Semantic features for face recognition

Face recognition is an important part of many computer vision applications. In this paper, we present a face recognition algorithm based on semantic feature extraction and tensor subspace analysis. The semantic features we employ include eyes and mouth, plus the region outlined by the three weight centres of the edges between them. We then evaluate these features using tensor subspace analysis. Singular value decomposition is used to solve the eigenvector problem and to project the geometrical properties to the face manifold. Experimental results demonstrate that our proposed algorithm performs well, and that it is capable of achieving more accurate convergence coupled with a lower computational demand compared to standard approaches.