Improving Multi-view Facial Expression Recognition in Unconstrained Environments

Facial expression and emotion-related research has been a longstanding activity in psychology while computerized/automatic facial expression recognition of emotion is a relative recent and still emerging but active research area. Although many automatic computer systems have been proposed to address facial expression recognition problems, the majority of them fail to cope with the requirements of many practical application scenarios arising from either environmental factors or unexpected behavioural bias introduced by the users, such as illumination conditions and large head pose variation to the camera. In this thesis, two of the most influential and common issues raised in practical application scenarios when applying automatic facial expression recognition system are comprehensively explored and investigated. Through a series of experiments carried out under a proposed texture-based system framework for multi-view facial expression recognition, several novel texture feature representations are introduced for implementing multi-view facial expression recognition systems in practical environments, for which the state-of-the-art performance is achieved. In addition, a variety of novel categorization schemes for the configurations of an automatic multi-view facial expression recognition system is presented to address the impractical discrete categorization of facial expression of emotions in real-world scenarios. A significant improvement is observed when using the proposed categorizations in the proposed system framework using a novel implementation of the block based local ternary pattern approach.