Deep Facial Expression Recognition Exploiting Categorical and Continuous Emotional Dependencies

Abstract

Facial Expression Recognition (FER) has been a topic of study in the field of computer vision and machine learning for decades. Despite huge efforts to improve the accuracy of FER systems, existing methods still are not generalizable and accurate enough for use in real-world applications. Most traditional methods use hand-crafted features for representation of facial images that often require rigorous hyper-parameter tuning to achieve favorable results. Over the past few years, deep learning methods have shown remarkable results in FER managing to achieve almost human performance in lab-controlled environments. However, recognizing facial expressions in real-world settings is still very challenging due to large variations, occlusions and the ambiguity of human emotion. Meanwhile, we have no clear evidence as to which emotion representation is more appropriate for FER. Numerous models describing the human emotional states have been proposed by the psychology community and the majority of FER systems use either the categorical or the dimensional model of affect. The goal of this diploma thesis is to explore the challenges that are present in the task and present novel deep learning techniques for recognizing facial expressions in-the-wild.