Developmental Network and Its Application to Face Recognition

As a very important technique in biometric recognition, face recognition has many applications in our daily life. It is a very complex problem influenced by the different light condition, pose, head angles, and so on. 108 face images of 27 subjects in ORL face database are efficiently recognized with the developmental network. To testify the effect of developmental network on the face recognition, the influences of different initializing methods for the weights from X layer to Y layer of the DN, and different neuron numbers and competing neuron number top-k in Y layer, are studied. Experimental results show that when the neuron number in Y layer is bigger than the number of subjects to be recognized, the recognition accuracy can reach over 95%. Otherwise, the recognition accuracy decreases significantly. Under the condition of the same neuron numbers in Y layer, with the increasing of fired neuron number k, the recognition accuracy decreases.