Complex Wavelet Feature Extraction for Video-based Face Recognition

A novel two dimensional complex wavelet transform (2D-CWT) for video-based face feature extraction is proposed. As 2D-CWT has its merit as follows: insensitive to image pixel shift, directional selectivity, and computation efficiency in the dual-tree structure, it turns out to be useful for face feature extraction in the video clips. In the proposed system, 2D-CWT features are extracted from the detected frontal face images and sent to an Artificial Neural Network (ANN) for recognition. The comparative experiments demonstrated that the proposed feature extraction method has the better recognition performance and less computation complexity than Gabor transform with the similar algorithm structure. The good face recognition results using the proposed feature extraction method are reported.