MPCA+MDA: A novel approach for face recognition based on tensor objects

This paper presents a novel approach to solve the supervised dimensionality reduction problem and feature extraction by encoding an image object as a general tensor of 2-D/3-D order. In this paper a multilinear principal component analysis (MPCA) for tensor object feature extraction and then a multilinear discriminant analysis (MDA), to find the best subspaces have been proposed. It should be noted that both of the algorithms work with tensor objects so the structure of the objects has been never broken. Therefore we achieve a better result than the result of the traditional methods. The focus of these algorithms is avoiding the curse of dimensionality. Finally, a comprehensive experiments on ORL and FERET databases has been provided by encoding face images as 2-D or 3-D tensors to demonstrate that MPCA+MDA algorithm based on higher order tensors has the potential to outperform the traditional vector-based subspace learning algorithms such as Eigenface and Fisherface, especially in the cases with small sample sizes and curse of dimensionality dilemma.