Graphics Reference
In-Depth Information
6.2 Measures of facial expressiveness
Human codings: The most commonly used system is the Facial
Action Coding System, an anatomy-based method to describe visually
distinguishable facial movements. The FACS does not interpret the
facial expression, but detects inseparable facial movements, the so-
called 44 facial action units (AUs). Individual muscles create the basis
for AUs, but some AUs are produced by groups of muscles. Some
muscles can even produce several different AUs. With the help of the
44 AUs, it is possible to describe all emotional (EmFACS) and non-
emotional movements (such as communication signals) of the face. The
analysis is based on a video recording. Furthermore, the intensity and
temporal resolution can be captured (onset, apex, offset).
The Active Appearance Model (Cohn, 2010) and CERT (Bartlett et
al., 2008) are able to automatically recognize the AUs. Such automated
video processing systems for the recognition of facial movements are
based on the gathering and classification of features (Wimmer and
Radig, 2007). The efficiency of these methods is significantly increased
by taking into account the modeling of the facial geography and thus
the separation of dynamic (contours, coloring and contrasts) and static
features (Skelley et al., 2006).
6.3 Measures of psychobiological activity
The psychobiological emotion recognition can be subdivided into
traditional psychological research (Stemmler and Wacker, 2010;
Kreibig et al., 2007) and the area of affective computing (Picard et al.,
2001). It has been noticed, however, that the two areas are converging
(Kolodyazhniy et al., 2011; Walter et al., 2010; Hrabal et al., 2012). In
the basic research area and the affective computing area, discrete (fear,
anger, joy, etc.) and dimensional (valence, arousal, dominance) models
are used. In both research areas, however, the dimensional approach
has become more prevalent. Both areas use similar parameters: blood
volume pulse (BVP), skin conductance level (SCL), respiration (RSP)
and electromyography (EMG). Von Kolodyazhniy and coworkers (2011)
added additional parameters. Especially reliable are the correlations
between the dimensions valence and corrugator and/or Zygomaticus
EMG (Tan et al., 2012), as well as between arousal and SCL or BVP,
respectively. In this process, predominantly action potential and
frequency analyses are determined. Frequency analyses are generally
suitable for real-time applications.
The main difference between basic research and affective computing
is the following: In the basic research domain, averages and standard
Search WWH ::
Custom Search