Physiology-based face recognition in the thermal infrared spectrum

Area of Science:

• Biometrics
• Computer Vision
• Medical Imaging

Background:

• Current face recognition systems depend on external facial features, which are susceptible to changes from environmental factors and aging.
• These superficial characteristics possess low permanence and can be easily altered, posing significant challenges for reliable identification.
• Existing methodologies attempt to mitigate these issues but are fundamentally limited by the transient nature of the features used.

Purpose of the Study:

• To propose a novel face recognition framework leveraging permanent, innate physiological information beneath the skin.
• To establish the feasibility of using bioheat patterns from thermal imagery for robust facial identification.
• To develop a method resilient to variations in facial pose and temporal changes.

Main Methods:

• Utilized thermal imagery to capture facial physiological patterns, specifically the superficial blood vessel network.
• Employed Bayesian framework for face segmentation and image morphology for blood vessel localization.
• Defined Thermal Minutia Points (TMPs) from the skeletonized vascular network as unique identifiers and incorporated multi-pose image collection for pose invariance.

Main Results:

• Experimental validation on laboratory and public databases demonstrated the effectiveness of the proposed physiological framework.
• The method showed particular strength in addressing the challenge of low feature permanence over extended periods.
• Achieved promising results, confirming the viability of using under-the-skin vascular patterns for face recognition.

Conclusions:

• The physiological framework for face recognition based on thermal vascular patterns is feasible and offers significant advantages over traditional methods.
• The Thermal Minutia Points (TMPs) approach provides a robust and permanent feature set for reliable identification.
• This research opens new avenues for developing advanced, permanent biometric systems and further methodological exploration.