Scholarlink

In recent months, the proliferation of deep learning-based software tools has significantly facilitated the creation of highly realistic face-swapped videos, commonly referred to as “DeepFake” (DF) videos. While digital video manipulation has been practiced for decades through traditional visual effects, recent advancements in deep learning have drastically enhanced the realism of synthetic content and its accessibility. These AI-generated media, popularly known as DeepFakes, pose a considerable challenge in terms of detection, as distinguishing between authentic and manipulated content requires sophisticated techniques. In this study, we propose a DeepFake detection framework leveraging Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs). The system employs a CNN to extract spatial features at the frame level, which are then utilized by an RNN to capture temporal inconsistencies across frames. By analyzing the sequential dependencies introduced by DeepFake generation techniques, our approach effectively identifies manipulated videos. We evaluate our model on a large dataset of fake videos, demonstrating its competitive performance in DeepFake detection while maintaining a relatively simple yet efficient architecture.

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

Hybrid CNN-RNN Deepfake Detection Framework Using ResNeXt-LSTM with Temporal and Spatial Artifact Analysis

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Rashmi Kumari,
  • Albin George,
  • Vaibhav Gupta,
  • Shivam Garg,
  • Subhranil Das,
  • Raghwendra Kishore Singh

摘要

In recent months, the proliferation of deep learning-based software tools has significantly facilitated the creation of highly realistic face-swapped videos, commonly referred to as “DeepFake” (DF) videos. While digital video manipulation has been practiced for decades through traditional visual effects, recent advancements in deep learning have drastically enhanced the realism of synthetic content and its accessibility. These AI-generated media, popularly known as DeepFakes, pose a considerable challenge in terms of detection, as distinguishing between authentic and manipulated content requires sophisticated techniques. In this study, we propose a DeepFake detection framework leveraging Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs). The system employs a CNN to extract spatial features at the frame level, which are then utilized by an RNN to capture temporal inconsistencies across frames. By analyzing the sequential dependencies introduced by DeepFake generation techniques, our approach effectively identifies manipulated videos. We evaluate our model on a large dataset of fake videos, demonstrating its competitive performance in DeepFake detection while maintaining a relatively simple yet efficient architecture.