Blending and Exaggeration of Animated Facial Expressions of Emotion

Facial expressions have a very important role in social communication. As the development of technology enables more and more realistic virtual humans, they may become much more common in our everyday lives, and believable facial behavior is crucial for the success of a virtual character. This thesis states that the facial expressions of virtual characters should be designed so that they would be able to express a rich variety of different emotions and emotional intensities, and that blending and exaggerating facial expressions can be used to bring variety to facial expressions. This thesis also considers the uncanny valley phenomenon, which implies that even though increasing realism increases familiarity and likability of cartoony characters, very realistic characters can appear eerie, disgusting and horrifying instead. This thesis explores the possibility to express complex emotions by blending the expressions of basic emotions: anger, disgust, fear, joy, sadness and surprise. The thesis introduces a physically-based facial model and a method of producing anatomically correct basic emotion blends. Using a series of questionnaires, it was found that while some of the blends were perceived as pure basic emotions or combinations of two basic emotions, some blends induced perceptions of complex emotions, such as shame or malicious joy. It is generally believed in the animation community that the expressions and movements of less realistic characters should be exaggerated to convey sufficient emotional intensity, but exaggerating highly realistic characters can lead to a strange appearance. This belief was supported by a study using a questionnaire to measure how people perceive the same facial expressions on different levels of realism and with different levels of exaggeration. Moreover, it was found that the maximum emotional intensity that can be reached without a strange appearance appears to be roughly the same on all levels of realism. In the uncanny valley literature, it is usually assumed that perceived strangeness is primarily associated with a negative emotional response, such as fear and disgust. This thesis introduces a study where emotional reactions to facial images were studied using facial electromyography and a self-assessment questionnaire. It was found that the primary emotional reaction to strangeness was amusement, instead of negative emotions. This thesis provides insight into how blending and exaggeration can be used to create believable facial animation with expressions of various emotions on various intensities. Furthermore, this thesis advances the understanding of perception of facial expressions in general, as well as the uncanny valley phenomenon.