Deepfake Video Detection

Deepfake Video Detection

1Harsh Wardhan, 2Megha Kashyap, 3Shrusti Deshmukh, 4Ankit Bhardwaj, ⁵Swapnil Chaudhari

Department of Computer Science and Engineering, MIT School of Computing, MIT Art, Design and Technology University, Loni Kalbhor, Pune, India

1 harshu.13raj@gmail.com ,2kashyapmegha243@gmail.com,3 deshmukhshrusti32@gmail.com,

4ankitbharadwaj9097@gmail.com ,⁵ swapnil.chaudhari@mituniversity.edu.in

Abstract— Deepfake videos have become a serious problem, as they use deep learning to create highly realistic fake content. While some are made for entertainment, others can spread misinformation, harm reputations, or invade privacy. Advanced deepfake techniques make it difficult to identify these videos just by looking at them, which is why better detection methods are needed. Our system combines Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), specifically Long Short-Term Memory (LSTM) networks, to detect deepfakes. CNNs are used to extract important details from individual video frames, while LSTMs analyze the sequence of frames to find patterns over time. This combination helps in identifying whether a video is real or fake by capturing both the details within frames and the connections between them. The system effectively identifies subtle inconsistencies and temporal mismatches in manipulated videos. It provides a robust method for addressing the challenges posed by deepfake content. We tested our approach on a variety of deepfake videos and achieved reliable results with a simple and efficient design, showcasing its potential for real-world applications.

Keywords— Deep learning, Deepfake Detection, convolutional Neural Network (CNN), Recurrent Neural Network (RNN),LSTM