High-Quality Video Watermarking Based on Deep Neural Networks for Video with HEVC Compression

Area of Science:

• Digital Image Processing
• Video Compression
• Steganography

Background:

• High-Efficiency Video Coding (HEVC/H.265) is a widely used video compression standard.
• Protecting intellectual property in digital video content is crucial.
• Existing watermarking methods often suffer from reduced quality or data loss under compression.

Purpose of the Study:

• To develop a transparent, high-capacity video watermarking method resilient to H.265/HEVC compression.
• To utilize neural networks for robust watermark embedding and recovery.
• To maintain high video quality post-watermarking and compression.

Main Methods:

• Employing a deep neural network trained for watermark embedding within the HEVC compression channel.
• Utilizing the chrominance channel (YUV420p) for watermark insertion.
• Testing watermark recovery from HEVC-compressed video frames using a constant rate factor (CRF) up to 22.

Main Results:

• Achieved high accuracy in watermark embedding, with Peak Signal-to-Noise Ratio (PSNR) values exceeding 44 dB.
• Demonstrated a watermark capacity of 96 bits for a 128x128 image.
• Successfully recovered watermarks from single video frames compressed with HEVC.

Conclusions:

• The proposed neural network-based watermarking method is effective for H.265/HEVC compressed videos.
• The technique offers a robust solution for transparent, high-capacity video watermarking.
• Maintained high video fidelity and ensured reliable watermark extraction post-compression.