Convolutional neural networks show promise for enhancing facial recognition technology

Thermal, infrared (IR), facial recognition technology has advanced apace recently. Research published in the International Journal of Information and Communication Technology, moves us another step toward a tenable system that overcomes some of the limitations of traditional visible-light systems.

Naser Zaeri of the Faculty of Computer Studies at the Arab Open University in Ardiya and Rusul R. Qasim of Kuwait Technical College in Abu-Halifa, Kuwait, explain how IR imaging sidesteps the problem of ambient lighting conditions and variations in skin tone seen with visible-light facial recognition.

The use of thermal imaging relies on capturing the unique heat patterns emitted by the face rather than reflected light. The heat pattern observed is determined almost wholly by a person’s facial vasculature and tissue structures beneath the skin.

These are consistent, broadly speaking, regardless of environmental lighting and skin tone. This could make thermal IR a much more reliable alternative to visible-light imaging for biometric identification.

However, thermal recognition has faced challenges. The technology often has to cope with degraded image quality due to factors such as noise, blurring, reduced spatial resolution, and temperature drift.

Additionally, variations in facial expression and pose can complicate the recognition process. Overcoming these issues requires advanced methods capable of accurately processing and recognizing faces even in less-than-ideal conditions.

Zaeri and colleagues have demonstrated the potential of Convolutional Neural Networks (CNNs) in enhancing the recognition of degraded thermal face images. CNNs are a class of deep learning models that have made a significant impact on the field of computer vision, thanks to their ability to automatically extract and learn complex features from raw images without requiring extensive pre-processing. This capability makes CNNs particularly well-suited to face the biometric challenge.

The team has worked with the well-known ResNet-50 CNN architecture. They applied it to a database of 7,500 thermal images in order to evaluate performance with images of different quality and where facial expression and pose are different.

The promising results show that this CNN-based system can achieve better recognition accuracy even with degraded thermal images and works across a range of scenarios. The work will have applications in security and the military world.

Thermal, infrared (IR), facial recognition technology has advanced apace recently. Research published in the International Journal of Information and Communication Technology, moves us another step toward a tenable system that overcomes some of the limitations of traditional visible-light systems.

Naser Zaeri of the Faculty of Computer Studies at the Arab Open University in Ardiya and Rusul R. Qasim of Kuwait Technical College in Abu-Halifa, Kuwait, explain how IR imaging sidesteps the problem of ambient lighting conditions and variations in skin tone seen with visible-light facial recognition.

The use of thermal imaging relies on capturing the unique heat patterns emitted by the face rather than reflected light. The heat pattern observed is determined almost wholly by a person’s facial vasculature and tissue structures beneath the skin.

These are consistent, broadly speaking, regardless of environmental lighting and skin tone. This could make thermal IR a much more reliable alternative to visible-light imaging for biometric identification.

However, thermal recognition has faced challenges. The technology often has to cope with degraded image quality due to factors such as noise, blurring, reduced spatial resolution, and temperature drift.

Additionally, variations in facial expression and pose can complicate the recognition process. Overcoming these issues requires advanced methods capable of accurately processing and recognizing faces even in less-than-ideal conditions.

Zaeri and colleagues have demonstrated the potential of Convolutional Neural Networks (CNNs) in enhancing the recognition of degraded thermal face images. CNNs are a class of deep learning models that have made a significant impact on the field of computer vision, thanks to their ability to automatically extract and learn complex features from raw images without requiring extensive pre-processing. This capability makes CNNs particularly well-suited to face the biometric challenge.

The team has worked with the well-known ResNet-50 CNN architecture. They applied it to a database of 7,500 thermal images in order to evaluate performance with images of different quality and where facial expression and pose are different.

The promising results show that this CNN-based system can achieve better recognition accuracy even with degraded thermal images and works across a range of scenarios. The work will have applications in security and the military world.