Face Recognition Based On Deep Belief Network

Face recognition technology is one of the most popular biometric technologies.It has become a hot topic in the field of machine vision,pattern recognition and artificial intelligence.It has been widely used in many fields such as information security,intelligent monitoring and human-computer interaction.For traditional face recognition method to extract the information features of a face is relatively single,and classification algorithms have limitations,resulting in uncontrolled environment recognition accuracy and robustness of the problem of insufficient.This paper around the two key technical problems of face image feature extraction and pattern classifier selected conducts a research and puts forward a face recognition algorithm combining with fusion feature and depth neural network.Aiming at the problem that a single feature cannot fully and efficiently express facial information,this paper combines the advantages of three-patch local binary pattern(TPLBP)feature and gradient direction histogram(HOG)feature and uses the idea of feature fusion to propose a feature construction strategy combining TPLBP feature and HOG feature.This strategy first extracts the TPLBP feature of partitioned face image and the HOG feature of global face image,and then constructs a new TPLBP/HOG fusion feature vector by combining two different feature vectors in a cascade manner.Considering that the original fusion feature dimension is large,this paper uses the method of principal component analysis(PCA)to reduce the dimension of TPLBP/HOG fusion feature in order to speed up the training speed.In order to solve the problem that deep belief networks unable to distinguish between image noise and target information,and ignore important facial texture information.This paper proposes a facial recognition method of fusion feature strategy based on DBN.Firstly,the image of face is preprocessed by adaptive equalized,thereby weakening the influence of illumination on face recognition.Then,TPLBP texture feature and HOG structure feature of extracted face image are fused to obtain the fusion feature with complementary information.Based on fusion feature after dimensional reduction,after the dynamic search of DBN depth model parameters to determine the optimal value,a face recognition algorithm based on DBN was proposed.Finally,experiments were conducted on the basis of ORL?AR?Yale-B?LFW face database.The experimental results show that:(1)TPLBP/HOG fusion feature is better than the performance of single feature in the uncontrollable environment,,which reflects that the HOG feature of global face image and the TPLBP feature of partitioned face image have a good complementary effect.Thereby The proposed TPLBP/HOG fusion feature can be more comprehensively express face image information.(2)The proposed method has greatly improved the accuracy compared with the traditional SVM,KNN and DBN algorithms,and has strong robustness.Compared with Deep CNNs algorithm,the recognition accuracy of proposed method is slightly lower and the recognition efficiency is higher.(3)The proposed algorithm are in line with the application requirements of face recognition in non-restrictive scenes of large crowds in face recognition speed,face recognition accuracy,face feature extraction and computer configuration requirements.Thus,the proposed algorithmhas great application space and potential in non-restrictive scenes of large crowds.