Semi-supervised nearest neighbor discriminant analysis using local mean for face recognition

Abstract

Feature extraction is the key problem of face recognition. In this paper, we propose a new feature extraction method, called semi-supervised local mean-based discriminant analysis (SLMNND). SLMNND aims to find a set of projection vectors which respect the discriminant structure inferred from the labeled data points, as well as the intrinsic geometrical structure inferred from both labeled and unlabeled data points. Experiments on the famous ORL and AR face image databases demonstrate the effectiveness of our method. © 2010 Springer-Verlag.