Automatic Affect State Detection using Fiducial Points for Facial Expression Analysis

Current advancements in digital technology indicate that there is an opportunity to enhance computers with automated intelligence in order to understand human feelings and emotions that may be relevant to systems performance. Furthermore, one of the most important aspects of the Ubiquitous Computing paradigm is that machines should be characterised by autonomy and context awareness to facilitate more intelligent interaction. Therefore, there is an opportunity to enhance computer systems with automated intelligence in order to permit natural and reliable interaction similar to human-human interaction. Although various techniques have been proposed for automatically detecting a user’s affective state using facial expressions, this is still a research challenge in terms of achieving a consistently high level of classification accuracy. The current research probes the use of facial expressions as an input perception modality for computer systems. Facial expressions, which are deemed the most effective input channel in the domain of Affective Computing, are generated from the movements of facial muscles from different regions of the face; primarily the mouth, nose, eyes, eyebrows, and forehead. Subsequently, due to the correlation between facial expressions and human emotions, it is foreseen that automatic facial expression analysis will endow computer systems with the ability to recognise human affective states. The presented study considers the use of facial point distance vectors within the representation of facial expressions, along with investigations into a range of supervised machine learning techniques, for affective state classification. Results indicate a higher level of classification accuracy and robustness is achievable, in comparison to using standard Cartesian coordinates from the fiducial points.