methods to group emotion styles. In addition, extracted styles are considered to generate facial expressions in virtual humans and are tested with users. After an evaluation using both quantitative and qualitative criteria, our results indicate that facial expression styles do exist and can be grouped using quantitative computational methods.