基于深度学习的多特征融合人脸鉴伪模型  

Multi-feature Fusion Face Authentication Model Based on Deep Learning

作　　者：李铮[1] 郑涛[2] 张小梅[2] Li Zheng;Zheng Tao;Zhang Xiaomei(China Information Technology Designing&Consulting Institute Co.,Ltd.,Beijing 100048,Chi-na;China United Network Communications Group Co.,Ltd.,Beijing 100033,China)

机构地区：[1]中讯邮电咨询设计院有限公司,北京100048 [2]中国联合网络通信集团有限公司,北京100033

出　　处：《邮电设计技术》2024年第8期58-61,共4页Designing Techniques of Posts and Telecommunications

摘　　要：人脸伪造给网络安全带来了重大挑战。针对现有人脸鉴伪模型特征单一、准确率低的问题,提出了一种基于深度学习的多特征融合人脸鉴伪模型。该模型设计了不同特征提取模块,用以获取不同尺度的特征表示。并学习如何有效融合这些语义信息以准确判定是否伪造,从而显著提升模型的准确率和鲁棒性。最后在公开数据集FaceForensics++上进行大量实验验证。实验结果显示,与现有方法相比,设计的模型有明显的性能提升。Face forgery poses a major challenge to network security.In response to the problem of existing face forgery models with single features and low accuracy,a multi-feature fusion face forgery model based on deep learning is proposed.The model designs different feature extraction modules to obtain feature representations at different scales.It also learns how to effectively fuse this semantic information to accurately determine whether it is forged,thereby significantly improving the accuracy and robustness of the model.Finally,a large number of experiments are carried out on the open data set FaceForensics++.The experimental results show that the designed model achieves significant performance improvement compared to existing methods.

关 键 词：人脸鉴伪 特征融合 深度学习 FaceForensics++数据集 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]