Scholarlink

<p>Now a days, Deepfake video is one of the major threats in social media platform. Deepfake videos of celebrities, politicians, and ordinary people spread false information, including of a devastating nature. Not only does it spoil their fame and sometimes leads to suicide. In order to save their good fame and life, developing detection methods is necessary. In the recent development of technology on communication systems, for reducing storage space, they are sharing compressed videos. Due to these advancements, video forensic departments find it more difficult to search for the originality of the videos. It cannot be stopped spreading of the information, but one can provide a solution to find the originality of the videos. To address this problem, a Robust Deepfake Video Detection (RDVD) method is proposed for compressive deepfake video detection using deep learning models such as Gated Recurrent Units (GRU) and Stochastic Gradient Descent (SGD), respectively. Additionally, we improve the detection by utilizing the Dempster-Shafer theory of fusion on outputs from the above two models. The FaceForensics++ dataset is utilized for experimentation. The dataset contains five kinds of Deepfake videos(Deepfake, Face2Face, FaceSwap, Neural Textures, FaceShifter) and original version of the Deepfake videos. The proposed model utilizes this dataset for model construction. The performance of the model is validated using various performance metrics. Our proposed method achieves promising detection results, with accuracies of 83.75% for Deepfake, 80.25% for Face2Face, 82.50% for FaceSwap, 81.66% for Neural Textures, and 79.75% for FaceShifter. In addition to accuracy, other evaluation metrics such as precision, recall, and F1-score were also measured, with the method consistently delivering better results compared to existing state-of-the-art techniques.</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

Detection of Deepfake Videos in Social Networks

  • 期刊论文

  • 发表日期：

    2025年9月19日

查看全文
  • T Bhuvaneswari,
  • R Chandra Guru Sekar,
  • K Shiva Raama Krishna,
  • S R Nishaal,
  • M Arun Kumar

摘要

Now a days, Deepfake video is one of the major threats in social media platform. Deepfake videos of celebrities, politicians, and ordinary people spread false information, including of a devastating nature. Not only does it spoil their fame and sometimes leads to suicide. In order to save their good fame and life, developing detection methods is necessary. In the recent development of technology on communication systems, for reducing storage space, they are sharing compressed videos. Due to these advancements, video forensic departments find it more difficult to search for the originality of the videos. It cannot be stopped spreading of the information, but one can provide a solution to find the originality of the videos. To address this problem, a Robust Deepfake Video Detection (RDVD) method is proposed for compressive deepfake video detection using deep learning models such as Gated Recurrent Units (GRU) and Stochastic Gradient Descent (SGD), respectively. Additionally, we improve the detection by utilizing the Dempster-Shafer theory of fusion on outputs from the above two models. The FaceForensics++ dataset is utilized for experimentation. The dataset contains five kinds of Deepfake videos(Deepfake, Face2Face, FaceSwap, Neural Textures, FaceShifter) and original version of the Deepfake videos. The proposed model utilizes this dataset for model construction. The performance of the model is validated using various performance metrics. Our proposed method achieves promising detection results, with accuracies of 83.75% for Deepfake, 80.25% for Face2Face, 82.50% for FaceSwap, 81.66% for Neural Textures, and 79.75% for FaceShifter. In addition to accuracy, other evaluation metrics such as precision, recall, and F1-score were also measured, with the method consistently delivering better results compared to existing state-of-the-art techniques.