摘要
随着深度伪造(Deepfake)技术的不断发展,犯罪分子可以利用造假图片伪造不在场证明,从而误导侦查方向以逃避法律责任。现有多数检测方法依赖于数据驱动,在跨压缩率、跨分辨率方面鲁棒性不强。研究Deepfake视频在脸部区域所遗留的伪影,建立一种基于Xception的双流网络检测模型,以实现对Deepfake图片的自动检测。利用Xception网络提取图片的全局空域特征,对脸部区域进行有效遮挡,凸显出脸部伪影并提取伪影特征。在此基础上,将空域特征与伪造特征2个支流的预测结果进行融合判别。在Deepfakes数据集上的实验结果表明,该模型的测试精度高达0.9864。
With the development of Deepfake technology,criminals may use fake pictures to forge alibi,misleading the direction of investigation to evade legal responsibility.Most existing detection methods are data-driven,and are not robust in terms of the cross-compression rate and cross-resolution.This article studies the artifacts in the face area in Deepfake videos,and on this basis designs a dual-stream network detection model based on Xception to realize automatic detection of Deepfake images.The model employs Xception network to extract the global spatial features of the picture.Then the face area is effectively occluded to highlight the artifacts of the face and extract the artifact features.Finally,the prediction results of the two tributaries with spatial features and forged features are fused and judged.The experimental results on the Deepfakes dataset show that the test accuracy of this method reaches 0.9864.
作者
耿鹏志
樊红兴
张翌阳
唐云祁
GENG Pengzhi;FAN Hongxing;ZHANG Yiyang;TANG Yunqi(School of Criminal Investigation,People’s Public Security University of China,Beijing 100038,China;Center for Research on Intelligent Perception and Computing,Institute of Automation,Chinese Academy of Sciences,Beijing 100190,China)
出处
《计算机工程》
CAS
CSCD
北大核心
2021年第12期156-162,共7页
Computer Engineering
基金
国家自然科学基金(61906199)
中央高校基本科研业务费专项基金(2021JKF203)
上海市现场物证重点实验室开放课题(2021XCWZK04)。