EUVIMEDAtklāt EUVIMED
Open access · CC BY 4.0

BMNet: Enhancing Deepfake Detection Through BiLSTM and Multi-Head Self-Attention Mechanism

Demao Xiong, Zhan Wen, Cheng Zhang, Dehao Ren, Wenzao Li

IEEE Access · 2025 · 13 · 21547-21556

DOI: 10.1109/access.2025.3533653 ↗

At a glance

Journal article in IEEE Access, 2025. Written by Demao Xiong and 4 co-authors. Cited 35 times so far (Crossref); 35 references. The full text is freely available (open access, CC BY 4.0). Funded by: Sichuan Province Science and Technology Department, Sichuan Province Major Science and Technology Project, Sichuan Province Philosophy and Social Science Research Project.

AbstraktsBibliogrāfiskie datiCiteRelated studies

Abstract (original language)

When forgery techniques can generate highly realistic videos, traditional convolutional neural network (CNN)-based detection models often struggle to capture subtle forgery features and temporal dependencies. Most existing models focus on feature extraction from static frames, neglecting the temporal correlation in videos, which decreases accuracy in detecting dynamic forged videos. Furthermore, the ability to detect localized forgery features remains insufficient. To address these limitations, we propose a deep forgery detection framework named BiLSTM Multi-Head Self-Attention Network (BMNet). By leveraging the Bi-Directional Long Short-Term Memory Network (BiLSTM) for modeling temporal dependencies between video frames, and the Multi-Head Self-Attention Mechanism (MHSA) for capturing features from different regions of an image, BMNet more effectively identifies dynamic and local forgery features. In our experiments, we extract features from 68 facial landmarks of each video frame and evaluate detection performance on four datasets: UADFV, FF++, Celeb-DF and DFDC. The results show significant improvements over traditional methods. We further validate the necessity of each network component through ablation studies, demonstrating that BMNet achieves accuracies of 95.54%, 92.18%, 80.20% and 84.72% on the FF++, UADFV, Celeb-DF and DFDC datasets, respectively, indicating its superior performance in deep forgery detection.

Funding

  • Sichuan Province Science and Technology Department
  • Sichuan Province Major Science and Technology Project
  • Sichuan Province Philosophy and Social Science Research Project
  • Undergraduate Education and Teaching Research and Reform Project of Chengdu University of Information Technology
  • Industry-School Cooperative Education Project of Ministry of Education

Cite

APA

Xiong, D., Wen, Z., Zhang, C., Ren, D., & Li, W. (2025). BMNet: Enhancing Deepfake Detection Through BiLSTM and Multi-Head Self-Attention Mechanism. IEEE Access, 13, 21547-21556. https://doi.org/10.1109/access.2025.3533653

Vancouver

Xiong D, Wen Z, Zhang C, Ren D, Li W. BMNet: Enhancing Deepfake Detection Through BiLSTM and Multi-Head Self-Attention Mechanism. IEEE Access. 2025;13:21547-21556. doi:10.1109/access.2025.3533653

BibTeX
@article{Xiong2025,
  author = {Xiong, Demao and Wen, Zhan and Zhang, Cheng and Ren, Dehao and Li, Wenzao},
  title = {BMNet: Enhancing Deepfake Detection Through BiLSTM and Multi-Head Self-Attention Mechanism},
  journal = {IEEE Access},
  year = {2025},
  volume = {13},
  pages = {21547--21556},
  doi = {10.1109/access.2025.3533653}
}
Start for free →