Scholarlink

Facial recognition, a computer vision task, employs machine learning algorithms to detect and recognize faces in images or videos. The challenge arises when face masks obscure facial features, significantly impacting recognition accuracy. High accuracy is paramount to prevent identity verification issues and enhance security and forensic applications. Support Vector Machine (SVM) excels in recognizing patterns but does not have the ability to extract features on its own. On the other hand, Convolutional Neural Networks (CNNs) can automatically extract high-level features from images, thus eliminating the need for manual feature engineering in SVM. While CNN typically uses the softmax activation function to classify the input vectors, recent studies favor SVM for better performance. Masked faces lead to high intra-class variations and inter-class similarity which in turn compromises the face recognition accuracy. Therefore, the study through an experimental research design leveraged the strengths of the two algorithms to develop a hybrid masked face recognition model. To optimize the performance of pre-trained CNN architectures, it focused on tuning the hyperparameters for VGG16 and MobileNetV2 networks. The grid search technique was utilized to investigate various hyperparameter values and identify the optimal combination systematically. The fine-tuned models were then integrated with a Linear Multiclass Support Vector Machine classifier (LMSVM) for the masked face recognition task. Results show that the VGG16-SVM hybrid achieved 98.27%, and MobileNetV2-SVM 94.36% in recognizing masked faces. The developed hybrid model can be deployed in healthcare settings to recognize patients and staff wearing surgical masks, ensuring seamless identification without compromising safety protocols.

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

Fine-Tuned Convolutional Neural Network and Linear Multiclass Support Vector Machine for Hybrid Masked Face Recognition

  • 章节

  • 发表日期：

    2025年

查看全文
  • Pauline W. Mwaka,
  • David G. Mwathi,
  • Edna C. Too

摘要

Facial recognition, a computer vision task, employs machine learning algorithms to detect and recognize faces in images or videos. The challenge arises when face masks obscure facial features, significantly impacting recognition accuracy. High accuracy is paramount to prevent identity verification issues and enhance security and forensic applications. Support Vector Machine (SVM) excels in recognizing patterns but does not have the ability to extract features on its own. On the other hand, Convolutional Neural Networks (CNNs) can automatically extract high-level features from images, thus eliminating the need for manual feature engineering in SVM. While CNN typically uses the softmax activation function to classify the input vectors, recent studies favor SVM for better performance. Masked faces lead to high intra-class variations and inter-class similarity which in turn compromises the face recognition accuracy. Therefore, the study through an experimental research design leveraged the strengths of the two algorithms to develop a hybrid masked face recognition model. To optimize the performance of pre-trained CNN architectures, it focused on tuning the hyperparameters for VGG16 and MobileNetV2 networks. The grid search technique was utilized to investigate various hyperparameter values and identify the optimal combination systematically. The fine-tuned models were then integrated with a Linear Multiclass Support Vector Machine classifier (LMSVM) for the masked face recognition task. Results show that the VGG16-SVM hybrid achieved 98.27%, and MobileNetV2-SVM 94.36% in recognizing masked faces. The developed hybrid model can be deployed in healthcare settings to recognize patients and staff wearing surgical masks, ensuring seamless identification without compromising safety protocols.