Autonomous Face Classification Online Self-Training System Using Pretrained ResNet50 and Multinomial Naïve Bayes.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Existing convolutional neural networks for face recognition require extensive training data and time, dependent on hardware.
• Pretrained models, like ResNet50, can efficiently encode face images after removing classifier layers.
• Real-time autonomous learning for face recognition presents challenges in data acquisition and model adaptation.

Purpose of the Study:

• To develop a novel, autonomous learning system for real-time face recognition.
• To enable continuous learning and classification of individuals without manual intervention.
• To investigate the role of novelty detection in autonomous face recognition systems.

Main Methods:

• Utilized a pretrained ResNet50 model for encoding face images captured by a camera.
• Employed Multinomial Naïve Bayes classifier for autonomous, real-time person classification.
• Integrated a novelty detection algorithm based on Support Vector Machine (SVM) to identify unknown faces for training.

Main Results:

• The system demonstrated successful autonomous learning and correct recognition of new faces under favorable conditions.
• The novelty detection algorithm proved critical for the system's ability to distinguish and learn new identities.
• False positives in novelty detection could lead to misclassification, assigning multiple identities or incorrect grouping.

Conclusions:

• The proposed autonomous system effectively performs real-time face recognition and learning.
• The novelty detection mechanism is a key component ensuring the accuracy and reliability of the system.
• Further refinement of novelty detection is essential to prevent misclassification errors in dynamic environments.