An efficient illumination invariant face recognition technique using two dimensional linear discriminant analysis

An efficient illumination invariant face recognition method based on two-stage two dimensional linear discriminant analysis (2S2DLDA) is presented in this paper. The proposed method uses a reflectance-illumination model (RI-Model) based on maximum filter to obtain illumination invariants of an image. Various combinations of two dimensional feature extraction techniques (PCA, 2DPCA family and 2DLDA family) with RI-Model are analyzed for the first time in the paradigm of face recognition problem. A vital unresolved problem of 2DLDA is that it needs large feature matrix for the task of recognition, as it considers only row correlation. 2S2DLDA method overcomes this problem by considering both row and column correlations. Nearest Neighbourhood (NN) classification approach is adopted for classification. For experimental purpose Yale B and Extended Yale B face databases were used. The performance superiority of the combination RI-Model and 2S2DLDA (proposed) among all other combinations is established through extensive experiments.