Secure Fingerprint Reconstruction-based Authentication System using WDBS-VSS and Geometric Mean Ronald Fisher Score Algorithm

This study introduces a comprehensive approach that synergizes cryptographic encryption, Wiener Direct Binary Search-centered Visual Secret Share (WDBS-VSS) and deep learning to fortify the security and precision of fingerprint authentication systems. During enrollment, authentic fingerprint images are collected from individuals, forming the core of the database. Collected images undergo pre-processing, which includes noise reduction, contrast enhancement and resizing. To address privacy concerns, genuine fingerprint images are encrypted using the Advanced Encryption Standard (AES), converting them into an unreadable format without proper decryption keys. Furthermore, the encrypted fingerprint images undergo the WDBS-VSS process. In the authentication phase, a query fingerprint-intended for authentication is pre-processed similarly to the enrollment phase. The pre-processed image then undergoes feature extraction using an improved Deep Belief Network (iDBN) coupled with the Self Improved Wild Geese Migration Algorithm (SI-WGMO). Following feature extraction, the Geometric Mean Ronald Fisher Scores (GMRFS) algorithm is employed which quantifies the discriminatory power of the extracted features. Lastly, classification is executed using an optimized threshold. The hybrid model, enhanced by both feature extraction and GMRFS, determines whether the query finger-print is genuine or fake based on the optimized threshold. This system enhances security and efficiency in fingerprint recognition for modern biometric applications. It provides robust protection against spoofing and tampering attacks and improves fingerprint matching accuracy by using geometric mean statistics. The system achieves up to 15 % higher accuracy and a security level of over 98%.