Real-Time Facial Patterns Mining and Emotion Tracking

In this paper, we present a real-time method for mining the facial patterns and tracking the facial emotion. We first recognize the facial patterns based on a nonsupervised idea of pattern mining from video. To raise the mining speed, we adopt some simple and rapid filters called Gabor-like filters to mine individual feature patterns. To evaluate the facial patterns, we match deformable templates and analyze the histogram of local region. To track the mouth contour, we model dynamically the color distribution of lip and skin by EM algorithm and segment lip color by the Bayes classification rule. Finally, we represent the facial emotion with an avatar by transferring some parameters. Through experiments, our method can rapidly mine the facial patterns and track different facial expressions of different people whether they wear glasses or not.