A Comparative Study of the Performances of Single-mode, Two-mode, and Three-mode Biometric Security Systems Using Deep Structured Learning Technique

Abstract

The automatic identification of subjects based on their physiological and behavioural attributes is referred to as biometric recognition. These attributes are specific to each subject and remain unchanged over the course of an individual's lifetime. The single-mode, two-mode biometric recognition systems still suffer problems of high rate of false positives and false negatives. Hence, in this paper multiple instances of three passive biometric modalities were captured and combined to address the drawbacks in two-mode and single-mode biometric systems using Convolution Neural Network_Genetic Algorithm (CNN_GA) which is a deep structured learning strategy. A database with 1026 training samples and 684 probing samples of face, ear, and iris modalities was used to test the system. The full system's design and implementation were completed on MATLAB R2016a programming platform. The system's performance was evaluated on the basis of sensitivity, specificity, precision, system accuracy, and computation time. The outcome demonstrates that the designed three-mode system outperformed the two-mode and single-mode counterparts at various threshold values of 0.20, 0.35, 0.50, and 0.76 in terms of sensitivity, specificity, precision, and system accuracy. As this study's findings demonstrate, combining data from several sources adopting deep structured learning strategy has been able to improve the performance of the systems in terms of sensitivity, specificity, precision, and system accuracy.