Facial expression recognition based on the daul-tree complex wavelet transform and supervised spectral analysis

A novel feature extraction method is proposed in this paper named DTCW-SA which is based on the dual-tree complex wavelet transform and supervised spectral analysis for facial expression recognition. Although holding the property of multi-resolution entirely, compared with traditional wavelet and Gabor transform, the attractive characteristics of DT-CWT are better orientation selectivity, approximate shift-invariance and lower redundancy. Different with existing DT-CWT, we extend images to appropriate size by interpolation before transform instead of copying the values of last row or column when decomposition of each scale. What's more, the method of supervised spectral analysis can dig nonlinear information hidden in the data. Experiments on JAFFE database and CK database illustrate the efficiency of DTCW-SA, and the highest average rate of six expressions reaches 97.8% on CK database.