基于生成式对抗网络的多人脸图像局部伪造特征智能检测方法  

Intelligent detection method for local forged features in multi-facial images based on generative adversarial networks

作　　者：朱振刚 严海兵[1] 杨萌[1] ZHU Zhengang;YAN Haibing;YANG Meng(Library,SUST,Suzhou 215009,China)

机构地区：[1]苏州科技大学图书馆,江苏苏州215009

出　　处：《苏州科技大学学报(自然科学版)》2025年第1期82-88,共7页Journal of Suzhou University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金项目(62373266)。

摘　　要：多人脸图像相较于单人脸图像,其复杂性更高。攻击者通常仅针对图像的局部区域进行篡改,加大了检测难度。为此,本文提出了基于生成式对抗网络的多人脸图像局部伪造特征智能检测方法。结合多人脸图像的模糊性分布特征,进行边缘识别检测与模糊信息分簇。并建立超分辨识别模型,以获取面部阴影区域特征分值,实现多人脸图像特征分割。在此基础上,利用生成式对抗网络生成逼真的虚假数据以欺骗区分器,结合径向基函数对支持向量机分类模型进行伪造检测。研究表明,所提方法能精准检测出人脸图像局部伪造特征,适用于多角度人脸图像伪造检测。Multi-facial images are more complex than singlefacial images.Attackers usually only tamper with the local areas of the images,which increases the difficulty of detection.To this end,an intelligent detection method for local forged features of multi-facial images based on generative adversarial network is proposed.Referring to the fuzzy distribution characteristics of multi-facial images,we detected the edge recognition detection and clustered the fuzzy information.The super resolution recognition model was established to obtain the feature score of facial shadow area and realize the feature segmentation of multi-facial images.Based on this method,the generative adversarial network generated realistic spurious data to cheat the discriminator,and performed forgery detection on the SVM classification model in combination with radial basis functions.The experimental results show that the proposed method can accurately detect the local forged features of facial images,which is suitable for multi-angle face image forgery detection.

关 键 词：人脸图像伪造 生成式对抗网络 模糊信息法 边缘感知 支持向量机 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]