Two-Stage Tensor Locality-Preserving Projection Face Recognition
Ying Liu, Dimitris A. Pados, Chia‐Hung Yeh · 2016
Locality-preserving projection (LPP) is an efficient dimensionality reduction approach that preserves local relationships within data sets and uncovers essential manifold structures. In this paper, we develop a two-stage tensor locality-preserving projection for face recognition, in which first-stage tensor LPP is performed in the original tensor space of face images and second stage tensor LPP is performed in the reduced-dimension tensor subspace of the first-stage projection. For classification, we seek a non-negative sparse representation in the final low-dimensional tensor subspace and determine the class of an unknown face image by minimum sparse representation error. Experimental studies demonstrate that our proposed two-stage tensor LPP scheme along with the non-negative sparse representation classifier effectively exploits the locality structure of face images and outperforms existing state-of-the-art face recognition schemes.