| 摘 要: 针对现有复制-移动伪造检测（CMFD）方法在大尺度篡改区域检测中对高级语义信息利用不足且计算效率偏低的问题，该研究提出一种融合全局上下文空间注意力（GCSA）与轻量化上采样模块的CMFD网络。该方法在基于浅层特征设计的BCM-Net基础上，在保留BCM-Net核心自相关计算模块的基础上引入GCSA模块以增强全局语义感知能力，有效捕捉大尺度篡改区域的高层关联信息；同时，设计轻量化的深度可分离转置卷积模块替代传统上采样结构，并引入联合损失函数与分类优化模块（CLF），在保持高精度的前提下显著降低计算开销并提升分割质量。在两个公开基准数据集上的实验结果表明，所提网络在检测精度和运行效率上均优于当前主流算法，尤其在大尺度篡改场景下展现出更强的鲁棒性与泛化能力。 |
| 关键词: 图像篡改检测 BCM-Net 空间注意力 轻量化 大尺度篡改 |
|
中图分类号:
文献标识码:
|
|
| Image Copy-Move Forgery Detection Based on an Attention-Optimized Lightweight Network Model |
|
Cai Zhen, Zhang Lin
|
Fujian University of Technology
|
| Abstract: To address the issues of insufficient utilization of high-level semantic information and low computational efficiency in existing copy-move forgery detection (CMFD) methods, particularly when detecting large-scale tampered regions, this study proposes a CMFD network that integrates a Global Context Spatial Attention (GCSA) module and a lightweight upsampling module. Built upon the BCM-Net, which is designed using shallow features, the proposed method retains the core self-correlation (SCORR) module of BCM-Net while introducing the GCSA module to enhance global semantic perception, effectively capturing high-level correlations within large-scale forged regions. Additionally, a lightweight depthwise separable transposed convolution module is designed to replace the traditional upsampling structure, and a joint loss function along with a Classification Fine-tuning (CLF) module is introduced to significantly reduce computational overhead while maintaining high precision and improving segmentation quality. Experimental results on two public benchmark datasets demonstrate that the proposed network outperforms current mainstream algorithms in both detection accuracy and operational efficiency, exhibiting stronger robustness and generalization ability, especially in large-scale tampering scenarios. |
| Keywords: image forgery detection BCM-Net spatial attention lightweight large-scale tampering |
