EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Improved Convergence in CNN-Based Video Forgery Detection through Chaotic Sinusoidal Zebra Optimization Algorithm

Michael O. Adio, John B. Oladosu, Alice O. Oke, Olufikayo A. Adedapo

FUOYE Journal of Engineering and Technology · 2025 · 10 (3) · 470-476

DOI: 10.4314/fuoyejet.v10i3.12 ↗

Kurz gefasst

Fachartikel in der Zeitschrift FUOYE Journal of Engineering and Technology, 2025. Verfasst von Michael O. Adio und 3 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Video forgery detection (VFD) has become a critical task in ensuring the authenticity of digital media, yet the effectiveness of Convolutional Neural Networks (CNNs) for this purpose is highly dependent on appropriate hyperparameter selection. Suboptimal configurations often lead to overfitting, poor convergence, and reduced generalization, thereby highlighting the necessity of robust hyperparameter optimization strategies. In this study, three models are evaluated: a baseline CNN, a CNN optimized using the Zebra Optimization Algorithm (ZOA), and a CNN further enhanced with the Chaotic Sinusoidal Zebra Optimization Algorithm (CSZOA). Experiments were conducted on a total of 270 videos (135 forged and 135 original), were sourced from Kaggle.com and categorized into five groups, each containing three forgery types: deletion, duplication, and insertion., assessing critical hyperparameters such as learning rate, number of filters, filter size, batch size, and objective function values. ZOA achieved its best performance at iteration 27 with a learning rate of 0.00597, 64 filters, a filter size of 3, and a batch size of 128, yielding an objective function value of 0.0146. CSZOA, integrating chaotic sinusoidal dynamics to enhance exploration and mitigate premature convergence, reached its optimal configuration at iteration 27 with a learning rate of 0.00705, 64 filters, a filter size of 3, and a batch size of 32, achieving a lower objective function value of 0.0129. Comparative analysis showed that CSZOA outperformed both the baseline CNN and ZOA-optimized CNN, delivering improvements of 99.51 % in detection accuracy, 99.38 % in precision, sensitivity of 99.31 % and 0.32 False Positive rate. These results confirm that chaotic sinusoidal enhancement not only improves convergence stability but also boosts detection reliability in VFD. Beyond video forgery detection, the developed CSZOA-based hyperparameter optimization framework has promising applicability to other computer vision tasks that require improved generalization and resilience against complex data variations.

Zitieren

APA

Adio, M. O., Oladosu, J. B., Oke, A. O., & Adedapo, O. A. (2025). Improved Convergence in CNN-Based Video Forgery Detection through Chaotic Sinusoidal Zebra Optimization Algorithm. FUOYE Journal of Engineering and Technology, 10(3), 470-476. https://doi.org/10.4314/fuoyejet.v10i3.12

Vancouver

Adio MO, Oladosu JB, Oke AO, Adedapo OA. Improved Convergence in CNN-Based Video Forgery Detection through Chaotic Sinusoidal Zebra Optimization Algorithm. FUOYE Journal of Engineering and Technology. 2025;10(3):470-476. doi:10.4314/fuoyejet.v10i3.12

BibTeX
@article{Adio2025,
  author = {Adio, Michael O. and Oladosu, John B. and Oke, Alice O. and Adedapo, Olufikayo A.},
  title = {Improved Convergence in CNN-Based Video Forgery Detection through Chaotic Sinusoidal Zebra Optimization Algorithm},
  journal = {FUOYE Journal of Engineering and Technology},
  year = {2025},
  volume = {10},
  number = {3},
  pages = {470--476},
  doi = {10.4314/fuoyejet.v10i3.12}
}
Kostenlos starten →