Reconstructive discriminant analysis: A feature extraction method induced from linear regression classification

Based on linear regression, a novel method called reconstructive discriminant analysis (RDA) is developed for feature extraction and dimensionality reduction (DR). RDA is induced from linear Regression classification (LRC). LRC assumes each class lies on a linear subspace and finds the nearest subspace for a given sample. But the original space cannot guarantee that the given sample matches its nearest subspace. RDA is designed to make the samples match their nearest subspaces. Concretely, RDA characterizes the intra-class reconstruction scatter as well as the inter-class reconstruction scatter, seeking to find the projections that simultaneously maximize the inter-class reconstruction scatter and minimize the intra-class reconstruction scatter. Actually, RDA can also be seen as another form of classical linear discriminant analysis (LDA) from the reconstructive view. The proposed method is applied to face and finger knuckle print recognition on the ORL, extended YALE-B, FERET face databases and the PolyU finger knuckle print database. The experimental results demonstrate the superiority of the proposed method.