A Facial Authentication-based Deepfake Detection Machine Learning Model

Abstract

In an era dominated by digital media, the escalating menace of media distortion, particularly propelled by the advancement of deepfake technology, has emerged as a critical concern spanning the realms of virtual landscapes and reality. The rise of deepfake technology has posed significant challenges to the authenticity of visual content in today's digital world. This study proposes a novel approach to deepfake detection using pixel analysis. By closely examining the pixel characteristics and patterns within manipulated images and videos, we developed an algorithm that can distinguish between real and fake content with high accuracy. Our algorithm combines two state-of-the-art deep learning models, Resnext and Long-Short Term Memory (LSTM), in a supervised machine learning framework. To enhance the performance of our algorithm, we applied standard pixel normalization during the preprocessing phase. Our proposed method achieved an impressive accuracy score of 95.6% on a public dataset of deepfake images and videos. This result demonstrates the efficacy of pixel analysis in detecting deepfakes. This research contributes significantly to countering the increasing threat of deepfake media manipulation, safeguarding the authenticity of visual content in today's digital world.