Today, a variety of biometric sensors like the fingerprint scanner are necessary attributes of a modern smartphone. We quite successfully used and stored under this castle a lot of personal data. However, there is no such system that guarantees 100% security. And if we talk about fingerprints, recently, the neural network has learned successfully to fake and even to create on their basis a “master key” to crack biometric identification systems.
As informs edition The Guardian, responsible for the development of a team of scientists from new York University, and the system is called DeepMasterPrints. As explained by project Manager Philip Bontrager,
“Perhaps our method will have wide application in the synthesis of fingerprints. As in the case of many studies in the field of security, demonstration of the shortcomings of existing authentication systems is considered to be an important part of developing a more secure replacement in the future.”
DeepMasterPrints operates using two properties of the authentication systems based on fingerprints. First, due to the fact that the scanners are often much smaller than the finger sensors are not reading the whole finger at once, and only part of it. If you’ve ever programmed system like Touch ID, then you have probably noticed that you are forced to place your fingers a few times and do it in several positions. This is necessary in order to make multiple fingerprint cards. And then the fun begins: when you put a finger, the sensor compares the received data with the whole base, and with any particular fingerprint in the database. So it is enough to match some part of “drawing” in order that the system worked correctly.
And here we come to the second property. The fact that some signs prints like a specific build, lines and swirls are more common than others. This means that the fake fingerprint, which contains a lot of very common, more likely will fit most of the fingerprints of the people.
The neural network DeepMasterPrints chanted a huge database of fingerprints and identify the key patterns. Based on these data, the neural network was not only able to fake a fingerprint, but also to create artificial fingerprints, which can meet multiple real counterparts. At the moment, the neural network can generate fingerprints, which work on average 23% of the scanners, which is quite a lot. It is hoped that the findings will help engineers to improve the technology of fingerprint identification.
This and other news you can discuss in our chat in Telegram.