Dual-Branch Noise-Guided Network for Image Splicing Forgery Detection

Image splicing forgery is a frequent manipulation technique in digital image forensics, presenting a significant challenge for image integrity verification. Existing detection methods often fail to fully exploit noise features, particularly the valuable noise information that signals tampered regions within an image. To address this issue, we introduce a novel model for image splicing forgery detection called dual-branch noise-guided network (NoGNet). We design a noise information enhancement (NIE) module to capture subtle noise differences, thereby enhancing the representation of noise features that are critical for identifying tampered areas. Furthermore, we propose a multi-view feature fusion (MFF) module that integrates features from multiple perspectives, allowing the model to capture richer, more comprehensive feature representations. Extensive evaluations on four datasets demonstrate that NoGNet achieves state-of-the-art performance in splicing forgery detection. Robustness experiments confirm that NoGNet demonstrates effectiveness in detecting image splicing forgery under various common attacks.