DEEP LEARNING-BASED GENERATION AND DETECTION FRAMEWORK FOR FACE MORPHING ATTACKS IN BIOMETRIC SYSTEMS

Abstract

The increasing reliance on biometric authentication systems, particularly facial recognition, has significantly enhanced security in domains such as banking, border control, surveillance, and digital identity verification. However, the emergence of sophisticated cyber threats, especially face morphing attacks, poses serious challenges to the reliability and integrity of these systems. Face morphing involves blending multiple facial images to create a single synthetic image that can deceive both human observers and automated recognition systems, enabling identity fraud and unauthorized access. This project proposes a deep learning-based face morphing attack detection framework designed to identify manipulated facial images with high accuracy. The system leverages Convolutional Neural Networks (CNNs) to extract complex features such as texture inconsistencies, pixel-level variations, and blending artifacts that are typically present in morphed images. A comprehensive dataset consisting of both genuine and morphed images is used to train and validate the model, ensuring robust classification performance. The proposed system integrates image preprocessing techniques, including normalization, resizing, face alignment, and enhancement methods, to improve input quality and model efficiency. It classifies images into two categories: genuine (verified) and morph attack (detected). Additionally, a web-based interface is developed to facilitate user interaction, allowing image uploads, real-time verification, and administrative monitoring through dashboards. The system also incorporates performance evaluation metrics such as accuracy, precision, recall, and F1-score, along with visualization tools for better analysis. To enhance security and usability, the framework includes real-time alert mechanisms and logging systems that ensure transparency, traceability, and prompt response to suspicious activities. Experimental results demonstrate that the proposed approach significantly outperforms traditional methods in detecting morphing attacks, achieving high accuracy and reliability