Unified frequency and graph learning framework for accurate and robust copy-move image forgery detection

Area of Science:

• Digital image forensics
• Computer vision
• Image processing

Background:

• Copy-move forgery (CMF) poses a significant challenge in digital image forensics.
• Conventional CMF detection methods face issues like high computational cost, poor localization, and low efficacy for minor forgeries.

Purpose of the Study:

• Introduce a novel Dual-Phase Forgery Identification Network (DPFIN) to enhance CMF detection efficacy.
• Improve resilience against complex forgeries and mitigate limitations of existing approaches.

Main Methods:

• A dual-phase preprocessing mechanism combining frequency domain analysis and adaptive visual enhancement.
• Utilizing a YOLOv5-based backbone for multi-scale feature extraction.
• Incorporating the Zebra Optimization Algorithm for stable training and generalization.

Main Results:

• Achieved 99.89% accuracy and 98.89% F1-score on the CoMoFoD dataset.
• Obtained 99.99% AUC and 96.8% F1-score on the CASIA v2 dataset.
• Demonstrated a low false positive rate of 0.21%.

Conclusions:

• The proposed DPFIN is a highly effective and reliable solution for detecting copy-move forgeries.
• DPFIN shows superior performance compared to conventional methods, especially for complex and minor forgeries.