A study of singular value decomposition of face image matrix

In this paper, singular value decomposition (SVD) and its stability, proportion invariance, and rotation invariance of image matrix are discussed in detail. Optimal approximation of rank reduction matrix is studied. As for the Olivetti Research Laboratory (ORL) face database, experimental results demonstrate that it is effective and feasible that singular values of face image matrix are used as recognition features in face recognition, and that singular value features take on separability, stability and independence, and are valid in feature extraction of face image.