Digital Hologram Watermarking Based on Multiple Deep Neural Networks Training Reconstruction and Attack

Area of Science:

• Computer Vision
• Digital Image Processing
• Cryptography

Background:

• Digital holograms are susceptible to unauthorized access and manipulation.
• Existing watermarking techniques may lack robustness against various attacks.
• Deep learning offers potential for advanced image security solutions.

Purpose of the Study:

• To propose a novel deep neural network (DNN) based method for embedding and extracting watermarks in digital holograms.
• To enhance the robustness and invisibility of watermarking through integrated attack simulation and holographic reconstruction within the DNN.
• To develop a re-training strategy for improving robustness against specific attacks.

Main Methods:

• A three-sub-network architecture for digital hologram watermarking.
• Integration of attack simulation and holographic reconstruction within the DNN for simultaneous training of invisibility and robustness.
• A proposed network training methodology utilizing hologram and reconstruction data.
• Iterative re-training based on robustness analysis against various attacks.

Main Results:

• The proposed DNN effectively embeds and extracts watermarks from digital holograms.
• The integrated attack simulation enhances watermark robustness against diverse threats.
• Quantitative evaluation demonstrates reliable performance against multiple attack scenarios.
• The re-training strategy successfully improves robustness based on attack analysis.

Conclusions:

• The developed DNN-based watermarking technique offers a reliable solution for securing digital holograms.
• Simultaneous training of invisibility and robustness within the network architecture is effective.
• The proposed method provides a robust and adaptable approach to digital hologram watermarking.