Detecting Deepfake Videos using Face Recognition and Neural Networks
Muthu Aravind Murugan, T. Mathu, S. Jeba Priya · 2024
Deepfake videos created using advanced artificial intelligence techniques, pose a significant threat to digital media credibility. This project introduces a holistic strategy for identifying these videos, incorporating face recognition, feature extraction, and an innovative deep-learning model. The methodology involves pre-processing video data, extracting facial features using the face recognition library, and training a neural network model on processed face-only videos. The project filters videos based on frame count, extracts face, and creates a curated dataset for the detection model. Face-only videos are loaded and pre-processed to train a custom neural network model, which combines a pre-trained ResNext CNN with an LSTM layer for temporal feature extraction. The model is trained using Adam optimization with a cross-entropy loss function, and after completion, it has an accuracy of 95% and is capable of differentiate between fake and real videos using a confidence score.