In this paper we study the facial expression recognition using both low level visual features and high level semantic rules. First, the facial landmark points are localized by Active Shape Model. The key expression regions are then extracted. Second, the Local Binary Pattern features are extracted and a deep neural network is trained by Restricted Boltzmann Machine. Third, the output of the neural network is used for semantic inference, and fuzzy inference system is adopted to implement the

more »

... level decision system. Finally, experiments are carried out on the Japanese Female Facial Expression Database. Results show that the proposed system constantly provides higher recognition rates over two traditional methods.