Semi-supervised bi-directional dimensionality reduction for face recognition

Abstract

A novel method for face recognition called semi-supervised bi-directional dimensionality reduction (SSBDR) is proposed. Based on semi-supervised learning, domain knowledge in the form of pairwise constraints besides abundant unlabeled examples are available, which specifies whether a pair of instances belong to the same class or not. Compared to the semi-supervised dimensionality reduction (SSDR), it can not only preserve the intrinsic structure of the unlabeled data as well as both the must-link (the same class) and cannot-link constraints (different classes) defined on the labeled examples in the projected low-dimensional space, but also constructs two image covariance matrices directly by the original image matrix in two directions which can reduce the dimension of the original image matrix in two directions. The validity of this method can be verified by the experiments on ORL face database.