Development of a user authentication module based on computer vision and machine learning technologies
TL;DR
The proposed authentication module employs facial recognition technology implemented using Python with OpenCV, dlib, and face_recognition libraries, leveraging a pre-trained ResNet-34 deep convolutional neural network for generating 128-dimensional facial embeddings.
Abstract
This paper addresses the development of a user authentication module based on computer vision and machine learning technologies. The study analyses traditional authentication methods, including knowledge-based, possession-based, and biometric approaches, identifying their respective limitations in the context of contemporary cybersecurity threats. A comprehensive comparison of modern libraries for biometric data processing—including OpenCV, Dlib, Face Recognition, DeepFace, and InsightFace—is presented, with particular attention to their accuracy, performance, and implementation complexity. The proposed authentication module employs facial recognition technology implemented using Python with OpenCV, dlib, and face_recognition libraries, leveraging a pre-trained ResNet-34 deep convolutional neural network for generating 128-dimensional facial embeddings. The system architecture integrates a web-based user interface, PHP-based server logic, and a Python image processing subsystem, enabling real-time biometric verification. Additionally, the module incorporates dynamic head movement detection using MediaPipe FaceMesh to enhance security against spoofing attacks. Functional testing utilising LFW and CelebA datasets demonstrated identification accuracy exceeding 96.8%. The results confirm the effectiveness of the proposed approach for practical implementation in information security, education, and access control systems.