Research On Anti-spoofing Face Recognition Algotithm And Its Application In Port Attendance

As an important means of biometric verification,facial recognition is used in all aspects of life,such as cell phone login,time and attendance,and access control.A good face recognition system needs fast and efficient face recognition algorithms in addition to face anti-spoofing detection features.However,the emergence of more and more potential presentation-based attacks(photo attacks,video playback attacks,and3 D attacks)has caused serious security risks and limited the further development of face recognition technology.Relying on the port anti-spoofing face recognition attendance system,this thesis researches face recognition algorithm and face antispoofing detection algorithm and implements an anti-spoofing port attendance system,the main work of this thesis is as follows:(1)To improve Arc Face by linearizing the Arc Face algorithm for the problem of angular oversaturation.By comparing the Arc Face face recognition algorithm with other algorithms experimentally,it is concluded that the Arc Face algorithm has the property of decision boundary stability.However,the cosine function in Arc Face is nonlinear,which may lead to angle oversaturation and thus the problem of insufficient angle optimization.The linearization idea of Lin Cos algorithm is used to deal with the nonlinear problem in Arcface,and the first K terms in the Taylor expansion are substituted for the cosine function,which can offset the effect of nonlinearity and prevent the model from overfitting,and also enhance the dispersion between different classes and the compactness between the same class.Finally,experiments show that the improved Arc Face algorithm is more accurate and effective.(2)A face anti-spoofing detection algorithm based on the fusion of color texture and r PPG frequency domain features is designed for the purpose that a single feature cannot detect all presentation-type attacks better.The feature extraction of face images is performed by two aspects,on the one hand,the color texture features are obtained by the improved MB-LBP algorithm,and on the other hand,the r PPG frequency domain features are extracted by signal processing techniques.These two aspects are then fused and trained using a classifier to finally determine whether it is a live body or not.Tested on 3DMAD dataset and MSU-MFSD dataset and compared with existing face anti-spoofing detection algorithms,the experiments show that the method has a high accuracy rate.(3)A face recognition based anti-cheating mobile attendance system is designed.In order to solve the difficulties of port attendance and the cheating problems of traditional attendance system,this thesis will incorporate face detection,face anticheating,face recognition and GPS technologies according to the attendance system of the port,which can realize the validity of employee punching location while verifying employee identity and preventing demo-type attacks.This system not only improves the efficiency of employee attendance in port enterprises,but also avoids the phenomenon of employee clocking cheating to a certain extent,which has some practical significance.