Image Forgery Detection Using Federated Learning On Edge Devices
Sara M. Saadeldein, Ramez M. Elmasry, Omar M. Fahmy, Mohammed A.‐M. Salem · 2023
Social media has become a big part of our lives, people often come across fake content like fake news and edited images. Nowadays, it's easy to edit images with high accuracy, making it hard to tell what's real and what's fake. From filters on Instagram and Snapchat to professional-level Photoshop editing, it's causing problems between individuals, cooperates, and even nations. Dealing with this issue is important to maintain trust and authenticity in the digital world. In this research, a user-friendly mobile app interface is implemented that allows users to upload images for analysis. These input images underwent JPEG recompression, and the difference between the original and recompressed images is calculated. This difference is then processed by a federated forgery classification CNN network. The analysis results were presented to the users through the mobile app. The proposed approach resulted in 87% training accuracy per client, 76% global training accuracy, and 83% global testing accuracy.