Deepfake Detection Using Keyframe Extraction, Global Feature Enhancement, and Temporal Analysis

Deepfake Detection Using Keyframe Extraction, Global Feature Enhancement, and Temporal Analysis

Parth Jagtap1, Yash Bhandare2, Atharva Nalawade3, Shrenik Kolhe4, Shubhangi Gaikwad5

1,2,3,4 Student, Department of Computer Engineering, Jayawantrao Sawant College of Engineering, Pune

5 Guide, Department of Computer Engineering, Jayawantrao Sawant College of Engineering, Pune

---------------------------------------------------------------------***---------------------------------------------------------------------

Abstract

This paper presents a novel deepfake detection approach that combines spatial and temporal analysis, leveraging keyframe extraction, global feature enhancement, and a dual network system integrating ResNeXt-50 and LSTM. This hybrid approach aims to detect AI-synthesized media by enhancing subtle artifacts in manipulated frames. Initial observations indi cate that spatial enhancements improve artifact detection while the LSTM effectively identifies temporal inconsistencies. This methodology, outlined with expected performance outcomes, contributes a robust solution for deepfake detection that is adaptable to future real-time applications.

Key Words:  Deepfake Detection, Global feature Enhancement, ResNeXt-50, LSTM, Temporal Analysis, Keyframe Extraction, GAN, AI-Synthesized Media