燕山大学学报2025，Vol.49Issue(4)：339-348,10.DOI:10.3969/j.issn.1007-791X.2025.04.007
频域辅助监督双流网络的人脸呈现攻击检测
Face presentation attack detection based on frequency auxiliary supervision two-stream network
摘要
Abstract
Face presentation attack detection is crucial for ensuring the security of face recognition system.Aiming at the problem of poor generalization ability of the model due to the sensitivity of extracted image features to light exposure in limited data scenarios with only visible light modal,a two-stream face presentation attack detection method based on multi-level frequency domain auxiliary supervision is proposed,which adopts a multi-level feature fusion strategy using the features of low frequency,medium frequency and high frequency components to make full use of the information of different frequencies.In order to ensure the effective transmission and utilization of high-frequency information,a residual fusion attention in level of frequency block is proposed.Collaborative learning of binary supervised stream and frequency supervised stream is driven by hierarchical cross attention.Experimental results on four public datasets show that the proposed method achieves an average HTER value of 14.01%and an AUC value of 92.50%in cross-dataset testing experiments.Compared to seven existing literature methods,the proposed method obtains the best cross domain generalization performance,which has advantages in learning generalized features of data with distribution differences.关键词
人脸呈现攻击检测/频域信息/双流网络/特征融合Key words
face presentation attack detection/frequency domain information/two-stream network/feature fusion分类
信息技术与安全科学引用本文复制引用
封筠,李景涛,贺晶晶,高宇豪..频域辅助监督双流网络的人脸呈现攻击检测[J].燕山大学学报,2025,49(4):339-348,10.基金项目
国家自然科学基金资助项目(61972267) （61972267）
石家庄铁道大学研究生创新资助项目(YC202449) （YC202449）
