Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (23): 183-190.DOI: 10.3778/j.issn.1002-8331.2206-0270

• Graphics and Image Processing • Previous Articles     Next Articles

Face Liveness Detection Algorithm Based on Metric Learning-Cross Batch Storage Mechanism

CAI Tijian, LIU Wenxin, CHEN Fuchun, CHEN Jun, LUO Ciyong   

  1. School of Information Engineering, East China Jiaotong University, Nanchang 330013, China
  • Online:2023-12-01 Published:2023-12-01

引入跨批存储机制度量学习的人脸活体检测

蔡体健,刘文鑫,尘福春,陈均,罗词勇   

  1. 华东交通大学 信息工程学院,南昌 330013

Abstract: Aiming at the problems of embedding feature aliasing and poor generalization ability in face liveness detection,it uses the method of anomaly detection to learn a compact representation space of living samples and get finer liveness features through pixel-level assisted supervision. To get a clearer classification boundary, multi-scale tuple loss is introduced to optimize the model, and the method of combining in-batch and inter-batch sample mining is used to expand the scope of sample mining to obtain more effective sample pairs. Finally, the robustness and generalization of the algorithm are demonstrated by comparing experiments on open datasets OULU, Replay Attack, and CASIA.

Key words: face liveness detection, anomaly detection, triplet loss, sample mining strategy

摘要: 针对人脸活体检测中存在的嵌入特征混叠、泛化能力差的问题,采用异常检测的方法来学习活体样本的一个紧凑表示空间,并通过像素级的辅助监督来获得更细粒度的活性特征;为了获得更清晰的分类边界,引入多尺度三元组损失来优化模型,采用批内和批间样本挖掘相结合的方法,来扩大样本挖掘范围,以获得更多有效的样本对。通过在公开数据集OULU、Replay Attack和CASIA上的对比实验,证明了算法的鲁棒性和泛化能力。

关键词: 人脸活体检测, 异常检测, 三元组损失, 样本挖掘策略