Development and Evaluation of Multimodal Biometric Datasets and Fusion Techniques for Enhanced Person Authentication

Abstract

Introduction: Biometric authentication has emerged as a critical component of identity verification systems in the current digital age. Fingerprints, face recognition, iris patterns, and voice are unique to each person and impossible to copy, making them suitable for secure identification. Spoofing assaults, in which an impostor uses phony biometric features, have generated severe concerns. To solve this, anti-spoofing algorithms based on machine learning are rapidly being included into biometric systems. These sophisticated models aid in the detection and differentiation of actual users from spoofing efforts, resulting in increased dependability and safety. Historical Background of Biometric Authentication.

Objectives: The work being done focuses on creating a machine learning-based biometric authentication system that includes spoof detection and user-friendly verification techniques. It includes a fallback mechanism to boost reliability. In addition, a web-based interface is being created to enable real-time authentication and dynamic user registration. The work also incorporates the use of anti-spoofing techniques to improve the system's resilience against presentation assaults. This modular and scalable technique seeks to deliver a safe, adaptable, and real-time solution appropriate for current biometric authentication applications.

Methods: The convenience and reliability of biometric authentication systems have made them a widely adopted method for secure identity verification. Because of its dependability and ease of use, biometric authentication systems have gained popularity as a secure identity verification technique. But these systems are becoming more and more susceptible to spoofing assaults, in which sensors are tricked by the use of phony fingerprints or printed pictures as biometric characteristics. The creation of a machine learning-based biometric authentication system that combines spoof detection with intuitive verification techniques. 

Results: The study results an integrated, dual-modal biometric identification system may greatly improve accuracy, reliability, and usability. The ML models trained on fingerprint and facial datasets had an average classification accuracy of more than 90%. The fallback approach assisted in resolving circumstances when fingerprint scans were weak or distorted, with facial recognition providing as a viable backup alternative. Anti-spoofing models were successful in detecting and rejecting fake inputs, particularly 2D picture assaults and silicone-based fingerprints. Spoof detection relied heavily on texture, depth, and motion information. The realtime web interface, developed using Streamlit, provides a dynamic and user-friendly environment for authentication and new user registration. Real-time registration simplified the dataset growth process and laid the groundwork for scalable biometric systems. 

Conclusions: The investigation describes a biometric authentication system that uses machine learning techniques to identify spoofing threats. The system distinguishes between legitimate and counterfeit biometric samples using classification algorithms and anti-spoofing tactics. It identifies complex spoofing attempts that standard systems typically miss. The system performs consistently under a variety of scenarios, balancing security and user convenience.