ABSTRACT
KEYWORDS: Steerable pyramid, Feature extraction.
1 INTRODUCTION
As high recognition accuracy, traditional multiresolution recognition methods, the discrete wavelet transform (DWT) and Gabor are very popular (Devi, B.J. 2010, Shen L.L. & Ji Z. 2009). With translational and rotational invariance, Contourlet and Curvelet, steerable pyramid transform (SPT) are similar with the two-dimensional DWT (Boukabou, W.R. & Bouridane, A. 2008, Mandal, T. et al. 2007, Su C.Y, Y.T Zhuang & L. Huang 2005). However, most of traditional multi-resolution methods are unsupervised learning. There are errors once the recognition process deviates the correct direction. Semi-supervised learning is the methods which part labeled samples are used and many unlabeled samples. In the learning process, clustering and classification are performed by constraint condition (Song Y.Q. et al. 2008, Cui P. & R.B. Zhang 2012). To enhance robustness and accuracy, we propose a steerable pyramid-based semi-supervised discriminant power (SPSDP) for face recognition.
