Robust Authentication System with Privacy Preservation for Hybrid Deep Learning-Based Person Identification System Using Multi-Modal Palmprint, Ear, and Face Biometric Features

Conventional biometric systems are vulnerable to a range of harmful threats and privacy violations, putting the users who have registered with them in grave danger. Therefore, there is a need to develop a Privacy-Preserving and Authenticating Framework for Biometric-based Systems (PPAF-BS) that allows users to access multiple applications while also protecting their privacy. There are various existing works on biometric-based systems, but most of them do not address privacy concerns. Conventional biometric systems require the storage of biometric data, which can be easily accessed by attackers, leading to privacy violations. Some research works have used differential privacy techniques to address this issue, but they have not been widely applied in biometric-based systems. The existing biometric-based systems have a significant privacy concern, and there is a lack of privacy-preserving techniques in such systems. Therefore, there is a need to develop a PPAF-BS that can protect the user’s privacy and maintain the system’s efficiency. The proposed method uses Hybrid Deep Learning (HDL) with palmprint, ear, and face biometric features for person identification. Additionally, Discrete Cosine Transform (DCT) feature transformation and Lagrange’s interpolation-based image transformation are used as part of the authentication scheme. Sensors are used to record three biometric traits: palmprint, ear, and face. The combination of biometric characteristics provides an accuracy of 96.4% for the Formula: see text image size. The proposed LI-based image transformation lowers the original Formula: see text pixels to an Formula: see text hidden pattern. This drastically decreases the database size, thereby reducing storage needs. The proposed method offers a safe authentication system with excellent accuracy, a fixed-size database, and the privacy protection of multi-modal biometric characteristics without sacrificing overall system efficiency. The system achieves an accuracy of 96.4% for the Formula: see text image size, and the proposed LI-based picture transformation significantly reduces the database size, which is a significant achievement in terms of storage requirements. Therefore, the proposed method can be considered an effective solution to the privacy and security concerns of biometric-based systems.