报告详情
  1. 活动首页
  2. 报告清单
  3. 报告详情
HFEP-Net:Hierarchical Forensic Evidence Propagation Network for Generalizable Video Deepfake Detection
编号：174 访问权限：仅限参会人 更新：2026-11-10 17:02:24 浏览：6次 Online

报告开始：2026年11月20日 15:00（Asia/Kolkata）

报告时间：15min

所在会场：[S4] Artificial Intelligence and Intelligent Computing [S4-3] Artificial Intelligence and Intelligent Computing

暂无文件

摘要
The increasing realism of generative artificial intelligence has made video deepfake detection increasingly difficult under unseen manipulations, compression artifacts, partial facial alterations, and cross-domain distribution shifts. Existing methods largely rely on global binary classification or isolated spatial, frequency, and temporal representations, which can suppress subtle forensic evidence and limit generalization. To address these challenges, we propose a Hierarchical Forensic Evidence
Propagation Network (HFEP-Net) for generalizable video deepfake detection. HFEP-Net jointly learns object-, frame-, and video-level forensic representations and introduces a Forensic Evidence Representation (FER) to characterize anomaly scores, confidence, and uncertainty across semantic levels. An Evidence
Propagation Graph (EPG) hierarchically propagates reliable evidence, while Reliability-Guided Evidence Aggregation (RGEA) suppresses unreliable evidence caused by compression and do main shifts. A Cross-Level Evidence Consistency Loss (CECL)  further enforces consistency among object-, frame-, and video level predictions. Experiments on FaceForensics++, Celeb-DF  v2, DFDC, and DeeperForensics-1.0 demonstrate that HFEP-Net achieves 90.17% accuracy, 90.15% F1-score, 96.25% ROC-AUC, and 9.80% EER. These results demonstrate the effectiveness of reliability-guided hierarchical forensic evidence propagation for robust and generalizable video deepfake detection.
关键词
Deepfake Detection, Computer Vision, Video Forensics, CNN, Multi-Level Supervision, Frequency Analysis.
报告人
Saritha K
Assistant Professor Shah and Anchor Kutchhi Engineering College

稿件作者
Saritha K Shah and Anchor Kutchhi Engineering College
Amiya Kumar Tripathy Don Bosco Institute of Technology
发表评论
验证码 看不清楚，更换一张
全部评论
重要日期

  • 会议日期

    11月20日

    2026

    至

    11月22日

    2026

  • 10月31日 2026

    初稿截稿日期

  • 11月07日 2026

    提前注册日期

  • 11月22日 2026

    注册截止日期

主办单位
United Societies of Science
承办单位
Pondicherry University
协办单位
IEEE Section
IEEE Madras Section
联系方式
登录查看完整联系方式
移动端
在手机上打开
小程序
打开微信小程序
客服
扫码或点此咨询