Scholarlink

<p>The rise of deepfakes and users’ susceptibility to online manipulation underscore the critical need for effective detection methods in scholarly research. Detecting multimodal deepfakes—particularly in inflammatory posts—poses unique challenges, as these combine multiple media types to increase believability and emotional impact. To address this, we propose the DEFUTE framework, an entropy-based method that assesses feature consistency across images and text through four modules: deepfake detection, image similarity, text similarity, and text-image matching. Using the DamCNN algorithm, DEFUTE identifies facial forgeries, while image and text similarity modules analyze key visual and semantic features. The text-image matching module verifies alignment between descriptions and images to spot discrepancies. DEFUTE demonstrates high accuracy in identifying deepfake content, with future efforts focused on enhancing generalization and integrating multimodal data for even greater precision.</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

Combating deepfakes: an entropy-aware framework for detecting forged visual and textual information

  • 期刊论文

  • 发表日期：

    2025年9月22日

查看全文
  • Hongchen Wu,
  • Zhaorong Jing,
  • Xiaochang Fang,
  • Hongxuan Li,
  • Meng Li,
  • Hongzhu Yu,
  • Bing Yu,
  • Huaxiang Zhang

摘要

The rise of deepfakes and users’ susceptibility to online manipulation underscore the critical need for effective detection methods in scholarly research. Detecting multimodal deepfakes—particularly in inflammatory posts—poses unique challenges, as these combine multiple media types to increase believability and emotional impact. To address this, we propose the DEFUTE framework, an entropy-based method that assesses feature consistency across images and text through four modules: deepfake detection, image similarity, text similarity, and text-image matching. Using the DamCNN algorithm, DEFUTE identifies facial forgeries, while image and text similarity modules analyze key visual and semantic features. The text-image matching module verifies alignment between descriptions and images to spot discrepancies. DEFUTE demonstrates high accuracy in identifying deepfake content, with future efforts focused on enhancing generalization and integrating multimodal data for even greater precision.