DEVELOPMENT OF AI/ML BASED SOLUTION FOR DETECTION OF FACE-SWAP BASED DEEP FAKE VIDEOS

Abstract

The rapid advancement of Generative Adversarial Networks (GANs) and deep learning has made it possible to create highly realistic face-swap deepfake videos, raising serious concerns for privacy, trust, and digital security. Existing detection approaches range from manual forensic review to heuristic, rule-based checks, both of which are slow, expert-dependent, and fail to generalize to new manipulation techniques. This paper presents an AI/ML based solution for the detection of face-swap deepfake videos. The proposed system extracts frames from an input video, detects and aligns facial regions using MTCNN/Dlib, and passes the processed facial data into a Convolutional Neural Network (CNN) based classifier trained to identify artifacts such as blending boundaries, unnatural skin textures, and irregular facial movements. Frame-level predictions are aggregated to produce a video-level REAL/FAKE classification along with a confidence score. The system is implemented using Python, TensorFlow/Keras, OpenCV, and MTCNN, and is deployed through a Flask-based web application for real-time video analysis. Experimental evaluation demonstrates that the CNN-based approach achieves strong detection accuracy, offering a practical and scalable alternative to manual forensic review for identifying manipulated faceswap videos. KEYWORDS: Deepfake Detection, FaceSwap Videos, Convolutional Neural Network, MTCNN, XceptionNet, Face Detection and Alignment, Flask Framework.