Scholarlink

<p>This paper extends our previous research on continuous mobile authentication via behavioral biometrics by introducing a deep learning framework based on Siamese Neural Networks (SNNs). Unlike our earlier work, which relied on one-class classifiers and deep autoencoders, the proposed approach integrates two novel gesture features, <i>Distance Travelled</i> (DT) and <i>Covered Area</i> (CA), to capture richer swipe dynamics. The model is trained and evaluated on the large-scale <i>BrainRun</i> dataset, enabling pairwise comparison of swipe gestures for real-time identity verification. To better reflect practical usage, we also implement a confidence-level strategy that accumulates authentication decisions over multiple swipes. Experiments across balanced and unbalanced datasets, and under different gesture types (horizontal, vertical, and combined), show a clear performance gain over previous methods, with the SNN achieving Equal Error Rate (EER) = 0.00% in several configurations. These results demonstrate the effectiveness of combining metric learning with engineered behavioral features, advancing the development of secure, adaptive, and unobtrusive authentication systems for mobile devices. Source code and models are available at: <a href="https://github.com/AnassSej/Siamese-Swipe-Authentication">https://github.com/AnassSej/Siamese-Swipe-Authentication</a></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

Continuous mobile authentication using siamese neural networks and enriched swipe features

  • 期刊论文

  • 发表日期：

    2025年11月10日

查看全文
  • Anass Sejjari,
  • Chouaib Moujahdi,
  • Noureddine Assad,
  • Haidine Abdelfatteh

摘要

This paper extends our previous research on continuous mobile authentication via behavioral biometrics by introducing a deep learning framework based on Siamese Neural Networks (SNNs). Unlike our earlier work, which relied on one-class classifiers and deep autoencoders, the proposed approach integrates two novel gesture features, Distance Travelled (DT) and Covered Area (CA), to capture richer swipe dynamics. The model is trained and evaluated on the large-scale BrainRun dataset, enabling pairwise comparison of swipe gestures for real-time identity verification. To better reflect practical usage, we also implement a confidence-level strategy that accumulates authentication decisions over multiple swipes. Experiments across balanced and unbalanced datasets, and under different gesture types (horizontal, vertical, and combined), show a clear performance gain over previous methods, with the SNN achieving Equal Error Rate (EER) = 0.00% in several configurations. These results demonstrate the effectiveness of combining metric learning with engineered behavioral features, advancing the development of secure, adaptive, and unobtrusive authentication systems for mobile devices. Source code and models are available at: https://github.com/AnassSej/Siamese-Swipe-Authentication