DeepFake Video Detection

Authors: Sakshi K, Rakshitha N, Thejashwini, Varshitha P

Abstract: The swift progress in deepfake creation methods has triggered significant worries about the genuineness and dependability of online video material. Manipulated facial appearances in deepfake videos and expressions present considerable dangers in areas like social media, journalism, and digital forensics.While current deep learning-based detection techniques have shown encouraging outcomes, numerous models demonstrate restricted generalization when utilized on unfamiliar datasets or live video feeds. This study introduces a deepfake video detection system that incorporates ResNeXt for spatial characteristics.extraction using an Attention-Driven Bidirectional Long Short-Term Memory (Bi- LSTM) model for temporal examination. ResNeXt adeptly extracts distinguishing facial characteristics from separate frames, whereas the attention-boosted Bi-LSTM selectively emphasizes significant temporal segments throughout video sequences.This integrated structure enhances the understanding of both spatial discrepancies and temporal dynamics.dependencies related to deepfake modification. Experimental findings indicate that the suggested method demonstrates excellent results on benchmark datasets. In spite of its effectiveness, issues connected to dataset reliance and immediate implementation persist, which are examined alongside possible future research pathways