Enhancing Face Forgery Detection in a Decentralized Landscape: A Federated Learning Approach with ResNet
Vinay Gautam, Himani Maheshwari, Raj Gaurang Tiwari, Ambuj Kumar Agarwal, Naresh Kumar Trivedi · 2023
Recent advancements in technologies could be the reason for fake image and video generation over the internet. This may be the cause of fake identity creation over the internet for forgery. These types of acts may be the reason for security issues in society. The legacy fake forgery method is not that capable of recognizing such forgery as the methods are trained with publicly available centralized datasets and never focus on privacy and security issues and adversely influence the forgery detection. Hence, the objective of this research is to provide decentralized ways to handle these issues effectively. The issue is taken care of with an effective federated learning-based deep learning technique. In the proposal, several deep learning models were used to generate a residual feature map from the available image dataset and later federated learning was used to generate a decentralized infrastructure for collaborative client machines. The complete experiment setup is established with a publicly available dataset and various variable parameters are used such as the number of clients and round of communication. Afterward, deep learning models’ performance is compared for forgery detection under federated learning environments, and it has been observed that ResNet deep learning outperforms with an accuracy rate of 87.83% in FaceForensic dataset.