Scholarlink

In today’s digital world, where what we see isn’t always what it seems, the rise of Deepfake technology presents a big challenge to trust in videos. These manipulated videos, created using advanced AI, it helps in distinguishing between real and fake and makes it easier to spread false information. This study addresses the urgent need for reliable Deepfake detection methods using deep learning (DL) techniques within the Deepfake Detection Challenge (DFDC) dataset. Leveraging the vast and diverse DFDC dataset, this research develops robust detection algorithms capable of discerning between authentic and manipulated media content. Recently, researchers focused on Deepfakes detections, but accuracy is very low due to the content availability, rapid growth in digital content creation techniques. We proposed enhanced deep learning approach uses the fused ResNet50 and LSTM models, for more accurate detection of Deepfakes. When we evaluated the model over the DFDC dataset, it attains better F1-score, accuracy, and Logloss values compared to the existing models. The proposed model is intended to reduce the risks posed by Deepfake, including the spread of disinformation and the erosion of reputation and security. The study therefore helps in combating the negative impact of Deepfakes through the application of the DL models toward protecting trust, credibility and integrity in the digital space.

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

Enhanced Deepfake Detection Using Deep Learning on Large-Scale Video Data: A Fused ResNet50 and LSTM Approach

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Naveen Ananda Kumar Joseph Annaiah,
  • B. Omkar Lakshmi Jagan

摘要

In today’s digital world, where what we see isn’t always what it seems, the rise of Deepfake technology presents a big challenge to trust in videos. These manipulated videos, created using advanced AI, it helps in distinguishing between real and fake and makes it easier to spread false information. This study addresses the urgent need for reliable Deepfake detection methods using deep learning (DL) techniques within the Deepfake Detection Challenge (DFDC) dataset. Leveraging the vast and diverse DFDC dataset, this research develops robust detection algorithms capable of discerning between authentic and manipulated media content. Recently, researchers focused on Deepfakes detections, but accuracy is very low due to the content availability, rapid growth in digital content creation techniques. We proposed enhanced deep learning approach uses the fused ResNet50 and LSTM models, for more accurate detection of Deepfakes. When we evaluated the model over the DFDC dataset, it attains better F1-score, accuracy, and Logloss values compared to the existing models. The proposed model is intended to reduce the risks posed by Deepfake, including the spread of disinformation and the erosion of reputation and security. The study therefore helps in combating the negative impact of Deepfakes through the application of the DL models toward protecting trust, credibility and integrity in the digital space.