The Research Of Face Recognition Methot Based On Deep Learning

With the continuous development of human society, people’s social identity is becoming increasingly importance. Face Recognition has been widely used in security, financial services, e-government and other areas. It’s important to improve the performance of Face Recognition for expansioning applications. The key to improve the robustness of Face Recognition is the robust representation of face feature, the more discrimination of feature selection and the good classification ability of classifier.On the basis of elaborating the theory and methods of Face Recognition Technology and Deep Learning, we start our research on the route of Deep AutoEncoder Neural Network, Convolution Neural Network and Face Recognition. Research and innovation of this paper include the following contents:(1) Summary the research status of Deep Learning. By studying the relevant literatures at home and abroad, we analyze the exisiting methods of Deep Learning, and introduce the basic model and theory and related evolution of Deep Learning. Deep Learning is a relatively new theory, it will lead to a long-term development by applying the idea of Deep Learning to the research of Face Recognition.(2) We propose a method of Face Recognition based on Deep AutoEncoder Neural Network. After the image pre-processing, we design a Deep AutoEncoder Neural Network Model. By fine-tuning, optimizing the sparsity parameters and the nodes of hidden layers and the numbers of hidden layers, we determine the optimal network model. We propose a method of extracting facial feature based on Deep AutoEncoder Neural Network, and Softmax classifier is used to classify feature. Experiments on ORL face database show that the algorithm in this paper achieves a more satisfactory results on the Face Recognition.(3) We propose a method of Face Recognition based on Deep Convolution Neural Network. Firstly, we pre-processe the image. And then we use a Linear Decoder to extracte the local statistical characteristics, which is a Convolution Kernel of Deep Convolution Neural Network Model we designed. By optimizing the size of sample blocks, the nodes of hidden layers and the scales of pooling, we get a optimal Network Model. Finally, we use the optimal Convolution Neural Network to extract convolution feature of the whole image. A large number of experiments on ORL face database show that the algorithm in this paper has a increased recognition rate compared with classical algorithm of Face Recognition.