Deep Learning based DeepFake Image Detection for Enhancing Social Media Security

Abstract

The rapid growth of deep learning technologies and social media has led to a massive amount of deepfake content. The deepfake media is often utilized for spreading misinformation, identity theft, falsification of data, spoofing, and hate speech. Studies indicate that over 96% of deepfake videos online are used for malicious purposes, with detection becoming increasingly critical as generation techniques evolve. Automatic deepfake detection is challenging because advances in content creation make generated content indistinguishable from the original. This paper presents deepfake detection using a deep convolutional neural network and a Long Short-Term Memory (LSTM) network. The DCNN captures spatial correlations and local connectivity in images, extracting facial features and micro-expressions across multiple hierarchical layers. In contrast, the LSTM is used to capture temporal dependencies and long-term correlations in deepfake images, particularly to analyze frame-to-frame inconsistencies and unnatural motion patterns that characterize synthetic media. The DCNN-LSTM achieves an improved accuracy of 95.5% compared with DCNN (95.1%) and LSTM (84.3%) for AVCeleb dataset.