Abstract:Micro-expression cognition is a vital useful input to develop
affective computing strategies in modern human-computer/robot
interaction. In this paper, an effective system for micro-expression
cognition and emotional regulation is described. As input, a
micro-expressional face is represented as a point in a 3D space
characterized by arousal, valence and stance factors. The capture
and recognition method of micro-expressions is based on a novel
combination of 3D-Gradient projection descriptor, multi-scale and
multi-direction Gabor filter bank and the gradient magnitude
weighted Nearest Neighbor Algorithm (NNA) in facial feature regions.
The main distinguishing feature of our work is that the emotional
regulation model does not simply provide the classification and jump
in terms of a set of discrete emotional labels, but that it operates
in a continuous 3D emotional space enabling a wide range of
intermediary emotional states to be obtained. The micro-expression
recognition method has been tested with the Yale University's facial
database and universal participants' facial database so that it is
capable of analyzing any adult subject, male or female in the
typical database and interactive process. Then the cognition and
emotion system has been applied to the human-robot interaction, and
the results are very encouraging and show that our micro-expression
cognition and emotion model is generally consistent with human brain
emotional regulation mechanisms.