Achievement

It is a difficult task for humans and computers and has applications in human computer interaction, consumer electronics, intelligent tutoring systems, etc. Current algorithms are able to detect discrete emotional states and facial expressions. However, an algorithm has yet to be realized that performs well in unconstrained, continuous video. We have developed three algorithms for continuous videos. They are based on (1) Emulating human visual system that processes dynamic and static information separately, and then fuses it. (2) Accounting for temporal dynamics over a larger time window in a hidden Markov model. (3) Computing the spatio-temporal morphology of the face in terms of SIFT-Flow energy. These algorithms have been validated them on a publicly available Grand Challenge data set. Published in ICIP 2012, ICPR 2012 and ICMI 2012 conferences.