Scholarlink

<p>Face tampering, involving the modification of facial images using advanced software and deepfake technology, poses a significant security threat. Traditional face manipulation detection methods often struggle with limited frequency information, hindering their effectiveness. To address this, we propose a novel face tampering detection framework that combines multi-level discrete wavelet transform (DWT) and vision transformer. By extracting multi-level frequency features using DWT and capturing global inconsistencies across local face areas through the vision transformer, our approach reveals discriminative details and frequency information. We evaluate our method on public datasets, achieving state-of-the-art accuracy of 99.86% and 99.92% on FaceForensics++ and Celeb-DF, respectively. These results demonstrate the efficacy of our framework in detecting deepfake face images, paving the way for more robust facial manipulation detection systems. The source code of our proposed algorithm and the related datasets are available at <a href="https://github.com/shawn-zhangxiang/MLFFE_ViT">https://github.com/shawn-zhangxiang/MLFFE_ViT</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deepfake face detection via multi-level discrete wavelet transform and vision transformer

  • 期刊论文

  • 发表日期：

    2025年1月23日

查看全文
  • Main Uddin,
  • Zhangjie Fu,
  • Xiang Zhang

摘要

Face tampering, involving the modification of facial images using advanced software and deepfake technology, poses a significant security threat. Traditional face manipulation detection methods often struggle with limited frequency information, hindering their effectiveness. To address this, we propose a novel face tampering detection framework that combines multi-level discrete wavelet transform (DWT) and vision transformer. By extracting multi-level frequency features using DWT and capturing global inconsistencies across local face areas through the vision transformer, our approach reveals discriminative details and frequency information. We evaluate our method on public datasets, achieving state-of-the-art accuracy of 99.86% and 99.92% on FaceForensics++ and Celeb-DF, respectively. These results demonstrate the efficacy of our framework in detecting deepfake face images, paving the way for more robust facial manipulation detection systems. The source code of our proposed algorithm and the related datasets are available at https://github.com/shawn-zhangxiang/MLFFE_ViT.