RIFoL: A robust image forgery localization network for noisy images.

Area of Science:

• Computer Vision
• Digital Image Forensics

Background:

• Image forgery localization is crucial for digital evidence integrity.
• Noise, such as additive Gaussian white noise, significantly degrades the performance of existing forgery localization techniques, limiting their practical application.

Purpose of the Study:

• To propose a robust image forgery localization network resilient to noise attacks.
• To enhance the detection and localization accuracy of forged images subjected to various noise types.

Main Methods:

• Introduced RIFoL, a novel network comprising a restoration module and a forgery localization module.
• The restoration module recovers forged traces from noisy images.
• The forgery localization module incorporates a Spatial Feature Enhancement Module (SFEM) and a Multi-Attention Feature Fusion Module (MAFM) for enhanced feature extraction.

Main Results:

• RIFoL demonstrates superior performance in detecting and localizing forged images attacked with noise.
• The proposed SFEM utilizes multi-scale attention, integrating channel and spatial information.
• The MAFM effectively fuses features across different scales via spatial and channel pathways.

Conclusions:

• The proposed RIFoL network significantly improves the robustness of image forgery localization against noise.
• The novel enhancement mechanisms within RIFoL provide state-of-the-art performance in identifying manipulated images under noisy conditions.