Investigating Spontaneous Facial Action Recognition through AAM Representations of the Face
Simon Lucey, Ahmed Bilal, F. Jeffrey · 2007
Face Recognition 276 actions.They typically occur in less controlled environments, with non-frontal pose, smaller face image size, small to moderate head movement, and less intense and often more complex facial actions.Spontaneous actions units are elicited indirectly from subjects through environmental variables (e.g., showing a subject something associated with happiness which then indirectly results in a smile).Although harder to collect and annotate, spontaneous facial actions are preferable to posed as they are representative of real world facial actions.Most automated facial action recognition systems have only been evaluated on posed facial action data [Donato et al., 1999, Tian et al., 2001] with only a small number of studies being conducted on spontaneous data [Braathen et al., 2001, Bartlett et al., 2005].This study extends much of the earlier work we conducted in [Lucey et al., 2006].We greatly expand upon our earlier work in a number of ways.First, we expand the number of AUs analyzed from 4, centered around the brow region, to 15, stemming from all regions of the face.Second, we investigate how representation affects both posed and spontaneous actions units by running our experiments across both kinds of datasets.Third, we report results in terms of verification performance (i.e., accept or reject that a claimed AU observation is that AU) rather than identification performance (i.e., determine out of a watchlist of AU combinations which class does this observation belong to?).The verification paradigm is preferable over identification as it provides a natural mechanism for dealing with simultaneously occurring AUs and is consistent with existing literature [Bartlett et al., 2005].