Face Recognition Based on Generalized Canonical Correlation Analysis

We have proposed a new feature extraction method and a new feature fusion strategy based on generalized canonical correlation analysis (GCCA). The proposed method and strategy have been applied to facial feature extraction and recognition. Compared with the face feature extracted by canonical correlation analysis (CCA), as in a process of GCCA, it contains the class information of the training samples, thus, aiming for pattern classification it would improve the classification capability. Experimental results on ORL and Yale face image database have shown that the classification results based on GCCA method are superior to those based on CCA method. Moreover, those two methods are both better than the classical Eigenfaces or Fishierfaces method. In addition, the newly proposed feature fusion strategy is not only helpful for improving the recognition rate, but also useful for enriching the existing combination feature extraction methods.