Transfer Learning of a Neural Network Using Deep Learning to Perform Face Recognition

Deep Learning can be used to solve different problems in image analysis and pattern recognition. Face recognition is it's one of the applications. The use of face recognition to solve different social problems like personal authentication and security is increasing rapidly. In past different face recognition techniques have been introduced like Fisherfaces, Eigenfaces, and LBPH. These methods have low accuracy, so neural networks are used. The deep learning based neural networks are the most accurate. The parameters of the network can be tuned to achieve high accuracy. Face recognition requires a data base of individuals to train a neural network using deep learning. The trained network is then capable of recognition. The framework of the facial recognition process using transfer learning of the pre-trained neural network is described in this paper. We have used AlexNet for recognition which is a pre-trained convolutional neural network. Transfer learning of this pre-trained network has given accuracy of 97.95%. It can be used to classify 1000 different people. For recognition purpose, it requires a large database at least one thousand images per class. The accuracy is although high as compared with techniques mentioned above.