Image Fake News Detection using Efficient NetBO Model

Abstract

Today, social networks have become a prominent source of news, significantly altering the way people obtain news from traditional media sources to social media. Alternatively, social media platforms have been plagued by unauthenticated and fake news in recent years. However, the rise of fake news on these platforms has become a challenging issue. Fake news dissemination, especially through visual content, poses a significant threat as people tend to share information in image format. Consequently, detecting and combating fake news has become crucial in the realm of social media. In this paper, we propose an approach to address the detection of fake image news. Our method incorporates the error level analysis (ELA) technique and the explicit convolutional neural network of the EfficientNet model. By converting the original image into an ELA image, it is possible to effectively highlight any manipulations or discrepancies within the image. The ELA image is further processed by the EfficientNet model, which captures distinctive features used to detect fake image news. Visual features extracted from the model are passed through a dense layer and a sigmoid function to predict the image type. To evaluate the efficacy of the proposed method, we conducted experiments using the CASIA 2.0 dataset, a widely adopted benchmark dataset for fake image detection. The experimental results demonstrate an accuracy rate of 96.11% for the CASIA dataset. The results outperform in terms of accuracy and computational efficiency, with a 6% increase in accuracy and a 5.2% improvement in the F-score compared with other similar methods.