Supervised Vector Angle Embedding Learning for Face Recognition

Based on constructing neighborhood graphs, a method called supervised vector angle embedding (SVAE) was presented for face recognition. A graph including both positive edges and negative edges was constructed. It put a positive edge on two samples in the same class and a negative edge on two different class samples within k nearest neighbor each other. The measure in SVAE was the angle between two vectors instead of modulus in traditional methods. It does not require the estimation of the parameter in heat weight function, and can reduce the influence of luminance on face recognition. When test sample was embedded into low-dimensional space with preserving neighborhood vector angle, a classification called angle nearest neighbor was used for face recognition. Experiments on Yale and UMIST databases demonstrated the proposed approach was superior to other methods in terms of recognition accuracy.