摘要
人工智能(Artificial Intelligence,AI)生成图像技术发展迅猛,高逼真内容对网络安全与社会信任构成重大威胁,而人类自主鉴别准确率仅约59%,接近随机猜测水平.现有检测方法普遍存在性能有限、跨模型泛化能力不足等问题,尤其无法有效捕捉生成图像中物理光照的不一致性.为此,本文提出融合光影敏感特征及Kolmogorov-Arnold(K-A)表示定理的特征融合鉴别方法(Light-enhanced Kolmogorov-Arnold Networks,L-KAN).在红绿蓝三原色(Red、Green、Blue,RGB)语义特征、频域特征和边缘特征的基础上,构建光影敏感特征.该特征通过整体光照分布、阴影面积及方向和多尺度光照梯度特性,捕捉生成图像中的光照异常.引入K-A表示定理进行特征融合,通过内外层函数协同作用,在保证特征互补性的同时有效抑制特征冗余.在3组公开数据集上,与9种先进方法进行对比,所提方法平均分类准确率均有显著提升.
The rapid advancement of artificial intelligence(AI)-generated image technologies poses significant threats to cybersecurity and public trust,as human visual detection accuracy remains as low as 59%,close to random guessing.Existing detection methods suffer from limited performance and poor generalization across generative models,particularly struggling to capture physical inconsistencies in illumination.To address this gap,we propose L-KAN(Light-enhanced Kolmogorov-Arnold Networks),a novel detection framework that integrates illumination-sensitive features with the Kolmogorov-Arnold(K-A)representation theorem.Building upon red-green-blue(RGB)semantics,frequency-domain cues,and edge information,we construct physically grounded features that encode global illumination distribution,shadow geometry,and multi-scale illumination gradients to expose lighting inconsistencies in synthetic images.Leveraging the K-A theorem for feature fusion,ours method synergizes inner and outer functions to enhance feature complementarity while suppressing redundancy.Experimental results on three public datasets demonstrate that L-KAN achieves a competitive performance compared with the state of the art methods.
作者
邓巧
姜林
刘乐新
唐吕鑫
杨英丽
DENG Qiao;JIANG Lin;LIU Le-xin;TANG Lü-xin;YANG Ying-li(School of AI and Advanced Computing,Hunan University of Technology and Business,Changsha,Hunan 410000,China;Xiangjiang Laboratory,Changsha,Hunan 410000,China)
出处
《电子学报》
北大核心
2025年第11期4077-4090,共14页
Acta Electronica Sinica
基金
湖南省教育厅科学研究项目(重点)(No.22A0441)
湘江实验室重大项目(No.23XJ01003,No.23XJ01009)。
