Paper Key : IRJ************568
Author: Gopika E S,Amith A S,Aparna A,Alwin Varghese,Athira Bose
Date Published: 05 Nov 2024
Abstract
The rise in computational power has made deep learning algorithms so advanced that creating highly realistic deepfake videos has become alarmingly easy. These face-swapped deepfakes can lead to political unrest, fake terrorism, revenge porn, and blackmail. In this work, we introduce a deep learning-based method to effectively distinguish AI-generated fake videos from real ones. Our system leverages a ResNeXt Convolutional Neural Network to extract frame-level features, which are then used to train an LSTM-based Recurrent Neural Network. This approach classifies videos as either deepfakes or real. To ensure real-time performance, we evaluate our method on a diverse, balanced dataset composed of FaceForensics++, the Deepfake Detection Challenge, and Celeb-DF, demonstrating how our system can achieve competitive results with a straightforward and robust approach.