Fingerprint recognition gets true `Fingerspitzengefühl`

Will we pay using our fingerprint, or enter a building just touching a sensor? Does our mobile phone recognize our fingerprint? It is possible, as far as Dutch PhD student Asker Bazen is concerned. He has improved the verification techniques, resulting in a better result even for deformed and damaged prints. Together with a higher speed, the new methods can take away existing reserves for implementing fingerprint verification. Bazen is finishing his PhD research at the faculty of Electrical Engineering of the University of Twente, on September 18. Of all biometric methods, fingerprints are the most practical, is the conviction of Bazen: it can be done with a simple sensor that is coupled in a smart way with a database. Including the system in a mobile device like a cell phone is no problem in the near future. Iris-detection is another possibility of detection: it is extremely reliable but you need a sophisticated and fairly large camera.

Recognizing fingerprint is a complex problem for image recognition experts. The optical or capacitive sensor turns the fingerprint into a grayscale image: a black and white picture. Put the picture over a picture in the database, you would say, and you see if both print match. In practice, this turns out to be far less easy: this only works for fingerprints that are purely identical, and are saved in the database in exactly the same way. But there are numerous factors modifying the print. The policeman has a ‘flat’ print in his database, while he finds a print on a round whisky glass at the crime scene. Someone with wet or cold fingers has a slightly different fingerprint than someone with dry fingers.

Ridges and valleys Asker Bazen therefore wants to digitally re-shape the picture slightly, for an elastic fit with the picture in the database. Shrink it a little bit here, and stretch them in other places. He therefore uses recognition of the minutiae. A fingerprint consists of ridges and valleys. Where they end or split up, the minutiae are found. Every fingerprint has between twenty and fifty of them. By matching the minutiae of both prints, a new drawing grid is constructed, compensating for elastic deformation.

One of the existing problems is the processing time, especially for databases containing thousands of prints. Ideally, recognition must take place in ‘real time’. The user doesn’t want to wait or get five or six error messages before being admitted. On the other hand, an intruder must not be allowed to get in easily. Bazen mentions the black list problem, in which a small group of people is not allowed in,, e.g. football hooligan. “In that case you have a small database of people not to be admitted. You don’t want people to be able to manipulate their print in order to enter a stadium without a permit.“

In his thesis, he concludes that elastic matching is very promising, but Bazen also found a method that may be more competitive in terms of speed. It uses the main angles of the lines in the fingerprint. Four ‘cloudy’ pictures indicate the concentration of the four main angles (0, 45, 90, 135 degrees). “We always assumed that this is just a way for a first rough evaluation, but the end result is promising as well. Error percentage is going down to about 0.5 percent, with a good chance of still improving this drastically. At the same time, the speed is about 100 times higher.”

With an early version of his elastic method, Bazen in the international Fingerprint Verification Competition. He then already ended in the upper regions. His refined version has been sent to the 2002 edition of FVC. “I expect a lot of it. This technique is better than two years ago. The higher speed is a major advantage as well.”

The good old ink-and-sheets method in this way gets a full digital follow-up. The improvements Bazen, are very good for improving user acceptance. Acceptance is in fact the true bottleneck for replacing part of the PIN-passes and keys. Improved performance and speed are the only way to gain this acceptance. The sensor itself was not part of Bazen’s research. There are flexible and cheap sensors on the market, he says.

Asker Bazen MscEE defends his PhD-thesis on September 18. His promotors are prof.dr. C.H. Slump and prof.dr.-ing. O. Hermann. Comparative illustrations are available.

Press contact: University of Twente, Corporate Communication, Wiebe van der Veen, tel +31 53 4894244, e-mail w.r.vanderveen@utwente.nl