CNN-Guided Dual-Chaotic Encryption and Wavelet Domain Embedding for Robust and Adaptive Image Watermarking
International Journal of Electrical and Electronics Research · 30.03.2026 · 14 (1) · 176
DOI: 10.37391/ijeer.140119 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Electrical and Electronics Research, online erschienen am 30.03.2026. Verfasst von Prof. Ajit Singh und 1 weiteren Autor:innen. Die Arbeit nennt 15 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
With the rise of digital content creation and sharing, protecting multimedia assets from illegal use, decoding, and piracy, without a doubt, is becoming more important as time passes. As such, this work proposes a novel digital image watermarking framework that combines Convolutional Neural Networks (CNNS), Discrete Wavelet Transform (DWT), and a dual chaotic encryption technique based on Logistic Map and Tent Map fusion. An adaptive encryption and watermark embedding in the frequency domain of the host image ensures imperceptibility, robustness, and security. Feature statistics are extracted from a normalized watermark using a two-layer CNN, dynamically creating initial conditions for chaotic maps. The high-entropy resultant mask is fused with the generated sequences and used to encrypt the watermark with modular arithmetic. The encrypted watermark is embedded in the LL sub-band of a DWT-transformed host image using alpha blending. The final watermarked image is then reconstructed using inverse DWT. CNN features are used to regenerate identical chaotic sequences to decrypt and retrieve the watermark during extraction. In order to validate the approach, it was tested on images from the COCO (Common Objects in Context) and ImageNet datasets. Initially, average PSNR (Peak Signal-to-Noise Ratio) values were larger than 41 dB, and SSIM (Structural Similarity Index) values were over 0.97, thus having good visual fidelity. The system was found to be resilient to typical attacks (Gaussian noise, cropping, JPEG compression, rotation), with PSNR values between 30.1dB and 37.8 dB and SSIM over 0.94. As such, the PSNR of “Sports Car” remained 37.74 dB in the case of noise and 37.81 dB under cropping for the noise “Laptop”. COCO images like “Dog” and “Person” demonstrated PSNR above 35 dB for most distortions. Chaos parameters were adaptively generated from CNN features using a dual chaotic map fusion, enhancing security and embedding them into the DWT domain to improve robustness. This integrated approach establishes a secure intelligent watermarking framework suitable for real-world applications such as copyright protection, secure transmission of medical images, and forensics.
Zitieren
APA
Singh, P. A., & Rajni (2026). CNN-Guided Dual-Chaotic Encryption and Wavelet Domain Embedding for Robust and Adaptive Image Watermarking. International Journal of Electrical and Electronics Research, 14(1), 176. https://doi.org/10.37391/ijeer.140119
Vancouver
Singh PA, Rajni. CNN-Guided Dual-Chaotic Encryption and Wavelet Domain Embedding for Robust and Adaptive Image Watermarking. International Journal of Electrical and Electronics Research. 2026;14(1):176. doi:10.37391/ijeer.140119
BibTeX
@article{Singh2026,
author = {Singh, Prof. Ajit and Rajni},
title = {CNN-Guided Dual-Chaotic Encryption and Wavelet Domain Embedding for Robust and Adaptive Image Watermarking},
journal = {International Journal of Electrical and Electronics Research},
year = {2026},
volume = {14},
number = {1},
pages = {176},
doi = {10.37391/ijeer.140119}
}
EUVIMED