Spoofify: A Hybrid Anti-Spoofing Facial Recognition System

Abstract

Facial recognition systems have gained widespread adoption as replacements for conventional authenti- cation methods due to their robustness, offering bet- ter security compared to systems such as PINs, pass- words, and fingerprints. However, despite their re- silience, facial recognition remains vulnerable to var- ious attacks , including picture, video, and replay at- tacks. To mitigate these risks, the implementation of a robust anti-spoofing mechanism is imperative. In this paper, we propose a novel hybrid anti-spoofing facial recognition system that combines unsupervised and supervised approaches to address spoofing vul- nerabilities effectively. Leveraging Generative Adver- sarial Networks (GANs) as the cornerstone model for anti-spoofing, our system employs their efficiency in feature learning to discriminate between genuine and spoofed images. The user identification is performed using a K-Nearest Neighbors (KNN) classifier, which analyzes facial features to match individuals against a database of registered users. We evaluate the perfor- mance of our proposed system on the CelebA-Spoof and CASIA-FASD Datasets, achieving a commend- able accuracy rate of 69% on CASIA-FASD. Through rigorous experimentation and analysis, we demon- strate the efficacy and reliability of our hybrid anti- spoofing facial recognition system, offering a promis- ing solution to enhance security in real-world appli- cations. The source code for the approach will be re- leased on: https://github.com/ayeshazia99/spoofify- a-hybrid-antispoofing-facial-recognition-system