Facial expressions reliably reflect an individual’s internal emotional state and form an important part of effective social interaction and communication. In clinical psychiatry, facial affect is routinely assessed, and any identified deviations from normal affective range and reactivity may signal the presence of a potential psychiatric disorder. An example is melancholic depression or ‘melancholia’ where facial immobility and non-reactivity are viewed as sensitive diagnostic indicators of the illness. However, affect in depressive disorders such as melancholia, and indeed psychiatric conditions more broadly, is largely assessed by clinicians, without biological or computational quantification. While such clinical assessment provides useful qualitative descriptors of illness features, the inherent subjectivity of this approach raises concerns regarding diagnostic reliability, and may hinder communication between clinicians. Methodological advances and algorithm development in the field of affective computing have the potential to overcome such limitations through objective characterization of facial features. Among these methods are implicit face analysis techniques, which are based on local spatio-temporal descriptors such as the space-time interest points and Bag-of-Words framework, and explicit face analysis techniques based on deformable model fitting methods such as Constrained Local Models and Active Appearance Models. In this chapter we overview these approaches and discuss their application toward detection and diagnosis of depressive disorders, in particular their capacity to delineate melancholia from the residual non-melancholic conditions.
|Title of host publication||Advances in Face Detection and Facial Image Analysis|
|Editors||Michal Kawulok, M. Emre Celebi, Bogdan Smolka|
|Place of Publication||Cham Switzerland|
|Number of pages||17|
|Publication status||Published - 2016|