The Face Recognition Based On Deep Learning

As one of the most popular biometrics technologies,face recognition technology has the characteristics of reliability,uniqueness and stability,making it widely used in many fields such as access control,document verification and judicial application.Neural network is the most noticeable direction of pattern recognition,especially in deep learning neural network,in which deep belief network(DBN)is a common deep learning architecture,DBN builds a multi-layer nonlinear network structure and rich training data,which is very hot in the field of face recognition.It has two advantages: The number of layers in the network structure could be deeper,which enabled the network structure to learn more elaborate information,and the non-tagged datas could be used.Based on a large number of research and literature on face recognition technology,this thesis has done the following works:(1)Aiming at the problem that only extracting a single feature can not express multi-angle informations in human face image very well,By using the advantages of local binary pattern(LBP)and histogram of oriented gradient(HOG),this paper proposes a new feature extraction method that fusions LBP and HOG.Firstly,the LBP and the HOG are used to extract the corresponding texture features and gradient features respectively.Then,the features are fused into a feature vector by means of vector connections to obtain richer facial features.Through the verification on the face databases of ORL,FERET and Yale,the results show that this method has a better detection effect in uncontrollable environment.(2)In view of the fact that the DBN ignores the texture and gradient informations of the image,that is,the problem that many effective features of the face image can not be learned well in the process of network learning,this paper proposes a new face recognition algorithm,First,the LBP feature and the HOG feature are fused,then the fusions features is used as the input for DBN.Through the verification on the face databases of ORL,FERET and Yale,the results show that the algorithm has better robustness in the real environment.(3)Based on the above,a face recognition smart key cylinder based on DBN is designed.The intelligent recognition system combines the ARM platform with the mobile APP,when the target object presses the doorbell to trigger the camera to collect a photo of the face.Then,the control panel extracts and fusions the features of the photo,and compares it with a pre-trained DBN model.If the comparison succeeds,the key cylinder automatically unlocks.If the matching fails,the photo is sent to the user's mobile phone APP by WIFI,and the user decides whether to unlock the key cylinder.Test shows that the cylinder has good practicality.