Deepfake Face Detection and Recognition Systems

Yash Shrirao, Dr. Ayesha Siddiqui

Abstract: High-powered computing and generative AI have completely changed how we see digital media. Deep learning cracked open new ground for creativity, but it also kicked off an era of deepfakes- so real they can swing elections, scam people, or wreck reputations with just one image. This research dives into deepfake face detection and recognition, especially with the chaos of real-world social media. On clean, controlled datasets like FaceForensics++ or Celeb-DF, models can boast 98% accuracy. But throw those same models at an actual social media stream and, suddenly, accuracy falls by 40 to 50% in AUC. The main culprit? Brutal video compression. Platforms like WhatsApp, Instagram, and Snapchat shrink videos so much that all the tiny signals detectors rely on just vanish.

Keywords: Deepfake Detection, Face Recognition, Artificial Intelligence, Convolutional Neural Networks (CNN), Feature Extraction, Social Media Security, Biometric Authentication, Adversarial Attacks, Real-Time Detection, Data Privacy