Facial movement can be described by three temporal parameters:
Unfortunately, when and how a movement appears and disappears, and how co-occurrent movements integrate with each other (coarticulation effects) are difficult to quantify. Once a change in facial expression is specified, the rate of change can determine the perceived meaning of the intended emotion. Duration of facial expression of emotion is often related to the intensity of the emotion. Facial expressions due to felt or unfelt emotion, or voluntary or involuntary facial expressions, differ in their timing parameters. A fake or polite smile might occur too early or too late; its onset or offset could be too fast or too slow. There is evidence to support that someone who is trying to convey an emotion methodically (trying to mask the real emotion, for instance) will have faster onset and offset than someone giving a spontaneous emotion [36]. Also, there is a report that studied some socially skilled and socially unskilled subjects and found that the socially unskilled people ``may possess basic verbal communication skills but may fail to respond to the subtle social cues governing the timing of emotional displays'' [55]. The sending and reading of the timing of expressions is important in determining the real emotion.
Few data and studies relate to the problem of computing the onset and offset of facial expression. The three parameters of a facial action (onset, apex and offset) can be obtained by human observers. However, these parameters typically lack precision for facial animation because facial action is a more complex pattern than a simple decomposition in three linear parameters. In the case of speech, facial expressions (not only lip movements but also eyebrow movements, smile and so on) and speech are coordinated by a synchronization phenomenon defining the apex of these facial expressions. For the lip motion, it is important to know when the target of the given lip shape should be reached. For example, is it on the pitch of the segment? Or at the beginning of the segment? Also when does the movement decay? Then the problem of coarticulation arises. The coarticulation of movements tries to achieve the required facial action goals in a given time; however, in some cases there is not enough time to accomplish an action. A few computer vision systems are being developed which are capable of extracting temporal characteristics of facial expressions. Yacoob et al. [152] tracks expression generation over time for recognition, while Essa et al. [44][42] observes change in facial pattern and relates it to both muscle and FACS descriptions to extract coarticulation in facial expressions.