A Novel Approach to Cancelable Biometrics Through Deep Learning Distortion Metrics

This paper introduces a novel approach to enhancing the security and privacy of biometric authentication systems through an AI-enhanced framework for cancelable biometrics. Traditional cancelable biometrics involve the transformation of biometric data to protect against unauthorized access and data breaches. Our method extends this concept by integrating artificial intelligence to generate distortions in biometric data, ensuring that the integrity of embeddings is maintained. We employ the Labeled Faces in the Wild (LFW) dataset in a cropped format, focusing on facial recognition, to evaluate the effectiveness of our AI-based transformations. The LFW Generator Model, a deep learning architecture designed for this purpose, demonstrates the ability to produce distortions that improve the Precision, Recall, and F1 Score of biometric authentication, as evidenced by our experimental results. These findings highlight the potential of AI to significantly enhance the security and efficiency of biometric authentication systems, opening new avenues for research and development in the field. This study lays the groundwork for further exploration into AI-enhanced cancelable biometrics, aiming to establish a new standard for privacy-preserving biometric authentication technologies.