Research On Face Anti-spoofing Based On Fusion Strategy

With the gradual commercial use of face recognition system,the era of "looking at the face" has brought new challenges into this field.It is urgent to automatically and efficiently distinguish the authenticity of images and resist presentation attacks to ensure the security of the system.To design method for face liveness detection with high detection accuracy,strong generalization ability and real-time needs is the focus of current research.This paper summarizes the current research results of face liveness detection at home and abroad,and designs several face anti-spoofing algorithms based on deep learning algorithm and fusion strategy.The main work of this paper is as follows:(1)The algorithms of face anti-spoofing were reviewed.After a lot of research,this paper divided the existing main methods into three categories: the method based on manual design feature representation,the method based on deep learning and the method based on fusion strategy.In this paper,these methods were reviewed,the advantages and disadvantages of each method were analyzed,and some points that need to be solved in face anti-spoofing were expounded,which provides convenience for later researchers.(2)A face liveness detection algorithm based on Mobile Netv2 was designed.Combined with the practical application of face anti-spoofing,the basic Mobile Netv2 is "slimming ",and the fine-tuned convolution network greatly reduces the parameters and computations amount without reducing the detection accuracy.Experiments on NUAA and Siw datasets show that this method has some validity,but the effect on complex data sets is general,so it is necessary to fuse other methods to use it.(3)A face liveness detection algorithm based on decision fusion is designed.Decision fusion of the results of several other popular in liveness detection algorithms using Stacking and voting in integrated learning effectively improved the accuracy and generalization ability of face liveness detection.(4)A face liveness detection algorithm based on feature fusion is designed.Based on the Mobile Netv2 network,the features were extracted from the RGB,HSV,LBP graph respectively,then fusion,and the real or fake of the face are judged by the Softmax layer.Finally,a lot of experiments have been done on NUAA and Siw dataset.The results shown that the fusion of different features could effectively improve the accuracy and generalization ability of the algorithm,and the lightweight network can meet the real-time requirements.