Face Anti-spoofing Based On Multi-feature Fusion

Trusted identity recognition is an important cornerstone of public security in the information age.As the face recognition system with the role of identity authentication is applied in all aspects of life,face anti-spoofing has also received widespread attention.This paper mainly focuses on the research of face anti-spoofing algorithm in trusted identity authentication.In face recognition system,uncontrolled camouflage attack has various ways,diverse scenes and diverse materials,which leads to the result that the target of camouflage attack is distinguished from each other and the distinguishable features are a few.Besides,the fact that the image acquisition domains are different also worsens the problem that there are less distinguishable features.In addition,as face anti-spoofing algorithms is based on deep neural networks,it has large computational complexity and high model complexity,which makes it difficult to deploy in application platforms with limited computing resources.This paper mainly studies the problems mentioned above in the following aspects.Firstly,starting from looking for various attacking objects,this paper adopts the edge information to amplify differences and improve the loss function of depth information monitoring network in existing algorithms so as to solve the problems,like background dependence of the attacking object and the lack of depth information caused by non-rigid motion.Besides,it also improves the performance of differential features by integrating edge feature network and depth information network to achieve a high-performance face anti-spoofing algorithm based on the fusion of face edge and depth information.Secondly,this paper deeply analyses the reasons for the difference of the target of camouflage attack from the existing cross-domain mobility of face anti-spoofing algorithms,using grouping regularization to strengthen the ability of automatic feature selection and noise suppression learning of the model,and improves the robustness of the algorithm through multi-loss joint learning,so as to achieve face anti-spoofing computation with high anti-spoofing performance and strong generalization ability.Finally,the existing face anti-counterfeiting algorithm based on deep learning shows poor performances in large computation,strict requirement for computing resources,and is difficult to achieve real-time in the application platform with limited resources.Using convolution reconstruction and feature re-calibration,the model parameters can be compressed as much as possible to reduce the complexity of the model while ensuring the accuracy of the model,so that the algorithm can be deployed on the application platform.Technical support is provided.