Scholarlink

Signal processing has its application to authentication which is the fundamental security tool in information security. Mostly image, audio, and video signals are being used for authentication tasks along with the inclusion of traditional and modern forms of machine learning. In this work, face recognition is studied as the well-known example of image signal processing for biometric recognition preferred over other biometrics including iris, finger, palm, vein, and many others. A novel bag of clas- sifiers approach is proposed for face recognition and three representative face recognition methods—hand-crafted feature-based (LBP), machine learning feature-based (Eigenface, Fisherface), and CNN feature-based are analyzed on three popular datasets - Yale, AR, and LFW. Two kinds of face detection methods (OpenCV, Dlib) are used. Accuracy, specificity, sensitivity, F1-score, MCC, and AUC measurements are used for performance comparison. The performance of the proposed method is found to outperform.

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

Performance Analysis of Classical, ML-Based, and CNN-Based Face Recognition Methods Using a Bag of Classifiers Approach

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Devendra Prasad,
  • Parveen Kumar,
  • Lalit Awasthi

摘要

Signal processing has its application to authentication which is the fundamental security tool in information security. Mostly image, audio, and video signals are being used for authentication tasks along with the inclusion of traditional and modern forms of machine learning. In this work, face recognition is studied as the well-known example of image signal processing for biometric recognition preferred over other biometrics including iris, finger, palm, vein, and many others. A novel bag of clas- sifiers approach is proposed for face recognition and three representative face recognition methods—hand-crafted feature-based (LBP), machine learning feature-based (Eigenface, Fisherface), and CNN feature-based are analyzed on three popular datasets - Yale, AR, and LFW. Two kinds of face detection methods (OpenCV, Dlib) are used. Accuracy, specificity, sensitivity, F1-score, MCC, and AUC measurements are used for performance comparison. The performance of the proposed method is found to outperform.