Research On Face Anti-Spoofing Algorithm Based On Deep Learning

With the continuous development of computer vision,face recognition has become a research hotspot in the field of computer vision.In order to prevent criminals from maliciously forging or stealing the facial features of others to attack the face recognition system,many researchers have begun to study face anti-spoofing.Face anti-spoofing,as an important guarantee in face recognition systems,has been widely used in many fields such as finance,intelligent security,teaching attendance and property management.The current single-modal face anti-spoofing has become increasingly mature,but in practical applications,due to the changing environment in complex scenes,the single-modal face anti-spoofing system has the problem of poor robustness.With the popularity of infrared and depth cameras,and infrared pictures can measure the heat radiated from the face,depth pictures contain richer structural features.Therefore,this paper focuses on the security issues in the biometric authentication system,and proposes a multi-modal face anti-spoofing method,which extracts features from different modal face images,uses a multi-modality feature fusion algorithm and facial attention.The mechanism algorithm optimizes traditional face antispoofing technology.The main work of this paper is as follows:1.A face anti-spoofing algorithm based on multi-modality feature fusion is proposed.In the feature fusion stage,the different features of the single-modal face are fused by different convolutional layers to increase the separability of the difference between the spoofing face and the real face,and then the neural network with the optimal number of layers is selected for different modes.Face features are fused,and the discriminator is trained using the more discriminative features to improve the accuracy of the face anti-spoofing algorithm.2.A face anti-spoofing algorithm based on attention mechanism is proposed.Design an end-to-end human face detection feature extraction network.By using two attention mechanisms,combining global features and local features of the face,more discriminative features are extracted for face anti-spoofing.The entire network is trained using joint loss of cross-entropy loss and central loss.The use of the attention mechanism reduces the overfitting of the model and improves the accuracy of the model to a certain extent.Finally,this paper experiments on the two algorithms proposed in this paper on the OULUNPU and CASIA-SURF datasets.Experimental results indicate that the algorithm proposed in this paper has strong advantages and can effectively solve the problem of multi-modality face anti-spoofing.