DETECTING DEEPFAKE FACES WITH HYBRID CNN-ViTs-LSTM MODEL AND REAL TIME WEB INTERFACE

Abstract

This project presents a web-based application for detecting deepfake face videos using advanced Data
Science and Big Data techniques with deep learning. The system combines Convolutional Neural
Networks (CNNs) for spatial feature extraction, Vision Transformers (ViTs) for global pattern
recognition, and Long Short-Term Memory (LSTM) networks for learning sequential frame changes.
Large datasets like FaceForensics++, DFDC, and Celeb-DF ensure robust training and high accuracy.
Built with Python, Flask, and Django, the application lets users upload videos for real-time analysis and
generates authenticity scores with visual tampering evidence. By processing large volumes of video data
and storing results securely, this modular system ensures scalability, security, and reliability. Overall, the
project aims to help verify video authenticity and build trust in digital media.