Facial expression recognition based on complex wavelet transform

In this paper we propose a novel medium single-tree complex wavelet transform (ST-CWT) for feature extraction which is the key point in facial expression recognition system. STCWT possesses whole attractive properties including orientation selectivity, approximate shift-invariance and efficient computation of DT-CWT and can imitate Gabor wavelet more closely. We use the magnitude part of six subimages of all 4 levels and the low-frequency part obtained after ST-CWT as the representation of the image, use PCA to reduce dimension and Euclidean distance for classification. We do experiments on JAFFE database and CK database separately, and the highest average recognition rate reaches 96.83%.