.png){kind=icon}
.png){kind=icon}
Return
Frequency-driven deep learning network for image splicing forgery detection
.png){kind=icon}
PRE
AI
DOI：10.1016/j.knosys.2025.114365.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Image splicing is a widely used technique for manipulating images in various social activities. Detecting splicing forgery is crucial in digital forensics to identify malicious image manipulation and protect information security. However, existing methods for detecting splicing forgery typically learn features in the spatial domain and struggle to effectively capture subtle features indicative of forgery, resulting in insufficient image splicing forgery detection accuracy. To address this challenge, we propose a novel deep-learning network named the frequency-driven deep-learning network (FreNet). Specifically, FreNet comprises three innovative modules: the frequency learnable module (FLM), the spatial-aware frequency learning module (SFLM), and the high-level feature-enhancement module (HFEM). The FLM effectively extracts high- and low-frequency features, thus enhancing frequency-domain representation and capturing subtle tampered features in splicing forgery images. The SFLM utilizes spatial information to guide frequency feature learning, thus enabling spatial-aware frequency feature learning. The HFEM enhances rich contextual and high-level semantic information through multilevel and multipath extraction and fusion. Extensive experiments on five benchmark datasets indicate that FreNet can achieve superior performance. Additionally, robustness experiments demonstrate the superior robustness of FreNet against various common attacks.
Journal
K
IF:
7.6
Papers:
1.2W
Citations:
4.5W