Face Recognition using combined Binary particle swarm optimization and Hidden layer of Artificial Neural Network

Face recognition is one of the challenging domains. We have seen artificial neural network perform very well in both detection and recognition. In this paper, we propose a novel method of feature extraction where features obtained at the end of hidden layer of neural network is utilised. This hidden layer output is our first level of features. On these features, we apply binary particle swarm optimisation (BPSO) to remove the redundancy, the few hidden units in the network. BPSO over hidden layer outputs can be implemented in two ways: 1) to apply BPSO over hidden layer in the training stage so the network is better optimised; 2) to directly use the BPSO on an optimised neural network's hidden layer output. Both the techniques performed well over traditional neural network and conventional BPSO. Experiments on FERET and LFW datasets show promising results.