Fisher's linear discriminant (FLD) and support vector machine (SVM) in non-negative matrix factorization (NMF) residual space for face recognition

摘要

A novel method of Fisher's linear discriminant (FLD) in the residual space is put forward for the representation of face images for face recognition, which is robust to the slight local feature changes. The residual images are computed by subtracting the reconstructed images from the original face images, and the reconstructed images are obtained by performing non-negative matrix factorization (NMF) on original images. FLD is applied to the residual images for extracting FLD subspace and the corresponding coefficient matrices. Furthermore, features are obtained by mapping the residual image to FLD subspace. Finally, the features are utilized to train and test support vector machines (SVMs) for face recognition. The computer simulation illustrates that this method is effective on the ORL database and the extended Yale face database B.