Deepfake Image Detection using Vision Transformer Models

TY - CONF

T1 - Deepfake Image Detection using Vision Transformer Models

AU - Ghita, Bogdan

AU - Kuzminykh, Ievgeniia

AU - Usama, Abubakar

AU - Bakhshi, Taimur

AU - Marchang, Jims

PY - 2024/6/1

Y1 - 2024/6/1

N2 - Deepfake images are causing an increasing negative impact on the day to day life and pose significant challenges for the society. There are various categories of deepfake images as the technology evolves and becomes more accessible. In parallel, deepfake detection methods are also improving, from basic features analysis to pairwise analysis and deep learning; nevertheless, to date, there is no consistent method able to fully detect such images. This study aims to provide an overview of existing methods of deepfake detection in the literature and investigate the accuracy of models based on Vision Transformer (VIT) when analysing and detecting deepfake images. We implement a VIT model-based deepfake detection technique, which is trained and tasted on a mixed real and deepfake images dataset from Kaggle, containing 40000 images.The results show that The VIT model scores relatively high, 89.9125%, which demonstrates its potential but also highlights there is significant room for improvement. Preliminary tests also highlight the importance of a large dataset for training and the fast convergence of the model. When compared with other deepfake machine learning and deep learning detection methods, the performance of the ViT model is in line with prior research and warrants further investigation in order to evaluate its full potential.

AB - Deepfake images are causing an increasing negative impact on the day to day life and pose significant challenges for the society. There are various categories of deepfake images as the technology evolves and becomes more accessible. In parallel, deepfake detection methods are also improving, from basic features analysis to pairwise analysis and deep learning; nevertheless, to date, there is no consistent method able to fully detect such images. This study aims to provide an overview of existing methods of deepfake detection in the literature and investigate the accuracy of models based on Vision Transformer (VIT) when analysing and detecting deepfake images. We implement a VIT model-based deepfake detection technique, which is trained and tasted on a mixed real and deepfake images dataset from Kaggle, containing 40000 images.The results show that The VIT model scores relatively high, 89.9125%, which demonstrates its potential but also highlights there is significant room for improvement. Preliminary tests also highlight the importance of a large dataset for training and the fast convergence of the model. When compared with other deepfake machine learning and deep learning detection methods, the performance of the ViT model is in line with prior research and warrants further investigation in order to evaluate its full potential.

KW - deepfake images

KW - deepfake detection

KW - Vision Transformer model

M3 - Paper

T2 - IEEE International Black Sea Conference on Communications and Networking

Y2 - 24 June 2024 through 27 June 2024

ER -