Published Online:April 2025
Product Name:The IUP Journal of Computer Sciences
Product Type:Article
Product Code:IJCS020425
DOI:10.71329/IUPJCS/2025.19.2.23-39
Author Name:P Shabana and R Balakrishna
Availability:YES
Subject/Domain:Engineering
Download Format:PDF
Pages:23-39
Biometric recognition systems play a vital role in identity verification and access control by leveraging unique physiological or behavioral traits for authentication, such as fingerprints, iris patterns, facial features and voiceprints. Ensuring robustness to both intra-class and inter-class variability is essential for reliable identification across diverse population and conditions. The advent of deep learning (DL) techniques has offered promising solutions to these challenges by enabling the automatic learning of discriminative features from raw biometric data. The paper proposes a multimodal multi-instance cancelable biometric recognition system that leverages hybrid DL techniques. Initially, DenseNet was introduced for deep feature extraction, facilitating the extraction of informative features from biometric images. To optimize these features effectively, a modified grammatical evolution (MGE) algorithmwas introduced for optimal feature fusion across multiple modalities and instances of biometric data. Furthermore,multi-attention deep neural network (MA-DNN) was employed for similarity matching computation in biometric recognition tasks, distinguishing between match and nonmatch classes. The multimodal templates obtained through optimal fusion not only enhance security against spoof attacks but also contribute to system robustness. It is found that the maximum accuracy achieved by the proposed system is 98.787% for multimodal multi-instance case, which is 13.142% more efficient than the existing systems. Similarly, the maximum area under the curve (AUC) achieved by the proposed system is 97.315% for multimodal multi-instance case, which is 13.142% more efficient than the existing systems.
Biometric authentication verifies identities using fingerprints, iris patterns, and face features.