Cutting edge development: Interview with Algorithm Specialist Fredrik Rosqvist

 

Fredrik-Rosqvis-black-and-white-300px

“The trend towards smaller sensors and tougher biometric performance has been a trigger for our success”, says Fredrik Rosqvist, Senior Algorithm Specialist and Algo lead at Precise Biometrics. His team of developers are making big strides to maintain and expand Precise Biometrics’ market leadership.

“We have done some fantastic improvements lately, both for biometric performance and latency, trimming our algorithms for different types of sensors”, says Fredrik Rosqvist, who is a key player in Precise Biometrics team of experienced developers.

The real momentum actually started in 2013, which was an exciting year for Precise Biometrics.  With the launch of iPhone 5S, the mobile market got its first major smartphone using fingerprint authentication. This was the start of a giant surge in the demand for algorithm solutions for digital authentication. Fredrik, who had joined Precise Biometrics five years earlier, was well prepared.

“Before I joined Precise I had worked as a specialist in digital image analysis. I had also worked with mobile applications, so I knew about the mobile industries’ demands on performance and user friendliness.

Prior to the surge in the mobile market Fredrik and his colleagues had developed Precise Match On Card™, an efficient algorithm solution for ID and smart cards with standard size fingerprint sensors. The product is used all over the world, with more than 160 million licenses sold. The work with smart card solutions gave Precise Biometrics a perfect platform for the mobile applications:

“A smart card’s memory and processing power is very limited, but we developed an algorithm that was powerful enough to operate efficiently in such an environment. We were ready for all types of tough demands on biometric performance”, says Fredrik.

What has been the biggest development since 2013?
“So far our fingerprint software has been integrated in over 140 mobile devices. Each second, more than 100 000 users are unlocking their device using our algorithm. The intensified usage of fingerprint technology has also led to a large shift towards increasingly smaller fingerprint sensors. This is due to smaller space on the mobile devices and the fact that smaller sensors often are more cost efficient. Demands on both security and convenience has risen. The first sensors we developed our algorithm for were 10 x 10 millimetres large. Today we support sensors sizes down to approximately 5 x 5 and 10 x 2.5 millimetres, and we continue to develop our solutions to support even smaller sensors. As the sensors are getting smaller, the challenge is to extract all the unique features of the fingerprint without losing any information.”

The requirements on our algorithms has also increased. The biometric performance, measured as the false rejection rate (FRR) at a certain false acceptance rate (FAR), now has to be 1-2 % FRR at 0.002 % FAR for all sensor sizes. A considerable drop from 3% FRR three years ago. Furthermore, the requirements on algorithm latency has been increased drastically: from one second to one tenth of a second. Another appreciated feature is an improving biometric performance over time.

“One key advantage with the patented Precise BioMatch™ software is that the combination of small memory usage and very fast matching enables us to store a lot of different views of the fingerprint. This greatly improves biometric performance over time as the algorithms learn more and more about the fingerprint. We call this process dynamic update and it also makes our algorithm adopt to changes in the fingerprint, such as if the finger sustained a cut or to variations in the finger condition during different seasons of the year.”

What are the current trends?
“Fingerprint recognition is increasing as a biometric modality for identification, it is now on the verge of being used in more consumer applications than mobile phones and tablets. I am of course talking about wearables and internet of things. On the hardware level we are also going to see a further shift towards optic and ultrasound sensors. Our algorithm solutions are sensor independent, works with all OS and hardware platforms, so we welcome this shift. Furthermore, we will see more and more phones with the sensor placed under the screen, which makes greater demands on both sensor image quality and algorithm performance. This is a very exciting time for our team of developers.”