Robust Detection and Localization of Image Copy-Move Forgery Using Multi-Feature Fusion

Area of Science:

• Digital Image Forensics
• Computer Vision
• Deep Learning

Background:

• Copy-move forgery detection (CMFD) is vital in image forensics.
• Existing deep learning models struggle with feature fusion and precise localization.
• Current methods often fail to fully utilize complementary RGB and noise domain features.

Purpose of the Study:

• To develop a robust method for detecting and localizing image copy-move forgery.
• To improve feature representation by fusing RGB and noise domain information.
• To enhance detection precision through advanced decoding and attention mechanisms.

Main Methods:

• A Multi-Feature Fusion Network (MFFNet) was designed, integrating RGB and noise domain features.
• A Lightweight Multi-layer Perceptron Decoder (LMPD) was developed for reconstruction and localization map generation.
• Cross-layer information aggregation and local/global attention mechanisms were employed for accurate prediction masks.

Main Results:

• The MFFNet model demonstrated superior feature representation through effective multi-feature fusion.
• The proposed decoder achieved more precise forgery localization compared to existing approaches.
• Experimental results confirmed enhanced robustness against JPEG compression, noise, and resizing.

Conclusions:

• The MFFNet offers a significant advancement in copy-move forgery detection and localization.
• The fusion of multi-domain features and advanced decoding improves detection accuracy and robustness.
• This approach provides a more reliable tool for image forensics analysis.