| In recent years,with the continuous breakthroughs in computer technology,image processing methods and pattern recognition theory,the accuracy of biometric detection and recognition has been greatly improved.Face recognition has become the most important research direction of biometric recognition due to its contactless and unobtrusive features.It is widely used in the fields of mobile payment,security verification system and employee identification.As a non-rigid object,face recognition rate is limited due to the influence of expression,posture,light source,age and hairstyle.Especially in the case of few samples or serious external interference,the traditional face recognition rate is low,which cannot meet the requirements of engineering application.Therefore,this paper carries out the research of face recognition algorithm to solve the core theory and key technical problems based on the background of the employee authentication and factory security monitoring in small and medium-sized enterprises.On this basis,the system is developed and applied.The main work and achievements of this paper are as follows:1.A face recognition method based on principal component analysis(PCA)and particle swarm optimization(PSO)for directed acyclic graph kernel support vector machine(PCA-PDKSVM)is proposed to solve the problem of limited face samples,fewer categories and less illumination interference.In this method,face images are linearly dimensionally reduced by PCA,then DKSVM model is designed and constructed with the principal component component as input and the face label as output,followed by particle swarm optimization for parameter optimization,and finally training and testing.The experimental results show that for ORL face database,the face recognition rate of this method is 99.1% and the recognition time is 0.234 s,which is better than the traditional method and can meet the application requirements of employee authentication in small and medium-sized enterprises.2.A face recognition method based on PCA and back propagation neural network(BPNN)is proposed for the situation of limited face samples,fewer categories and more illumination interference.This method firstly reduces the dimension of face image by PCA,then designs the BPNN model with the principal component as input and the face label as output,and then proposes a hidden layer number optimization estimation method.Experimental results show that for F-AR face database,the face recognition rate of this method is 99.5%,slightly worse than that of pa-pdksvm,but the recognition time is 0.114 s,which is suitable for scenes requiring strict recognition time.3.A face recognition method based on improved LeNet-5(ILeNet-5)is proposed for the situation of limited face samples,many categories,and severe illumination and occlusion interference.Firstly,the convolution layer and pool layer number,convolution kernel number,number of neurons and activation function are designed and optimized to automatically extract facial features.Then the structure,number of layers and number of neurons are designed and optimized.At last,Softmax classifier is used for face recognition.The experimental results show that for AR face database,the recognition rate of this method is 95.3% and the recognition time is 0.182 s,which is better than the traditional face recognition method and can meet the application requirements of factory safety monitoring in small and medium-sized enterprises.4.Designed and developed a face recognition system based on Visual C# 2015 language,SQL Server2014 data management system and Devexpress control for the application scenarios of employee authentication and factory security monitoring in small and medium-sized enterprises.The system first generates COM components from PCA-PDKSVM algorithm and ILeNet-5 algorithm on Matlab platform,and then calls them by Visual C#.The system can realize face sample recognition,input and weight update.The test results show that the recognition rate of the system is high and the time delay is short. |
PDF Full Text Request