Trimodal machine learning based biometrics system.
Journal:
Scientific reports
Published Date:
Jul 1, 2025
Abstract
Biometrics-based authentication systems have recently been considered as one of the safest methods to secure our data or possessions. In the literature, the solutions based on one or two measurable traits are broadly described. There was also claim that more features than two can make the samples acquisition procedure too difficult to accomplish a high accuracy recognition level. In this work, the authors presented their own fully-automated biometrics system based on three traits collected from human finger-its fingerprint, geometry and veins template. The experiments were divided into two areas. The first related to the design and preparation of the device for fast, secure and convenient samples collection. The second phase related to the design, implementation and evaluation of the fully automated system for human identity recognition based on three traits. Two ideas for feature vector creation are proposed-they are based on simple mathematical operations on the feature vectors obtained from each trait. Evaluation stage was performed with the database consisting of 100 users, each described by 5 samples per trait. The best results are collected with Multilayer feed-forward Neural Networks and reached 99.2% for precision and 97.5% for recall. These results are aligned with the main aim of the study that is efficiency of biometrics-based human identity recognition increasement.
