High-Quality Video Watermarking Based on Deep Neural Networks and Adjustable Subsquares Properties Algorithm.

Area of Science:

• Computer Vision
• Digital Signal Processing
• Multimedia Security

Background:

• Digital video watermarking is crucial for copyright protection and content authentication.
• Existing methods often struggle to balance watermark capacity, transparency, and robustness against compression.
• The H.265/HEVC codec presents unique challenges for watermark embedding due to its advanced compression techniques.

Purpose of the Study:

• To develop a high-capacity and transparent video watermarking technique for H.265/HEVC compressed videos.
• To ensure the watermark is imperceptible to the human eye while maintaining video quality.
• To achieve robust watermark survival after HEVC compression and enable accurate recovery.

Main Methods:

• Utilizing deep neural networks for watermark data encoding.
• Implementing an adjustable subsquares properties algorithm for watermark embedding.
• Employing the H.265/HEVC codec for video compression.
• Evaluating watermark performance using Peak Signal-to-Noise Ratio (PSNR) and Bit Error Rate (BER).

Main Results:

• Achieved PSNR values of 40 dB or higher, indicating high video quality preservation.
• Demonstrated complete watermark recovery (BER=0) from single frames compressed up to CRF 16.
• The adjustable subsquares algorithm allows for variable watermark capacity and resistance.

Conclusions:

• The proposed deep learning-based watermarking method offers a robust and transparent solution for H.265/HEVC videos.
• The technique effectively preserves original video accuracy while ensuring watermark security and resilience.
• The adjustable properties provide flexibility for different application requirements.