Supervised and unsupervised face recognition method base on 3CCA

Feature fusion is widely used in face recognition as a new technology. As we all know, face recognition methods can be divided into two parts: supervised methods and unsupervised methods. So we want to propose a new algorithm which can get a better group of effective discriminant vectors for recognition by combining supervised methods and unsuperviseds method together. Hence we introduce a novel method for feature extraction and feature fusion based on the canonical correlation analysis (CCA). In particular, we extend the traditional CCA method to 3 sets CCA (3CCA), which can extract canonical correlation features from three different feature sets. Finally, we implement the supervised (LDA, DCV, DLDA) and unsupervised (PCA, LPP, SPP) methods for face recognition based on 3CCA. At the end of this paper, we compare our methods with other face recognition methods on AR face database, the experiment results show that our proposed approaches have better recognition performance compared with single supervised and unsupervised feature extraction algorithms.