Research And Application Of Liveness Detection Algorithm In Face Recognitio

Advancements in technology have led to the widespread use of facial recognition technology.However,malicious individuals deceive facial recognition systems through methods like printing photos or recording videos.To address this security concern,developers emphasize the importance of liveness detection.Existing facial liveness detection algorithms are often complex or inaccurate,hindering their deployment on performance-limited embedded devices.This paper proposes the following contributions:(1)A lightweight liveness detection method based on image texture and multisupervision is introduced to achieve model compression.It employs lightweight convolutional modules,incorporating techniques such as texture feature extraction,multimodal fusion,multiple supervision,and auxiliary loss functions to enhance network performance.(2)A knowledge transfer-based liveness detection method is proposed,involving lightweight and auxiliary networks.It introduces soft labels and employs multiple networks to generate semantically rich depth image soft labels for knowledge transfer.The approach utilizes multi-network intermediate feature distillation,allowing the lightweight network to mimic the complex network’s feature representation.Experimental results show that these knowledge transfer mechanisms effectively enhance the accuracy and robustness of the lightweight network.(3)A facial recognition system based on embedded Linux and embedded processor RK3399 has been implemented,verifying the practicality of the algorithm proposed in this paper.