The Face Recognition Base On Deep Belief Network

With the development of our information level,automation become more and more common around us.surveillance cameras have become everywhere,in some places there have been fingerprint and face recognition access control system.In these field,face recognition technology plays an important role.Face recognition is an important part of biometrics recognition.It is a technology covering a variety of disciplines such as computer,image processing,signal processing,pattern recognition and psychology.Although the study of face recognition has a long history but still has a high research value.The Deep Belief Network(DBN)and the comparative divergence algorithm were proposed by the scholar Geoffrey Hinton in 2006.This opens a new way to deep learning,combining the depth of learning with large data present a possible path to artificial intelligence.The deep belief network is a generating model,and by training the weight of the neurons between them,we can let the whole neural network generate the training data according to the maximum probability.DBN has many uses it is a generation model that it can be used to generate data and extract features.add the classifier on the top of the network it can be used to transform the classification model for classification,in this paper we will use the DBN as a classification model for face recognition.1,Using the deep belief network and the contrast divergence algorithm to design and train the face recognition system,the experiment shows that the face recognition system based on the deep belief network can effectively accomplish the classification task.2,In the design of the network when the deep of the belief network layer,training network parameters such as the impact of the identification of factors to be analyzed and discussed,obtained pre-training related parameters and network layer of the optimal value.3 in the CMU PIE,Yale,UMIST face database to identify the system is useful,and the light,resolution,expression,attitude changes on the impact of the identification of experimental results,and with the traditional face recognition method for comparison,The test results show that the recognition algorithm of this paper can reach 90%of the recognition rate of human face,and the recognition rate can be improved by 1%to 2%compared with the traditional algorithm.