A Method for Deepfake Detection Using Convolutional Neural Networks

Abstract—

This paper proposes a method of countering spoofing attacks by improving the resilience of face-based biometric authentication systems to digital face manipulation attacks on the biometric input module. The proposed method of digital face manipulation detection (deepfake detection) is based on a convolutional neural network trained on a large dataset containing various types of manipulations, images of different quality, and a large number of identities and as a result achieves an accuracy of at least 99%. Experiment results also indicate high performance of the proposed approach compared to other available methods tested on the same dataset. The method can be used to improve the security of biometric authentication systems by reducing the risk of unauthorized access.