Deepfake Face Detection Using CNN-LSTM Hybrid Model for Video-Based Forgery Identification

Abstract

With the rapid advancement of artificial intelligence and deep learning technologies, deepfake videos have emerged as a significant threat to digital media authenticity. Deepfakes are synthetically generated or manipulated videos in which a person’s face or expressions are altered using deep learning techniques such as Generative Adversarial Networks (GANs). While these technologies have applications in entertainment and media, they pose serious risks in areas such as misinformation, identity theft, and cybersecurity. Therefore, the development of reliable deepfake detection systems has become increasingly important.This project proposes a deep learning-based system for detecting deepfake videos using a hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) architecture. The system is designed to analyze facial features extracted from video frames and classify them as real or fake. The implementation utilizes Python along with libraries such as TensorFlow, Keras, OpenCV, and Tkinter for building an interactive graphical user interface.The system begins by preprocessing a dataset containing labeled images of real and fake faces. Facial regions are detected using a Haar Cascade classifier, and images are resized to a uniform dimension. These processed images are then used to train a CNN-LSTM model. The CNN component extracts spatial features from individual frames, while the LSTM captures temporal dependencies across sequential frames, making the system more robust for video-based analysis.During training, the dataset is split into training and testing sets. Performance metrics such as accuracy, precision, recall, and F1-score are computed to evaluate the model. The trained model is saved and reused for detecting deepfakes in uploaded videos.For real-time detection, the system processes video frames, detects faces, and predicts whether each frame is real or fake. Based on the majority of predictions across frames, the system classifies the entire video. A user-friendly interface allows users to upload datasets, train models, and test videos.The proposed system provides a scalable and efficient solution for deepfake detection. It demonstrates the effectiveness of combining spatial and temporal analysis for improved accuracy. Future enhancements may include the use of advanced deep learning architectures and larger datasets to further improve performance.