Scholarlink

<p>DeepFake technology transforms digital media, blurring the gap between real and manipulation. With an increase in its misuse, the necessity for efficient detection has grown to new heights. Despite comprehensive research, a systematic comparison of DeepFake detection models is lacking, complicating the task for researchers to identify the most effective methods. This survey addresses the gap by systematically classifying detection models into CNN, GAN-based, transformer-based, and hybrid approaches, providing a clear comparison of their strengths and limitations. Also, it examines the advancement of these models, evaluating significant advancements that have influenced their development over time, as reflected in several deep learning models, which achieves an AUC of 99.80% on the CelebDF dataset, and an accuracy of 97.82% on the FaceForensics++ dataset. It also provides insights into commonly used datasets, evaluation measures, and a systematic analysis of detection methodologies. This survey integrates and organizes significant research contributions, facilitating access to essential information and enhancing the researcher’s knowledge of current methodologies and their relative efficacy. This creates a systematic foundation for future studies, directing efforts toward more effective DeepFake detection systems.</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 detection in images and videos: a survey on models, datasets, and evaluation metrics

  • 期刊论文

  • 发表日期：

    2026年2月27日

查看全文
  • Nuzhat Afzal Bhat,
  • Kaisar Javeed Giri

摘要

DeepFake technology transforms digital media, blurring the gap between real and manipulation. With an increase in its misuse, the necessity for efficient detection has grown to new heights. Despite comprehensive research, a systematic comparison of DeepFake detection models is lacking, complicating the task for researchers to identify the most effective methods. This survey addresses the gap by systematically classifying detection models into CNN, GAN-based, transformer-based, and hybrid approaches, providing a clear comparison of their strengths and limitations. Also, it examines the advancement of these models, evaluating significant advancements that have influenced their development over time, as reflected in several deep learning models, which achieves an AUC of 99.80% on the CelebDF dataset, and an accuracy of 97.82% on the FaceForensics++ dataset. It also provides insights into commonly used datasets, evaluation measures, and a systematic analysis of detection methodologies. This survey integrates and organizes significant research contributions, facilitating access to essential information and enhancing the researcher’s knowledge of current methodologies and their relative efficacy. This creates a systematic foundation for future studies, directing efforts toward more effective DeepFake detection systems.