Deepfake Video Detection and Classification Through Dynamic Spatio- Temporal Inconsistency Analysis

The escalating threat of deepfake videos in the digital age necessitates the development of advanced detection methods to safeguard against the proliferation of manipulated visual content. Recent research has suggested using convolutional neural network is used as an effective tool to detect deepfakes in the networks. However, most techniques fail to capture the inter frame dissimilarities of the collected media streams. We proposed an ensemble approach for deepfake detection by combining power of the convolutional neural networks and the recurrent neural networks, specifically RegNet and LSTMs. This research introduces a novel approach employing a hybrid model that combines RegNet(Regularized Networks) and LSTM (Long Short-Term Memory networks) for robust deepfake video detection. Focusing on subtle yet critical facial cues patterns. Our methodology surpasses a commendable accuracy threshold of 90%. The integration of RegNet, known for its lightweight and high-performance architecture, with LSTM networks enhances the model’s temporal understanding capabilities. To ensure practical applicability, videos with frames less than 100 are excluded from consideration. This research not only presents a technically sound and efficient deepfake detection model but also contributes to the broader mission of securing digital content integrity in an era of evolving media manipulation techniques.