Deep Learning-Based Forgery Attack on Document Images.

Area of Science:

• Computer Vision
• Digital Forensics
• Machine Learning

Background:

• Digital document images are prevalent in online services.
• Deep learning enables end-to-end text editing in images.
• Existing methods struggle with complex characters and backgrounds.

Purpose of the Study:

• To develop a low-cost document forgery algorithm for practical image editing.
• To address limitations of current text editing algorithms in complex scenarios.
• To create realistic forged documents that evade detection.

Main Methods:

• Disentangling textual and background information in source images.
• Utilizing text skeleton as auxiliary information for complex structures.
• Incorporating texture continuity in the loss function.
• Applying post-processing to mitigate forgery traces from print-and-scan distortions.

Main Results:

• Reduced Mean Squared Error (MSE) by approximately 2/3.
• Improved Peak Signal-to-Noise Ratio (PSNR) by 4 dB.
• Enhanced Structural Similarity Index Measure (SSIM) by 0.21.
• Visually superior reconstruction quality for complex characters and textures.
• Successful evasion of existing document authentication systems.

Conclusions:

• The proposed method offers a significant advancement in document forgery.
• It effectively handles complex document elements and realistic distortions.
• The algorithm demonstrates practical applicability in identity document alteration.
• Forged documents can successfully deceive current authentication systems.