首页|期刊导航|四川轻化工大学学报(自然科学版)|一种改进CFL-Net的图像篡改检测算法

一种改进CFL-Net的图像篡改检测算法OA

An Improved Image Tampering Detection Algorithm Based on CFL-Net

中文摘要英文摘要

随着信息技术的快速发展,图像篡改问题在各领域中逐渐凸显,使得图像篡改检测技术对维护信息真实性愈发重要.针对当前图像篡改检测算法普遍存在的特征融合不充分、对篡改区域关键特征捕获能力不足的问题,提出一种改进CFL-Net的图像篡改检测算法.具体改进包括:在原有的ResNet-50特征提取网络中嵌入挤压-激励(SE)模块,通过通道权重自适应调整强化模型对篡改特征的关注;引入注意力特征融合(AFF)模块,对RGB分支特征与噪声分支特征进行动态融合,实现双分支特征的有效互补;同时加入边缘损失函数,引导模型重点关注篡改区域的边缘细节信息.实验结果表明,与U-Net、PSCC-Net、MVSS-Net、CFL-Net 4个模型相比,改进的算法在5种广泛使用的数据集上总体表现优异,具有较好的检测和定位能力.

With the rapid advancement of information technology,the issue of image tampering has become increasingly prominent across various fields,making image tampering detection technology increasingly vital for safeguarding the authenticity of information.To address the current issues in image tampering detection algorithms,such as insufficient feature fusion and limited ability to capture key features of tampered regions,an improved image tampering detection algorithm based on CFL-Net is proposed.An Squeeze-and-Excitation(SE)module is added to the original ResNet-50 feature extraction network to enhance the model's focus on tampering features.Additionally,an Attentional Feature Fusion(AFF)module is introduced to fuse features from the RGB and noise branches,enabling effective complementary fusion of the dual-branch features.An edge loss function is also incorporated to help the model focus on the edge information of tampered regions.Results show that,compared to U-Net,PSCC-Net,MVSS-Net,and CFL-Net,the improved algorithm performs well across five widely used datasets,demonstrating superior detection and localization capabilities.

刘胜波;何小利

四川轻化工大学计算机科学与工程学院,四川 宜宾 644000四川轻化工大学计算机科学与工程学院,四川 宜宾 644000

信息技术与安全科学

图像篡改检测注意力机制特征融合特征提取

image tampering detectionattention mechanismfeature fusionfeature extraction

《四川轻化工大学学报(自然科学版)》 2026 (1)

54-61,8

2024教育部第三期供需对接就业育人项目(2023122021680)视频图像篡改检测算法开发项目(E10204710)自贡市重点科技计划项目(自贡市医学大数据与人工智能研究院)(2022ZCYGY16)企业信息化与物联网测控技术四川省高校重点实验室项目(2022WYY02)四川轻化工大学教学改革研究项目(JG-24025)

10.11863/j.suse.2026.01.06

评论