Fingerprint-Based Cryptographic Identity: A Custom Recognition Pipeline with Key Pair Generation

Abstract

In this paper, we proposed a personalized fingerprint recognition model that could yield a high average accuracy of 95% across many datasets (”FVC2000- FVC2002- FVC2004 datasets”) and surpass all the baseline experiments. The model utilized modern preprocessing techniques, robust feature extraction—orientation, texture, and frequency-based descriptors—and a LightGBM classifier, and a core point identifier, fine-tuned using stratified cross-validation and stratified feature selection. One of the main novel aspects of the work is the generation of secure biometric-based public/private key pairs, coupled with a password specific to the user derived from the extracted fingerprint features. This paradigm adds a dual layer of security by linking biometric authentication directly to cryptographic key material. We conducted exhaustive experiments on the FVC 2000-FVC 2002-FVC 2004 datasets to demonstrate the reliability and versatility of the model across a variety of fingerprint features. The proposed pipeline is an ideal candidate for privacy- preserving biometric-based applications, such as secure access control, identity management, and biometric encryption frame- works.