Reconnaissance faciale des émotions

: Un regard multidisciplinaire

Abstract

Facial emotion recognition, although yet in a developping phase, has been a growing trend for the last decades, challenging a multidisciplinary body of research. The emotional cues on our faces are of the highest interest for the GAFA-like industry giants, seeking to enrich their contents with user-contextual data. Actors of surveillance and state security matters challenge the techonology, with the expectation that an « augmented »facial recognition, able to discern
our emotional affects would allow them to conduct better behaviour analysis and prediction. In this thesis, we first review the traditional models in the theories of emotions. We then illustrate the most dominant of them, originating from psychologist Paul Ekman’s works, and depict how it currently governs most of the conceptual construction principles of existing systems. We also review the contruction methodologies of training datasets, and codification schemes. As a
conclusion, we assess the scientific validity of the concept of emotion recognition by questionning theoretic foundations and ground thruth construction.

Date of Award	Jun 2018
Original language	French
Awarding Institution	University of Namur
Supervisor	Claire Lobet-Maris (Supervisor)