Objects and Action Detection of Human Faces through Thermal Images Using ANU-Net.

Area of Science:

• Computer Vision
• Biometrics
• Machine Learning

Background:

• Thermal cameras offer advantages in low illumination and beyond the visible spectrum for surveillance.
• Face identification in thermal imagery is challenging due to limited visual data and similar facial heatmaps.
• Cross-modal identification (thermal to visible or vice versa) presents significant difficulties.

Purpose of the Study:

• To propose a robust method for detecting objects and actions on thermal human face images.
• To enhance face identification accuracy in thermal imaging.
• To classify facial features like hats, glasses, and orientation in thermal images.

Main Methods:

• Image pre-processing: resizing and median filtering for grayscale conversion.
• Feature extraction using Principle Component Analysis (PCA).
• Feature selection via Horse Herd Optimization (HOA), face detection with LeNet-5, and classification using ANU-Net with Monarch Butterfly Optimization (MBO).

Main Results:

• The proposed method achieves superior performance compared to state-of-the-art techniques for thermal face recognition.
• Experiments on the Terravic Facial Infrared Database demonstrate high precision in various facial recognition tasks.
• Successful detection of objects and actions on thermal faces, including specific features like hats and glasses.

Conclusions:

• The developed method significantly improves face recognition accuracy in thermal imaging.
• The integration of HOA, LeNet-5, ANU-Net, and MBO provides a powerful framework for thermal face analysis.
• This research offers a promising solution for security and surveillance applications utilizing thermal imaging.