Scholarlink

<p>Due to the COVID-19 pandemic, face masks’ widespread adoption in public has spurred an urgent need to investigate the efficacy of masked face recognition systems. In this research, a new approach of leveraging sparse representation and deep learning techniques, termed Sparse Representation based Classification incorporated with tailored CNN (Sparse CNN) is proposed to enhance the performance of masked face recognition. This method utilises Convolutional Neural Network (CNN) as feature extractor and integrates Sparse Representation based Classification (SRC) as a classifier to improve the accuracy of the masked face recognition systems. With the combination of both these techniques, the proposed Sparse CNN leverages the stability against noise and high accuracy at recognising interference issues of sparse representation while utilizing CNN’s shift invariant property to minimise alignment issues for creating a more robust and reliable masked face recognition system. The experimental results prove that Sparse CNN complements each other and achieves accuracies of 88.89% and 93.41% on Labelled Faces in the Wild Stimulated Masked Face Recognition Dataset (LFW-SMFRD) and Real-World Masked Face Recognition Dataset (RMFRD) respectively.</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

Sparse CNN: leveraging deep learning and sparse representation for masked face recognition

  • 期刊论文

  • 发表日期：

    2025年1月31日

查看全文
  • Ming Chun Yo,
  • Siew Chin Chong,
  • Lee Ying Chong

摘要

Due to the COVID-19 pandemic, face masks’ widespread adoption in public has spurred an urgent need to investigate the efficacy of masked face recognition systems. In this research, a new approach of leveraging sparse representation and deep learning techniques, termed Sparse Representation based Classification incorporated with tailored CNN (Sparse CNN) is proposed to enhance the performance of masked face recognition. This method utilises Convolutional Neural Network (CNN) as feature extractor and integrates Sparse Representation based Classification (SRC) as a classifier to improve the accuracy of the masked face recognition systems. With the combination of both these techniques, the proposed Sparse CNN leverages the stability against noise and high accuracy at recognising interference issues of sparse representation while utilizing CNN’s shift invariant property to minimise alignment issues for creating a more robust and reliable masked face recognition system. The experimental results prove that Sparse CNN complements each other and achieves accuracies of 88.89% and 93.41% on Labelled Faces in the Wild Stimulated Masked Face Recognition Dataset (LFW-SMFRD) and Real-World Masked Face Recognition Dataset (RMFRD) respectively.