Skin detection with near-infrared sensor technology at BEZ

Background and Motivation

One of the major challenges in biometrics is ensuring robustness against presentation attacks. Such attacks can take many forms, including facial masks or falsified fingerprints, and their complexity is increasing in light of recent advances in artificial intelligence. Moreover, security assessments are seldom transferable between systems, meaning that resilience against circumvention must typically be re-evaluated whenever software or hardware components are modified.

In view of these factors, regular evaluation and improvement of existing PAD technologies and the development of new PAD technologies are essential.

The BEZ is currently pursuing two PAD strategies:

• Near-infrared presentation attack detection (NIR-PAD)
• Three-dimensional presentation attack detection (3D-PAD)

Research Objective in the BEZ Context

NIR-based PAD is already successfully employed in facial biometrics. Research at the BEZ aims at enhancing existing systems and extending the application of NIR-PAD to additional biometric modalities, such as fingerprint recognition.

Brief description of the technology

Skin recognition is based on the measurement of NIR radiation. When a sample (such as skin or counterfeit materials) is illuminated with NIR radiation, only a certain portion is reflected, while the rest is absorbed by the sample. This reflected portion, also known as the spectrum, can be detected using various sensors and differs from material to material. The spectrum is created by the interaction of the chemical components of the material, with organic compounds in particular being responsible for the characteristic curve of the spectrum. This makes NIR spectra particularly well suited for detecting skin or polymer-based biometric fakes such as 3D-printed fingerprints. Both consist of organic molecules that produce a uniquely identifiable spectrum. For the detection of skin, this technology has the decisive advantage that skin colour has no influence on the spectrum of the skin. Thus, NIR sensing allows for consistent and robust detection irrespective of skin type.

Methodology

We use two different devices to measure the NIR spectra.

The first is a combination of two spectrometers with halogen lighting, one for the visual range (280 nm – 900 nm) and one for the NIR range (860 nm-1760 nm). The combination of the two spectrometers enables the widest possible spectrum to be recorded.

The second device is a self-developed NIR point sensor with LED illumination. It is significantly smaller and cheaper than the spectrometers, making it more suitable for practical applications. Both devices illuminate the sample presented and then measure its reflection spectrum. The spectrometers are primarily used to measure comprehensive reference data, while the point sensor performs the actual skin recognition.

Reflection spectra of the forehead, palm and back of the hand are recorded from the test subjects. The data collected will be used to improve skin recognition for fingerprint and facial recognition systems.

Student Projects and Practical Projects

Ongoing projects:

•  Planning and construction of a gonioreflectometer for measuring angle-dependent reflection properties
• Generation of uniform skin classifiers for the point sensors from the spectrometer data

Completed work:

• Materials science fundamentals of skin and biometric forgeries for facial and fingerprint biometrics
• Light scattering as a material property for distinguishing between human skin and biometric artefacts