DEVELOPMENT OF FRAMEWORK FOR DECISION MAKING IN BIOMETRIC AUTHENTICATION

Abstract

Mobile phones are widely utilized for high-security applications, such as financial transactions, where personal authentication with a high degree of accuracy and precision is needed. Therefore, biometrics-based authentication solutions are required to avoid security breaches and attacks during high-security transactions. Nowadays, mobile phones have many biometric authentication systems like iris, fingerprint, and face recognition. However, fingerprints or facial recognition-based systems in mobile phones may not be as applicable in pandemic situations like Covid-19, where hand gloves or face masks are mandatory to protect against unwanted exposure of the body parts. The biometric research literature has shown relatively few efforts focused on providing an effective authentication system that supports the user samples impacted by external factors (like gloves, wet hands, face masks) and contextual factors (location, time, and network connection). Therefore, this thesis focuses on investigating methods and evaluating frameworks for effective biometric authentication in mobile phones in the presence of such external and contextual factors. In our work, we propose a multimodal biometric authentication framework for smartphones utilizing touchscreen swipe and keystroke dynamics that can handle the input biometric samples impacted by external factors (like wet hands, and gloved hands). This system uses machine learning-based classifiers to lessen the impact of hand gloves and sanitized wet hands during the authentication process. An experiment employing several classifiers yielded the best authentication accuracy of about 99 percent with 197 users on the Samsung Galaxy S20 device. In light of the COVID-19 pandemic, the proposed multimodal behavioral biometric authentication framework could be widely applicable to smartphones. Another proposed system in our work is the use of keystroke dynamics for feature phones. We have suggested an approach to incorporate the user’s typing patterns to enhance the security of the feature phone. We have applied the k-Nearest Neighbours classification with fuzzy logic and achieved an Equal Error Rate of 1.88%. viii The experiments are performed with 25 users on the Samsung On7 Pro C3590 device. Finally, methods for face recognition with masked faces are investigated as part of this research study. Through this work, we present an approach using the Haar cascade classifier for face detection with Local Binary Patterns Histograms (LBPH) face recognizer. In this proposed work for masked face recognition, an accuracy of 86% is achieved when a Haar feature based cascade classifier with LBPH face recognizer is used which further improves to 97% when used in conjunction with fuzzy logic.