Local Fisher Discriminant Embedding for face recognition

An appearance-based face recognition approach called the local Fisher discriminant embedding (LFDE) method is proposed in this paper. By using LFDE, the face images are mapped into a face subspace for analysis. Different from linear discriminant analysis (LDA), which effectively sees only the Euclidean structure of face space, LFDE finds an embedding that preserves local information and obtains a face subspace that best detects the essential face manifold structure. Different from locality preserving projections (LPP) and unsupervised discriminant projections (UDP) which ignore the class label information, LFDE searches for the project axes on which the data points of different classes are far from each other while requiring data points of the same class to be close to each other. We compare the proposed LFDE approach with PCA, FDA, LPP, and UDP on two different face databases. Experimental results suggest that the proposed LFDE approach provides a better representation and achieves higher accuracy in face recognition.