Research On Face Recognition Technology Key Algorithms In The Process Of Security

In recent years,face recognition technology has been applied in many fields such as large airports,railway stations and other traffic hub security areas,and the security of people travelling has been improved.However,because of the non-coorperative factors such as facial expression,lighting condition variation and occlusion,the accuracy of recognition will be affected to some extent.In view of the above problems,the research of face recognition technology in the process of security detection is studied in the paper.The specific research contents are as follows.(1)Building face image data sets.Hanwang E camera and desktop resident ID card reading machine are used to collect images,and the face image and the ID card image are obtained,then the data sets of face and card images is established,where the data sets contains 342 face images(171 people,2 images per person).In addition,a large number of celebrity images are downloaded to establish a celebrity image data set,where the data set contains 6204 face images.(2)Designing a convolutional neural network recognition.The model is mainly based on the structure of LetNet-5 model,which adopts one input layer,two convolution layers,two pooling layers,two full connection layers and one output layer.Here,in the output layer the probability of matching the input image and class label is obtained through the softmax classifier,and the classification result can be acquired;and relu function and cross entropy function are used as activation function of the model and the loss function of the model respectively;in addition,in order to prevent over-fitting phenomenon in model training,dropout layer is added to the model.The experimental results show that the convolutional neural networks model can achieve better recognition rate for 1:1 and 1:N recognition.(3)Implementing multiple open source interfaces calls.The data sets of the celebrity pictures and the person's syndrome images are compared and tested by calling Baidu interface,Face++ interface and Dlib model interface respectively in Python language.The experimental results show that comparing with Baidu interface and Dlib model interface,Face++ face contrast interface can get better recognition results,where the accuracy rate of celebrity pictures comparison can reach 97.44%,and the accuracy rate of the personal identification images contrast can reach 87.72%.(4)Proposing a fusion method based on Naive Bayes.The face comparison results of several open source interfaces are fused based on Naive Bayesian theory as the final comparison result.In the process of experiment,the personal identification images and the celebrity images are used to test respectively,where the accuracy rates can reach 98.72% on celebrity image data set and 88.89% on personal image data set.The experimental results show that the fused face matching method can obtain better matching effect.In summary,for security inspection the fusion method based on Naive Bayes is more suitable.At the same time,the method has great promotion meaning and practical application significance for the enterprise attendance,window business,intelligent login and other authentication occasions.