DEEPFAKE VIDEO DETECTION USING LONG-RANGE ATTENTION-BASED DEEP NEURAL NETWORKS

Abstract

The rapid advancement of generative models has led to the proliferation of DeepFake videos, which manipulate facial features and expressions to create hyper-realistic fake content. These videos pose serious threats to digital integrity, public trust, and cybersecurity. This research proposes a DeepFake detection framework leveraging long-range attention-based deep neural networks capable of modeling global dependencies across video frames. The architecture utilizes a hybrid convolutional and transformer-based pipeline that captures both local texture inconsistencies and long-distance temporal relationships. Experimental evaluations on benchmark datasets such as FaceForensics++ and DFDC demonstrate superior accuracy and robustness compared to traditional convolutional models. The results highlight the effectiveness of long-range attention mechanisms in identifying subtle spatiotemporal artifacts, offering a scalable and explainable solution for real-world DeepFake detection.