Revealing Photoshop Inpainting Traces Under JPEG Compressions.

Area of Science:

• Digital Image Forensics
• Computer Vision
• Image Processing

Background:

• Photoshop inpainting creates subtle artifacts difficult to detect, especially after JPEG compression.
• Existing forensic methods face a trade-off between detecting inpainting and maintaining robustness to compression.
• JPEG compression degrades high-frequency details crucial for many forensic detectors.

Purpose of the Study:

• To develop a JPEG-resistant method for localizing Photoshop inpainting artifacts.
• To overcome the limitations of current forensic techniques that struggle with compressed images.
• To provide a robust solution for image tampering detection in real-world scenarios.

Main Methods:

• A multi-frequency representation framework using parameterized frequency-selective filters.
• Independent training of detectors for each frequency band, followed by fusion.
• Theoretical frequency-domain analysis to explain robustness to JPEG compression.
• Multi-quality JPEG augmentation during training to simulate compression variations.

Main Results:

• The proposed method consistently outperforms existing forgery localization techniques under various JPEG compression levels.
• Demonstrated effectiveness on both synthetic and real-world Photoshop inpainting datasets.
• Validated performance on images shared via social media platforms (WeChat, Weibo, Twitter).

Conclusions:

• The multi-frequency representation strategy offers a principled and effective approach to robust image forensic analysis.
• The method successfully balances discriminability of inpainting artifacts with robustness to JPEG compression.
• The proposed technique shows significant promise for practical applications in digital forensics and online media security.