Research And Implementation Of Face Anti-Spoofing Algorithm

Face recognition systems have a wide range of application needs in security and payment systems due to their security and convenience.To ensure the security of face detection systems,face anti-spoofing system detects whether there are presentation attacks such as printing,video replay,mask disguise,etc.to deceive the face detection system with given image or video data.Face anti-spoofing task is challenging due to algorithm poor performance under domain shift between the source domain and unseen target domain.Current face anti-spoofing techniques mostly focus on treating the live detection task as a pure binary classification problem or introducing auxiliary tasks to supervise the training process of the classification task,while ignoring the modeling of domaininvariant feature space.To address this problem,this paper proposes a style augmentation based domain generalization algorithm.Based on the assumption that the spoofing face can be disentangled into domain features and live features,the algorithm introduces domain adversarial techniques to model domain-independent feature space.We also introduce a style augmentation module to generate multi-styled data.Besides,a domain label smoothing strategy is designed to further improve the stability of algorithm training.To obtain multi-scale spoofing cues from live features,we design a spoof cue exploration module to generate clearer decision boundaries using contrast learning techniques,and finally output robust detection results.Image-based face anti-spoofing techniques might ignore dynamic signals such as remote Photoplethysmography the face part leading to erroneous detection results.To exploit the temporal information in video data,this paper proposes a video-based face anti-spoofing algorithm.The algorithm first extracts multi-frame spatial information using an image encoder and feeds it to a spatial-temporal attention module to combine spatial-temporal information so that obtain more accurate detection results.In this paper,we design and implement rich comparison experiments and ablation studies to validate the effectiveness of the above proposed domain generalized image face anti-spoofing algorithm based on style augmentation and the video face antispoofing algorithm based on spatial-temporal attention.Tested in cross-domain scenarios,the image and video face anti-spoofing algorithms proposed in this paper have better performance compared to other face anti-spoofing methods.Based on the algorithm proposed in this paper,a prototype system supporting user interaction is built and extensive experiments are conducted to evaluate the performance of the algorithm and the system.The results show that the algorithm proposed in this paper has better performance compared to other face anti-spoofing algorithms and the system is robust.