Highlights from the Madrid Report

Documentation of biometric performance


About the study:
Under the management of professor Raul Sanchez-Reillo at University Carlos III of Madrid, a highly respected biometric industry resource, extensive comparative tests involving 3 market leading sensors have been executed. The study involved collection of more than 80,000 fingerprints on each of the sensors from 600 people over a period of 4 months. More than 100 million comparisons using the collected images were done using industry-leading biometric algorithms.

The Madrid study was performed independently of NEXT. Capacitive sensors from two market-leading suppliers were used and ISO-standard best practices methodology was applied. A sequence of sensors for fingerprint collection was randomized for every test subject. The samples represent real life realistic mass market audiences. All operators and subjects involved with the test were objectively kept from knowing the origin of the sensors. All test subjects were requested to do the acquisition in two sessions, with a minimum 2-week break between each test.


Why this study?
We wanted to have an independent third party perform a credible study to prove beyond all reasonable doubt what any experienced biometric industry expert will confirm: a reliable fingerprint system cannot compromise on sensor area.

Key findings:
The study documents dramatic biometric system related system performance variations. Sensor size is proven as the key factor determining the overall performance of a sensor system. Other variables such as sensing principle and sensor resolution play a far less significant role.

When the active area of a fingerprint sensor is halved from 200 to 100 mm2 , the false rejection rates grow dramatically as the study shows. A typical system configured to 1 in 10,000 false acceptance rate (FAR) will achieve less than 1% false reject rate (FRR) for a full-sized area sensor (200mm2 ), while the average FRR increases dramatically to more than 5% for a 100mm2  sensor. This proves that the NEXT large size area sensor is far superior to any commercial touch sensor with only 10×10 or even 8×8 mm active area.

For small sensors, some improvement can be realized by «stitching» during initial finger enrollment. Here multiple finger placements are required to reconstruct a complete fingerprint from smaller fragments. While this method is subject to several international patents, it can only partially compensate the lack of sensing area.

Key conclusions:
A significant percentage of any given population will not be able to or be interested in using a small sensor system with such high false rejection rates.

NEXT sensor performance is far superior to competing small size sensors. The Madrid study documents that the NEXT Active Thermal™ sensor performs similarly to similar-sized, expensive and government approved fingerprint readers.

The study also documents that the ability to acquire and the capture time using the NEXT Active Thermal™  sensors is in line with the results of the 2 competing capacitive sensors. The NEXT sensor was even superior to other sensors in the study measured by NIST NFIQ assessment of image quality.

The Madrid study has been performed in an operator-guided, in-door environment. In real life applications variables related to user behavior and environment add further complexity.  In a out-door use case with impatient or careless users, the frequency of wet fingers, dirty fingers and skewed placements will increase. In such use cases the dependency of sensor size is even more critical – the false rejections rates of small sized sensors deteriorates to levels found unacceptable by most users.

Raul Sanchez-Reillo
at University Carlos III of Madrid