Scholarlink

<p>Human authentication via ear biometrics is an appealing research prospect because of the growing demand for security, surveillance, and access control. The human ear has unique biometric traits that offer various advantages over traditional biometric faces, fingerprints, irises, etc. The contour of the ear provides highly discriminative information for each individual, which is robust in nature. An ear detection technique involving three deep learning models, namely, a DNN, an autoencoder and a GNN, is presented in this study. A public dataset is applied for the training and testing of the models. Additionally, a shape-based feature extraction method is employed where the best 100 feature points are collected from the ear contour by fitting an ellipse on the ear shape. A comparative assessment of three deep learning models is performed. Essentially, this study presents the taxonomy of autoencoder modelling ear detection systems, which will advance a new direction for future researchers in the improvement of speed, accuracy and minimization of complexity in the proposed feature extraction algorithm.</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

Exploration of a shape-focused autoencoder for improved ear biometrics

  • 期刊论文

  • 发表日期：

    2025年5月24日

查看全文
  • Hrithik Pal,
  • Soubhik Acharya,
  • Priti Paul,
  • Bitan Misra,
  • Jungpil Shin,
  • Nilanjan Dey

摘要

Human authentication via ear biometrics is an appealing research prospect because of the growing demand for security, surveillance, and access control. The human ear has unique biometric traits that offer various advantages over traditional biometric faces, fingerprints, irises, etc. The contour of the ear provides highly discriminative information for each individual, which is robust in nature. An ear detection technique involving three deep learning models, namely, a DNN, an autoencoder and a GNN, is presented in this study. A public dataset is applied for the training and testing of the models. Additionally, a shape-based feature extraction method is employed where the best 100 feature points are collected from the ear contour by fitting an ellipse on the ear shape. A comparative assessment of three deep learning models is performed. Essentially, this study presents the taxonomy of autoencoder modelling ear detection systems, which will advance a new direction for future researchers in the improvement of speed, accuracy and minimization of complexity in the proposed feature extraction algorithm.