The Research On Face Recognition Based On Deep Learning

Face recognition is an identity authentication technology that uses the face as users' ID.Compared with fingerprint recognition and other traditional contact authentication methods,face recognition has the advantages of non-contact,fast and convenient.In terms of gait recognition,the accuracy of face recognition is extremely advantageous,so face recognition has become widely accepted as identity authentication.With the rapid development of deep learning,the tasks of target classification,recognition,and segmentation have made great progress and have been applied in many fields.As a typical classification task,the face recognition has also been greatly developed by deep learning algorithm.Compared with the traditional face recognition method,face recognition based on deep learning extracts highly abstract semantic information of the image through automatic optimization of the model,causing a huge advantage in recognition accuracy.In the paper,the face detection algorithm is used to segment the face and detect key points in the picture,and the cut face is aligned with the standard template based on the five key points of the face(both eyes,nose tip,and two mouth corners).the deep convolutional neural network that is consist of residual block extracts features from the face picture and calculates the cosine distance from the facial features in the library to determine which category the detected face belongs to,and uses the live detection algorithm to judge whether the face is captured by the camera is from a real legitimate user.The main work of this paper is:1.The key points detection algorithm is used to locate the key points of the upper eyelid,the lower eyelid and the corner of eyes.When the distance between the two corners of each eye is kept constant,the user's closed-eye and opened-eye state is inferred by the distance between the upper and the lower eyelids.The blinking action is used to determine whether it is a real person.In addition,you can set the times of blinking and time between opening eyes and closing eyes to form a secret action key to enhance safety.2.By the face detection algorithm,faces and face key points in the training set and the test set are located,and the affine transformation is performed based on the key points of the face of the standard template picture,and the face detection is applied again to the transformed picture,after which the face is split and scaled to the same scale.3.The residual block is used to form a deep residual network,and the three kind of residual blocks are verified and compared.The network is jointly trained by the Softmax Loss and Center Loss to reduce the intraclass distance.After the training,the output of the first fully connected layer is used as the feature vector of the face,and the useless features are removed by Principal Component Analysis(PCA).Finally,the cosine distance is used to judge the eigenvalue of the two feature vectors.4.The paper uses Web software to realize face recognition system,which integrates functions such as face detection,face recognition and live detection and so on.It includes two modules: register legal users and real-time face recognition.After registering legal users,the feature vectors of the legal user's face are saved in a document.When the user use the face recognition to log in the system,the feature vector of the face picture captured by the camera is obtained by the trained model,then the cosine distance between this feature vector and the feature vector in the document is calculated.The algorithm judges that the face is belong to the legitimate user if the average of all cosine distances is less than a certain threshold.In addition,it needs to be identified by live detection before logging in system.