Scholarlink

<p>Recent advancements in biometric security systems, which are designed to reliably verify an individual’s identity, have garnered significant attention in the fields of security and access control. However, traditional unimodal biometric systems encounter several challenges, including intra-class variability, vulnerability to spoofing, and noise interference, which can undermine their effectiveness. To overcome these limitations, this study introduces a novel penta-modal biometric recognition system that integrates five distinct biometric traits: iris, fingerprint, palm print, ear, and face. The proposed system employs a weighted similarity approach, utilizing Local Binary Pattern (LBP) feature maps that are extracted from query images using a Convolutional Neural Network (CNN). These extracted features are then compared against enrolled templates stored in a database. The operation of the system follows a structured five-step process: (1) pre-processing of biometric data, (2) extraction and classification of LBP feature maps via CNN, (3) score computation for each biometric modality, (4) weighted score-level fusion, and (5) final authentication. The performance of the penta-modal biometric system was rigorously evaluated using various public benchmark datasets, including CASIA-FingerprintV5, CASIA-FaceV5, CASIA-PalmprintV1, CASIA-Iris-V3, the AMI ear dataset, and the MEPCO multimodal dataset, as well as synthetically generated multimodal databases. The system achieved an impressive accuracy rate of 99.85% on the virtual databases, demonstrating significantly enhanced security and reliability compared to traditional unimodal systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Robust penta-modal biometric identification through deep learning and weighted score fusion

  • 期刊论文

  • 发表日期：

    2025年2月7日

查看全文
  • Arivazhagan Selvaraj,
  • Newlin Shebiah Russel,
  • Mridula Seenivasan

摘要

Recent advancements in biometric security systems, which are designed to reliably verify an individual’s identity, have garnered significant attention in the fields of security and access control. However, traditional unimodal biometric systems encounter several challenges, including intra-class variability, vulnerability to spoofing, and noise interference, which can undermine their effectiveness. To overcome these limitations, this study introduces a novel penta-modal biometric recognition system that integrates five distinct biometric traits: iris, fingerprint, palm print, ear, and face. The proposed system employs a weighted similarity approach, utilizing Local Binary Pattern (LBP) feature maps that are extracted from query images using a Convolutional Neural Network (CNN). These extracted features are then compared against enrolled templates stored in a database. The operation of the system follows a structured five-step process: (1) pre-processing of biometric data, (2) extraction and classification of LBP feature maps via CNN, (3) score computation for each biometric modality, (4) weighted score-level fusion, and (5) final authentication. The performance of the penta-modal biometric system was rigorously evaluated using various public benchmark datasets, including CASIA-FingerprintV5, CASIA-FaceV5, CASIA-PalmprintV1, CASIA-Iris-V3, the AMI ear dataset, and the MEPCO multimodal dataset, as well as synthetically generated multimodal databases. The system achieved an impressive accuracy rate of 99.85% on the virtual databases, demonstrating significantly enhanced security and reliability compared to traditional unimodal systems.