Face Recognition using Canonical Correlation Analysis

Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) are well known techniques for face recognition. Both PCA and LDA by themselves have good recognition rates. We propose Canonical Correlation Analysis (CCA) for combining two feature extractors to improve the performance of the system, by obtaining the advantages of both. CCA finds the transformation for each extractor dataset and maximizes the correlation between them. The feature extractors used in our experiments are combinations of PCA, LDA and Discrete Cosine Transform (DCT)-based PCA. We conduct tests on Yale Face Database, AR Face Database and FERET Database, and we show that the combination of two method works better than the individual methods.