Answers are grounded in this paper and cite their sources.
Demonstrates robust zero-watermarking in digital media, indicating enhanced copyright protection.
In the era of massive digital content distribution, balancing the trade-off between watermark imperceptibility and robustness remains a critical challenge. This paper proposes a novel robust zero-watermarking framework that integrates deep learning with chaotic cryptography to secure digital images without altering their original visual quality. Unlike traditional methods that rely on single-domain features, our approach employs a hybrid feature extraction mechanism: it utilizes the VGG-19 deep convolutional neural network to capture high-level semantic representations invariant to geometric distortions, fused with Local Binary Patterns (LBP) to preserve local texture robustness against signal processing attacks. To address dimensional redundancy and feature loss, PCA-based variance-threshold selection is applied to the VGG-19 activations, retaining only components that explain at least 99% of the total variance. Furthermore, multi-scale LBP descriptors at three radii ( R = 1 , 2 , 3 ) are employed to enrich local texture representation. The two descriptors are integrated via an energy-based attention-weighted fusion strategy, where adaptive weights are derived from the ℓ 2 -energy of each normalised descriptor, ensuring that the more informative modality contributes proportionally to the final feature key. A core innovation of this work is the introduction of a novel Chaotic Ring-Shift Permutation algorithm driven by the Tent-Tent System (TTS). Unlike conventional linear scrambling, this algorithm executes dynamic, bi-directional circular shifts based on chaotic coordinates, effectively obliterating spatial correlations. The TTS is specifically selected over single-layer chaotic maps owing to its expanded full-spectrum chaotic range and statistically uniform output distribution, properties that have been demonstrated to be superior to classical low-dimensional maps . The final zero-watermark is constructed by logically embedding the encrypted copyright logo into the binarized hybrid features via an XOR operation. Experimental results demonstrate that the proposed scheme achieves superior robustness against rotation, scaling, and filtering attacks while ensuring an enormous key space derived from a precision of 1 0 − 15 , which renders brute-force attacks computationally infeasible. This makes the framework highly suitable for robust copyright protection and content authentication in unsecured networks.
No takes yet. Share an insight, caveat, or question.
Herbadji et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: