BLINK & MOUTH-MOVEMENT DEEP LEARNING AUTHENTICATION SYSTEM

Abstract

Traditional authentication systems rely on Knowledge-Based Authentication (KBA) methods such as usernames and passwords which are vulnerable to phishing, brute-force attacks, and credential stuffing. These systems cannot verify whether the actual account owner is physically present; they only confirm knowledge of a secret. BioSecure AI introduces a biometric liveness and identity verification platform designed to overcome these limitations.The system replaces static passwords with real-time facial authentication using a webcam to capture live video frames of theuser. To prevent spoofing through photos or recorded videos, dynamic liveness challenges such as blinking or head movement are issued during authentication. Server-side AI analysis is performed using Google Gemini 2.5 Flash to verify three key conditions: the presence of a real human face, successful completion of the liveness challenge, and a match between the live capture and the registered reference image. Client-side face detection is implemented using BlazeFace with TensorFlow.js to provide real-time tracking and user guidance before frames are transmitted for verification. The application is developed using React 19, Vite 6.2, and TypeScript,
ensuring performance and scalability across modern web browsers. By

combining biometric verification with AI-powered liveness detection, BioSecure AI enhances security, improves user experience, reduces reliance on passwords, and offers a scalable solution
for secure digital authentication.