LIGHTWEIGHT AND ROBUST CNN-BASED WATERMARK DETECTION: A COMPARATIVE STUDY OF CNN, RESNET-50, AND EFFICIENTNET-B0 ARCHITECTURES
Problems of Information Society · 10.02.2026 · 17 (1) · 44-49
DOI: 10.25045/jpis.v17.i1.05 ↗Kurz gefasst
Fachartikel in der Zeitschrift Problems of Information Society, online erschienen am 10.02.2026. Verfasst von Abdurahman Vagifli.
Abstract (Originalsprache)
Digital watermarking has emerged as a critical technology for safeguarding intellectual property rights, particularly in an era where digital content generation is accelerating exponentially. In this study, we present a comprehensive evaluation and comparative analysis of three prominent deep learning-based models—namely a simple Convolutional Neural Network (CNN), ResNet-50, and EfficientNet-B0—for the task of invisible watermark detection. These models are tested using images embedded with a binary logo watermark through a CNN-based encoder-decoder system. Our experimentation leverages 2,000 labeled images from the MS COCO dataset, evenly split between clean and watermarked classes. We conduct thorough evaluations based on several key metrics, including detection accuracy, precision, recall, F1-score, area under the ROC curve (AUC), model storage size, and inference latency. Our results show that all models demonstrate strong detection capabilities, with ResNet-50 and EfficientNet-B0 reaching near-perfect performance. Notably, EfficientNet-B0 offers an optimal balance of performance and efficiency, making it ideal for real-time watermark verification in practical deployment scenarios.
Zitieren
APA
Vagifli, A. (2026). LIGHTWEIGHT AND ROBUST CNN-BASED WATERMARK DETECTION: A COMPARATIVE STUDY OF CNN, RESNET-50, AND EFFICIENTNET-B0 ARCHITECTURES. Problems of Information Society, 17(1), 44-49. https://doi.org/10.25045/jpis.v17.i1.05
Vancouver
Vagifli A. LIGHTWEIGHT AND ROBUST CNN-BASED WATERMARK DETECTION: A COMPARATIVE STUDY OF CNN, RESNET-50, AND EFFICIENTNET-B0 ARCHITECTURES. Problems of Information Society. 2026;17(1):44-49. doi:10.25045/jpis.v17.i1.05
BibTeX
@article{Vagifli2026,
author = {Vagifli, Abdurahman},
title = {LIGHTWEIGHT AND ROBUST CNN-BASED WATERMARK DETECTION: A COMPARATIVE STUDY OF CNN, RESNET-50, AND EFFICIENTNET-B0 ARCHITECTURES},
journal = {Problems of Information Society},
year = {2026},
volume = {17},
number = {1},
pages = {44--49},
doi = {10.25045/jpis.v17.i1.05}
}
EUVIMED