SOLUTION FOR DETECTION OF FACE-SWAP BASED FAKE VIDEOS

Recent advancements in deep learning have dramatically simplified the creation ofhighly realistic deepfake (DF) videos, which were traditionally the domain of skilled visual effectsprofessionals. This surge in AI-synthesized media poses challenges in sectors such as politics,security, and entertainment. Detecting deepfakes is increasingly difficult due to the realism of thesevideos. In this paper, we propose a deepfake detection approach that combines ConvolutionalNeural Networks (CNNs) for extracting detailed frame-level features with Recurrent NeuralNetworks (RNNs) for modeling temporal inconsistencies across video sequences. The CNNidentifies critical features such as facial landmarks and textures, while the RNN focuses on detectingunnatural transitions between frames, including irregular facial movements or subtle manipulations.Our method is capable of handling videos of varying lengths and has demonstrated competitiveperformance when tested on a comprehensive dataset of deepfake videos.