Research On The Face Liveness Detection Algorithm Based On Dual-Stream Convolutional Neural Network

Face liveness detection is a technology designed to determine whether the person being detected is a real person.It can effectively curb the abuse of face recognition systems and prevent attackers from committing spoofing by using the forged faces of legitimate users.Face liveness detection technology is widely used in fields such as finance and insurance to ensure the security and authenticity of user information.It should be noted that face liveness detection technology is not perfect,and attackers may still bypass detection using various means.So it is necessary to constantly improve the accuracy of the technology to improve the security of the entire system.Traditional face liveness detection methods usually use handcrafting features for face spoofing detection,and methods based on deep learning can also distinguish true and false faces through a single feature.However,these methods may face the bottleneck of accuracy and universality.In addition,the effectiveness of existing methods is affected by lighting conditions and equipment.Given the above problems,this thesis proposes two kinds of face detection methods that have good generality and can effectively cope with different illumination conditions.Specifically,the research content and innovative results of this thesis are as follows:(1)To address the problem of low generality of traditional face liveness detection algorithms and methods based on single difference cues for distinguishing real and fake faces,this thesis implements feature fusion by dual-stream convolutional neural networks to improve the generality of the algorithm.(2)To solve the problem that the existing face liveness detection algorithms are generally sensitive to lighting conditions,this thesis explores different methods for extracting highfrequency reflection information of faces and proposes a new method.And then incorporates the methods into dual-stream convolutional neural networks to reduce the sensitivity of the algorithms to different lighting conditions.(3)In view of the problem that the number of parameters based on the dual-stream convolutional neural network is large,this thesis adopts a lightweight backbone network and makes continuous improvements to reduce the number of parameters while improving the performance of the algorithm.(4)In this thesis,the attention mechanism is introduced to improve the fusion ability of temporal and spatial information of the algorithm,and the attention mechanism is improved to enhance the performance of the algorithm.(5)The proposed method is evaluated extensively in three mainstream face liveness detection datasets,and cross-dataset tests are also conducted to verify the universality of the method.The proposed algorithm for light treatment is also subjected to relevant experiments to verify the robustness of the method to light conditions.The experimental comparison shows that the two methods are satisfactory.