Feature Fusion Based Face Recognition Using EFM

This paper presents a fusing feature Fisher classifier (F3C) approach for face recognition, which is robust to moderate changes of illumination, pose and facial expression. In the F3C framework, a face image is first divided into smaller sub-images and then the discrete cosine transform (DCT) technique is applied to the whole face image and some sub-images to extract facial holistic and local features. After concatenating these DCT based facial holistic and local features to a facial fusing feature vector, the enhanced Fisher linear discriminant model (EFM) is employed to obtain a low-dimensional facial feature vector with enhanced discrimination power. Experiments on ORL and Yale face databases show that the proposed approach is superior to traditional methods, such as Eigenfaces and Fisherfaces.