Deep transfer learning for ear recognition: A comparative study

Ali Abd Almisreb, Nursuriati Jamil, Syamimi M. Norzeli, Norashidah Md Din

Research output: Contribution to journalArticle

Abstract

Transfer Learning is an efficient platform to solve problems with little amount of data. In this paper, the performance of three well-known Convolution Neural Network (CNN)-based learning model that are AlexNet, VGG16 and VGG19 for human identification based on ear images are compared. The respective convolution neural networks (CNNs) are fine-tuned to customize it to the ear images dataset. The last fully connected later is replaced with an-other fully connected layer to recognize 10 classes instead of 1000 classes. A total of 3,000 ear images are captured and augmented from 10 male subjects aged between 18 to 27 years old. To train the fine-tuned CNN –based networks, 2,500 images are used and the remaining 500 image are allocated for validation. The proposed fine-tuned CNN-based networks performed well in ear recognition as validation accuracy achieved 100% for all 10 male subjects.

Original languageEnglish
Article number80
Pages (from-to)490-495
Number of pages6
JournalInternational Journal of Advanced Trends in Computer Science and Engineering
Volume9
Issue number1.1 Special Issue
DOIs
Publication statusPublished - 01 Jan 2020

All Science Journal Classification (ASJC) codes

  • Computer Science (miscellaneous)
  • Electrical and Electronic Engineering

Fingerprint Dive into the research topics of 'Deep transfer learning for ear recognition: A comparative study'. Together they form a unique fingerprint.

  • Cite this