Prototype-based class-specific nonlinear subspace learning for large-scale face verification

In this paper, we describe a face verification method which is based on non-linear class-specific discriminant subspace learning. We follow the Kernel Spectral Regression approach to this end and employ a prototype-based approximate kernel regression scheme in order to scale the method for large-scale nonlinear discriminant learning. Experiments on two publicly available facial image databases show the effectiveness of the proposed approach, since it scales well with the data size and outperforms related approaches.