EUVIMEDEUVIMED entdecken
Crossref

DCNN-Based Deepfake Detection Using a Fused Model

S. Naga Raju, DarbhasayanamShreeya, Lade Prashanth, SamudralaPrashanthi, Aleti Kiran

Journal of Electronics and Power Engineering · 30.04.2026 · 3 (1) · 9-16

DOI: 10.69996/jepe.2026002 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Electronics and Power Engineering, online erschienen am 30.04.2026. Verfasst von S. Naga Raju und 4 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In this paper, we propose a new video-level DeepFake detection framework to address the challenge that counters the synthetic media which is becoming increasingly realistic through exploiting identity consistency and multiple-model fusion. Facial regions are detected and aligned on sampled video frames using MTCNN, then identity-preserving deep facial embeddings are obtained with a pretrained InceptionResnetV1 model. Rather than classifying independently at the frame-level, the approach temporally pools the embeddings to capture statistical regularities and identity-level coherence at the videolevel. Two complementary classifiers are employed: a lightweight neural discriminator on the embeddingbased statistical features, and an Interpretable Logistic Regression that considers the internal identity similarity within different frames. The outputs of these classifiers are then combined through a calibrated meta-classifier, providing more robustness and useful confidence information to the detection. The proposed framework is tested on a balanced partition of the DarbhasayanamShreeya Dept. of Computer Science & Engineering (AI&ML) Kakatiya Institute of Technology & Science Warangal, India SamudralaPrashanthi Dept. of Computer Science & Engineering (AI&ML) Kakatiya Institute of Technology & Science Warangal, India Celeb-DF dataset with a strict train/validation/test split and with the inclusion of comprehensive forensic metrics, i.e., AUC, EER and low false-positive-rate detection performance. Experimental results demonstrate that the proposed multi-stage fusion strategy and uncertainty-aware decision scheme can always achieve better performance than individual models.

Zitieren

APA

Raju, S. N., DarbhasayanamShreeya, Prashanth, L., SamudralaPrashanthi, & Kiran, A. (2026). DCNN-Based Deepfake Detection Using a Fused Model. Journal of Electronics and Power Engineering, 3(1), 9-16. https://doi.org/10.69996/jepe.2026002

Vancouver

Raju SN, DarbhasayanamShreeya, Prashanth L, SamudralaPrashanthi, Kiran A. DCNN-Based Deepfake Detection Using a Fused Model. Journal of Electronics and Power Engineering. 2026;3(1):9-16. doi:10.69996/jepe.2026002

BibTeX
@article{Raju2026,
  author = {Raju, S. Naga and DarbhasayanamShreeya and Prashanth, Lade and SamudralaPrashanthi and Kiran, Aleti},
  title = {DCNN-Based Deepfake Detection Using a Fused Model},
  journal = {Journal of Electronics and Power Engineering},
  year = {2026},
  volume = {3},
  number = {1},
  pages = {9--16},
  doi = {10.69996/jepe.2026002}
}
Kostenlos starten →