摘要
近年来,人工智能生成内容(artificial intelligence generated content,AIGC)蓬勃发展,为文学创作、新闻报道、智能对话、音视频生成、艺术创作等应用带来了新的机遇,但同时也引发了伪造信息泛滥的隐患,对国家、社会和个人造成潜在威胁,针对这种技术滥用的检测与防范逐渐成为研究热点.本文首先介绍了AIGC伪造技术,从局部伪造和整体生成两个方面介绍了基于一般生成模型的内容生成技术.此外,本文根据生成模态种类的不同梳理归纳了目前基于大模型的整体内容生成技术.然后,从被动检测和主动防御两个维度出发,系统地归纳、分析并比较了当前主流的AIGC伪造取证方法.针对被动检测技术,从图像、视频、音频和文本角度介绍了单模态伪造检测方法,从基于视觉–音频信息和基于视觉–文本信息两个角度介绍了多模态伪造检测方法.针对主动防御技术,从主动干扰技术和主动水印技术两个类别详细介绍和探讨了伪造防御方法.最后,进一步讨论了AIGC伪造内容被动检测和主动防御面临的挑战和发展方向.
In recent years,artificial intelligence generated content(AIGC)has developed rapidly,offering new opportunities in fields such as literature generation,news report,intelligent dialogue,audio-video generation,and art.However,it has also led to concerns about the spread of fake information,posing potential threats to security,society,and individuals.As a result,detecting and preventing the misuse of such technology has become a pivotal research direction.This paper first introduces AIGC forgery techniques,including two perspectives:partial manipulation and complete generation.Furthermore,we review and compare current large-scale generative model-based generation approaches,categorized by the generated modality.Subsequently,we comprehensively examine AIGC forgery forensics techniques,including passive detection and proactive protection.For passive detection techniques,we present unimodal forgery detection methods for images,video,audio,and text.We also explore multimodal detection techniques,focusing on approaches utilizing visual-audio information and visualtextual information.For proactive protection techniques,we discuss forgery prevention methods,including active interference techniques and active watermarking techniques.Finally,we further discuss the current challenges and future research directions for both passive detection and proactive protection of AIGC-forged content.
作者
刘晓龙
刘欢
赵耀
倪蓉蓉
李晓龙
郭茂祖
Xiaolong LIU;Huan LIU;Yao ZHAO;Rongrong NI;Xiaolong LI;Maozu GUO(School of Intelligence Science and Technology,Beijing University of Civil Engineering and Architecture,Beijing 102616,China;Beijing Key Laboratory of Super Intelligent Technology for Urban Architecture,Beijing University of Civil Engineering and Architecture,Beijing 102616,China;School of Computer Science and Technology,Beijing Jiaotong University,Beijing 100044,China;Beijing Key Laboratory of Science Fiction Audio and Video Intelligent Processing,Beijing Jiaotong University,Beijing 100044,China)
出处
《中国科学:信息科学》
北大核心
2025年第9期2250-2288,共39页
Scientia Sinica(Informationis)
基金
国家自然科学基金(批准号:U24B20179,62120106009,62336001)
北京交通大学基本科研业务费(批准号:K25JBZY00030)
北京市自然科学基金(批准号:4232021)
山东省自然科学基金(批准号:ZR2022LZH011)资助项目。
关键词
深度伪造
视频换脸
AIGC
被动检测
伪造检测
主动防御
deepfake
video face swap
AIGC
passive detection
forgery detection
proactive protection
