Robust zero-watermarking for color images using hybrid deep learning models and encryption.

Area of Science:

• Computer Science
• Image Processing
• Cryptography

Background:

• Digital watermarking is crucial for copyright protection, but traditional methods lack robustness against attacks.
• Existing zero-watermarking techniques often fail to balance security, imperceptibility, and resilience to distortions.

Purpose of the Study:

• To develop a robust zero-watermarking technique for color images that overcomes the limitations of existing methods.
• To enhance copyright protection for digital media through a secure and distortion-resilient approach.

Main Methods:

• Combines Local Binary Patterns (LBP) with deep features from VGG19's CONV5-4 layer.
• Employs Discrete Wavelet Transform (DWT) and Discrete Cosine Transform (DCT) for enhanced feature representation.
• Integrates a chaotic encryption scheme (Lorenz system, Logistic map) for scrambling and an XOR operation for watermark generation.

Main Results:

• Achieves high resilience against scaling, noise, filtering, compression, and rotation attacks.
• Demonstrates a low Bit Error Rate (BER < 0.0022) and high Normalized Cross-Correlation (NCC > 0.9959).
• Maintains optimal Peak Signal-to-Noise Ratio (PSNR > 34.0692 dB) for attacked images, ensuring visual quality.

Conclusions:

• The proposed zero-watermarking method offers superior robustness and security compared to existing algorithms.
• Effectively protects multimedia copyright with imperceptible and secure ownership verification.
• Presents a promising solution for reliable digital copyright protection in color images.