Face emotion recognition based on infrared thermal imagery by applying machine learning and parallelism

Area of Science:

• Computer Vision
• Machine Learning
• Biomedical Imaging

Background:

• Facial expression identification is challenging due to varying lighting and environmental conditions.
• Traditional methods struggle with occlusions and illumination variations.
• Infrared thermal imaging offers a robust alternative for facial emotion recognition.

Purpose of the Study:

• To develop novel infrared thermal image-based approaches for accurate facial emotion recognition.
• To enhance processing efficiency by focusing on salient facial regions.
• To improve the robustness and accuracy of facial emotion identification systems.

Main Methods:

• Facial images were divided into four parts, with active regions (eyes, lips) selected for analysis.
• A Convolutional Neural Network (CNN) with ten-folded cross-validation was employed for improved recognition.
• Parallelism techniques were integrated to reduce training and testing processing times.
• Decision-level fusion was applied to consolidate results and boost overall accuracy.

Main Results:

• The proposed method achieved a high recognition accuracy of 96.87%.
• Processing time was reduced by 50% through parallelism and region-focused analysis.
• The use of active facial regions demonstrated effectiveness in improving recognition performance.
• Ten-folded cross-validation enhanced the reliability of the CNN model.

Conclusions:

• Infrared thermal imaging provides a promising solution for reliable facial emotion recognition, especially in challenging conditions.
• Focusing on specific active facial regions significantly improves both accuracy and processing efficiency.
• The developed approach demonstrates robustness and high performance, paving the way for practical applications.