Related Experiment Videos
A hybrid attention-based spatio-temporal model for deepfake video detection
Sambhav Gupta1, Ojasvi Pandey1, Deepika Varshney1
1Department of Computer Science & Engineering and Information Technology, Jaypee Institute of Information Technology, Noida, India.
Scientific Reports
|July 8, 2026
Summary
This study introduces an Illusion Interception Tool to detect sophisticated deep fake videos. The novel hybrid system combines spatial and temporal analysis for improved accuracy in identifying manipulated media.
Area of Science:
- Computer Science
- Artificial Intelligence
- Digital Security
Background:
- Deep fake technology poses significant risks to information integrity, privacy, and digital security.
- Existing deep fake detection methods often lack effectiveness against advanced forgeries due to limited spatial or temporal analysis.
Purpose of the Study:
- To develop a robust deep fake detection system, the Illusion Interception Tool.
- To address the limitations of current methods by employing a hybrid approach for enhanced detection capabilities.
Main Methods:
- A hybrid framework integrating ResNeXt convolutional neural networks (CNNs) for spatial analysis.
- Long short-term memory (LSTM) networks with soft attention for temporal dynamics analysis.
- Combined analysis to detect both pixel-level artifacts and frame-wise inconsistencies.
Main Results:
- The proposed method achieved up to 94.8% accuracy on benchmark datasets like Face Forensics++ and Celeb-DF.
- Demonstrated significant improvements in accuracy and robustness compared to existing techniques.
- Ablation studies confirmed the effectiveness of the combined spatial and temporal analysis.
Conclusions:
- The Illusion Interception Tool offers a novel and effective approach to deep fake video detection.
- The hybrid framework shows strong generalization capabilities on unseen data.
- This system represents a significant advancement in combating the spread of manipulated media.