Research On Face Anti-spoofing Of Single Frame Image Based On Deep Learning

In recent years,face recognition technology has been widely used in various specific scenarios.However,the face recognition system is commonly attacked and spoofed by criminals with legal face photo.In order to improve the security of face recognition systems and make them distinguish real living faces from spoofing face attacks have become an urgent problem in face recognition system security.The methods of face anti-spoofing have been greatly improved from initial manual feature-based method to deep learning-based method.However,there are more and more types of single-frame image spoofing means,and there is an urgent need for a high-quality and large-scale face anti-spoofing dataset.In addition,liveness detection needs deeper feature information rather than normal convolutional neural network.Based on this,this paper proposes a mixed multi-channel feature extraction residual network,which can effectively extract features and detect the real living face and spoofing face.Furthermore,the trained network can be used to construct a real-time,non-interactive rapid face liveness detection.The main work is listed as follow:(1)Fully investigate the common face liveness detection methods and research status,starting from the security of face recognition technology to conduct face liveness detection method research,analyze and study several major methods of spoofing attacks,sort out and summarize face liveness detection fundamentals and traditional manual feature-based method,especially the advantages and difficulties of face liveness detection method based on deep learning.(2)Deep learning-based face liveness detection method requires a lot of data support.To this end,this paper investigates several common face anti-spoofing datasets,and then participate in construction the construction of the CW-FASD(Cloudwalk Face Anti-Spoofing database).The data set contains thousands of volunteers who use different collection devices to capture face videos and pictures in multiple scenarios.In order to enrich the data set,data enhancement work will also be carried out in the later stage to complete the data set of one million data.(3)Study on the face liveness detection method of single-frame image face based on mixed multi-channel feature extraction residual network.The methods of face spoofing attacks are complex and changeable,and most spoofing attack face manufacturing techniques are sophisticated,and after face detection and alignment processing,it is almost difficult to distinguish them from real living faces with the naked eye.When the traditional convolutional neural network is used to extract the facial features of a single-frame image,it is difficult to extract the difference between the real live face and the spoofing attack face.In addition,the traditional convolutional neural network often deepens or widens the network to extract deeper feature information,which will cause the network to degrade and increase the amount of network calculation.To this end,this paper designed a mixed multi-channel feature extraction residual network,the network mainly includes: 1 7 × 7 convolutional layer,8mixed multi-channel feature extraction residual modules,1 global average pooling,and fully connected.There are two kinds of mixed multi-channel feature extraction residual modules,which are the core modules of the mixed multi-channel feature extraction residual network;global average pooling can globally integrate the feature information output from the convolutional layer,and accept multi-scale Input to improve the robustness of the network model.(4)Design and carry out relevant verification experiments.Mainly verify the effectiveness of the mixed multi-channel feature extraction residual network in single-frame image face liveness detection,and compare whether the residual network is superior.Experimental results show that the proposed mixed multi-channel feature extraction residual network is effective and superior to the residual network,and the performance and efficiency of face liveness detection have been further improved.