Multilinear locality preserving canonical correlation analysis for face recognition

We propose in this paper a multilinear locality preserving canonical correlation analysis (MLPCCA) method for face recognition. Motivated by the fact that both spatial structure information within each face sample and local geometry information among multiple face samples are useful for facial image feature extraction, we utilize them simultaneously and derive an improved canonical correlation analysis algorithm - MLPCCA - to seek multiple sets of pairwise projection bases to maximize the correlation of two facial image sets. The proposed MLPCCA method is designed to characterize the potential nonlinear correlation of two image sets by utilizing both the spatial and local geometrical information, hence is more suitable for face recognition across large pose and illumination variants. Experimental results are presented to demonstrate the efficacy of the proposed method.