A group of scientists from Pohang University of Science and Technology (POSTECH) led by Professor Junsuk Rho has developed a new biometric authentication method that identifies a user without storing a fingerprint as digital data. The work was published in the journal Nature Sensors.
Conventional fingerprint recognition systems convert the pattern into a digital code and compare it with a pre-stored template. Even if the template is encrypted, the authentication process requires access to biometric data, which creates potential vulnerabilities. In addition, unlike a password, a fingerprint cannot be replaced if it is compromised.
Instead of seeking more secure storage methods, the researchers decided to abandon digital storage and comparison of fingerprints altogether. In their system, computations are performed through the physical interaction of ultrasound with the finger rather than in a computer.
The key element of the technology is a metasurface (an artificial structure that can precisely control the direction, intensity and phase of ultrasonic waves). In this case it works similarly to the elements of an artificial neural network. The scientists designed the system so that ultrasound passing through the metasurface is directed to different regions depending on the fingerprint presented. If the fingerprint is registered, acoustic energy concentrates in one zone; if not, in another. Authentication is determined by a simple comparison of signal intensities in these zones, without reconstructing the fingerprint image or extracting its features.
The researchers also learned to control the phase, amplitude and switching (on/off) of ultrasound, which gave the metasurface more degrees of freedom for processing information. Unlike traditional diffractive neural networks, which mainly modulate phase, here the intensity and transmission of waves are additionally controlled.
In experiments with the fingerprints of four people, the system achieved an accuracy of 97.92% on 192 verification samples. An additional numerical analysis using an international fingerprint database showed accuracy of up to 99.75% for one registered user. An overall improvement in recognition was also observed as the number of degrees of freedom of the metasurface increased.
According to the authors, the value of the work is that biometric authentication moves from a "store and compare" approach to "physical computation and direct decision". Since the fingerprint does not need to be stored separately, this may reduce the risk of biometric data leakage. Potential areas of application are finance, healthcare, defense, as well as cars, smartphones and wearable devices.
However, for practical adoption, the current experimental system, operating at a frequency of about 40 kHz, must be miniaturized to the megahertz (MHz) range used in industrial sensors and refined to support multiple registered users. The team plans to develop the platform by combining multilayer metasurfaces with digital technologies.
"Our research changes the very starting point of biometric authentication," the POSTECH press service quotes the head of the work, Professor Rho, as saying. "Abandoning separate storage of biometric information, we expect, will lead to the creation of a new authentication method that will reduce the risk of fingerprint leakage as a result of hacker attacks."