Paper Key : IRJ************029
Author: Nigade Jaydeep Popatrao,Shinde Nikhil Kailas,Avatade Rohan Ramdas,Bagal Yashkumar Ramchandra,Prof. Rutuja Taware
Date Published: 10 Nov 2024
Abstract
The proliferation of AI-driven deepfake technology has resulted in a growing need for advanced detection mechanisms capable of identifying realistic digital media manipulations. This paper introduces a novel hybrid detection model combining Convolutional Neural Networks (CNN) for spatial feature extraction and Long Short-Term Memory (LSTM) units within a Recurrent Neural Network (RNN) for temporal consistency analysis. Utilizing publicly available datasets, such as Face Forensics++ and the Deepfake Detection Challenge (DFDC) dataset, our system demonstrates superior accuracy in detecting manipulated content across a variety of resolutions and manipulation techniques. With an overall accuracy of 92%, this approach addresses current detection limitations and offers insights for future research in audio-visual deepfake detection. Keywords: Deepfake detection, Hybrid detection model, Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), Temporal consistency analysis, FaceForensics++ dataset.
DOI Requested