基于ConvNeXt的伪造人脸检测方法  

Fake Face Detection Method Based on ConvNeXt

作　　者：何德芬 江倩 金鑫[1,2] 冯明[1,2] 苗圣法[1,2] 易华松 He Defen;Jiang Qian;Jin Xin;Feng Ming;Miao Shengfa;Yi Huasong(School of Software,Yunnan University,Kunming 650504;Engineering Research Center of Cyberspace,Yunnan University,Kunming 650504)

机构地区：[1]云南大学软件学院,昆明650504 [2]跨境网络空间安全教育部工程研究中心(云南大学),昆明650504

出　　处：《信息安全研究》2025年第3期231-240,共10页Journal of Information Security Research

基　　金：国家自然科学基金项目(62101481,62261060);云南省基础研究计划项目(202401AT070470,202301AW070007,202201AU070033,202201AT070112,202301AU070210);云南省科技厅重大科技专项(202202AD080002);云南省迟学斌专家工作站项目(202305AF150078)。

摘　　要：由深度生成模型生成的虚假图像越发逼真,这些图像已经超越了人眼的识别能力.这种模型已成为编造谎言、制造舆论等非法活动的新工具.虽然当前研究者已经提出了很多检测方法检测伪造图像,但泛化能力普遍不高,因此,提出了一种基于ConvNeXt的伪造人脸检测方法.首先在ConvNeXt的第2个和第3个下采样模块后添加极化自注意(polarization self-attention,PSA)模块,使网络具有空间注意力和通道注意力的性能.其次在ConvNeXt的尾部设计一个信息富余模块(rich imformation block,RIB),以丰富网络学习到的信息,通过该模块对信息进行处理后再进行最终的分类.此外,网络训练使用的损失函数是交叉熵损失与KL(Kullback-Leibler)散度的结合.在当前主流的伪造人脸数据集上作了大量的实验,实验结果表明该方法在FF++高质量数据集上无论是准确率还是泛化性都超过所有对比方法.The fake images generated by deep generative models are becoming increasingly realistic,surpassing the human eye’s ability to detect them.These models have become new tools for illegal activities,such as fabricating lies and creating public opinion.Although current researchers have proposed many detection methods to detect fake images,their generalization ability is typically limited.To address this issue,we proposed a fake face detection method based on ConvNeXt.Firstly,we add a PSA(polarization self-attention)module after the second and third downsampling modules of ConvNeXt,enhancing the network’s spatial and channel attention performance.Secondly,a RIB(rich imformation block)is designed at the end of ConvNeXt to enrich the information learned by the network.The information is processed through this module before final classification.Furthermore,the loss function used in network training is a combination of CrossEntropy loss and KL(Kullback-Leibler)divergence.Extensive experiments on the current mainstream fake face datasets demonstrate that our method surpasses all comparative methods in accuracy and generalization on the FF++C23dataset.

关 键 词：神经网络 深度学习 伪造人脸 特征提取 伪造图像检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]