Research on a method of facial expression recognition

Facial expression recognition is necessary for designing any human-machine interfaces. A novel facial expression recognition method with wavelet packet decomposition and neural network ensemble is proposed in this paper. Firstly, the facial expression images are converted by wavelet packet decomposition method into many decomposed regions. Secondly, single BP neural networks are trained by the K-L transformed features of every decomposed region. Using the trained BP neural network to recognize facial expressions, 14 decomposed regions with the best recognition rate are chosen. At the same time, the decomposition mode of the wavelet packet is determined. Finally, the sub-networks are trained by the features of the 14 decomposed regions, and a facial expression classifier of the neural network composed with 9 sub-networks is determined, with a satisfactory recognition performance.