Neural networks can create synthetic fingerprints to trick biometric scanners


At the end of the line : Biometric security is widely used on mobile devices for convenience. This method of identity validation is generally considered “secure enough” for most applications, although researchers have repeatedly shown that such measures can be bypassed.

Most of the methods to thwart biometric security involve replicating the identity of a specific user, such as reproducing fingerprints from photographs or using images to deceive facial recognition systems. Recently, however, researchers at New York University and Michigan State University demonstrated a much more surprising tactic that uses neural networks to generate synthetic fingerprints.

The researchers trained the neural network using thousands of real fingerprint images and used a “generator” to create synthetic fingerprints. These fingerprints were then fed into another neural network, a “discriminator,” which is designed to classify the false impressions as real or generated, thereby improving their authenticity through trial and error.

The resulting DeepMasterPrints (named after master keys that can open many different locks) can be used in dictionary-style attacks against fingerprint verification systems with varying degrees of success, depending on the security strength of the user. target system.

At the lowest level of security on capacitive testing, researchers were able to use a DeepMasterPrint to trick the system 76.67% of the time. The mid-level security level was tricked 22.5% of the time, while the top-level security solution was tricked only 1.11% of the time.

Main image via Andrey VP, Shutterstock


Comments are closed.