Face recognition by using generalized RBF neural networks

We put forward a new efficient face recognition scheme. It is robust to environmental noises, facial expressions and variations in pose. First, we reduce the dimensionality of the face patterns by Discrete Cosine Transform (DCT). Next to decrease the influence of the illumination and pose and to extract the most significant features of the pattern, the optimal set of discriminant vectors is used to process the DCT coefficient matrix. At last we put forward the concept of Generalized Radial Basis Function Neural Networks (GRBFNNs) to realize the identification of the person. And in order to accelerate the training process and improve the quality of the solution of the GRBFNNs, Particle Swarm Optimization (PSO) instead of back propagation algorithm is used to effectively train the GRBFNNs. The experimental results show that our scheme is effective.