An Efficient Deepfake Detection System Using ConvoReinAutoNet and GeoFisherNet

This research proposes an efficient deepfake detection system using a hybrid optimization model and a new deep learning approach. This system is divided into two phases: (i) the training Phase and (ii) the detection Phase. The decision phase is the ultimate decision maker, wherein a new deep learning approach referred to as ConvoReinAutoNet(CRAN) is introduced by levering the layers of Convolutional Neural Networks (CNN), Deep Reinforcement Learning (DRL), and Autoencoders, respectively. The training phase is enriched with new feature fusion and a hybrid optimization-based optimal feature selection approach. The extracted temporal and texture features (newly introduced Improved Local Ternary Patterns (I-LTP)) from the pre-processed images of the deepfake database are fused using the new GeoFisherNet. The newest hybrid optimization method called Marine Predator Customized White Shark Optimizer (MCWO) is used to select the best features among the combined features, which represents the combination of both the Marine Predator Algorithm (MPA) and White Shark Optimization Algorithm (WSO). The suggested model has been implemented in python and validated in terms of detection efficiency over the existing approaches.