Improving the Efficiency and Effectiveness of Fingerprint Recognition Systems

By Naoyuki Onoe, Associate Director, at Sony Research India

7th January 2022

This article talks about the efforts made toward improving the efficiency and effectiveness of ‘Fingerprint Recognition Systems’ and ways to prevent or mitigate spoofing attacks. This is based on a research paper titled, ‘A Unified Model for Fingerprint Authentication and Presentation Attack Detection’ which is co-authored by :
  • Naoyuki Onoe (Sony Research India Pvt. Ltd.)
  • Prof. Anoop Namboodiri, Additya Popli and Saraansh Tandon (IIIT Hyderabad)
  • Joshua J. Engelsma (Michigan State University, USA)
  • Atsushi Okubo (Sony Group Corporation, Japan)
Due to their widespread usage in many different applications, fingerprint recognition systems are a prime target for attackers. One of the most widely known methods of attack is known as a presentation attack (PA), which can be realized through the use of commonly available materials like Gelatin, Play-doh, and silicone or more expensive and sophisticated 3D printing techniques (these subsets of presentation attacks are also known as spoof attacks). To counter these, various fingerprint presentation attack detection (FPAD) approaches to automatically detect and flag spoof attacks prior to performing authentication have been proposed.

Collaborative Background: Prior to launching Sony Research India, we searched for, and evaluated Indian institutes capable of the necessary research so that we could collaborate with them efficiently. After some evaluation, we discovered that IIIT Hyderabad possesses an established Biometrics lab in collaboration with the internationally well-known professor, Anil Jain from Michigan State University.

We visited IIIT Hyderabad in 2019 and met with Professor Anoop, who pursued his Ph.D. under the guidance of Professor Anil at Michigan State University. In India, we found the fingerprint authentication system is widely used, covering University attendance as well. However, the professor also mentioned that fingerprint spoofing is quite easy and widely done by the students at the University.

How is a typical finger recognition system susceptible to spoofing?

A typical system can be subject to spoofing in several ways, ranging from the use of easily-available materials like Play-doh to the more complex methods such as 3D printing. In addition to this, the function of matching and spoofing are independent of one another, rather than sequential. This causes a gap of continuity in the entire process, leaving space for errors. Simply put, the system reads the fingerprint in one phase and then checks for spoofing, rather than having an integrated method to it, which saves time and increases efficiency.

What approach is advisable to enhance the efficacy of a fingerprint scanning system?

We consider deep learning with a focus on multi-task learning method to solve the problem. At the time of creating fingerprint systems, multi-task learning was a hot topic in the Computer Vision domain but was never applied to fingerprint-related research.

What are the findings and conclusion of the research?

To conclude, the existing fingerprint recognition method comprises Fingerprint Presentation Attack Detection (FPAD) by a matching module. These two tasks are often treated independently using separate algorithms. However, experimental results indicate that these tasks are indeed related. In practice, this enables us to train a single joint model capable of performing FPAD and authentication at levels comparable to published stand-alone models, while reducing the memory and time of the fingerprint recognition systems by 50% and 40%.
We have also shown that our algorithm applies to patch-based fingerprint recognition systems, as well as full image recognition systems. In our ongoing research, we are investigating ways to further reduce the memory and computational complexity of fingerprint recognition systems, without sacrificing system accuracy. This will have tremendous benefits for finger-print recognition systems running on resource-constrained devices and communication channels.

Graphical representation of the findings

Tarang Chugh, Anil K.Jain, “Fingerprint Spoof Generalization”, International Conference on Biometrics (ICB) 2019,

Additya Popli, Saraansh Tandon, Joshua J. Engelsma, Naoyuki Onoe, Atsushi Okubo, Anoop Namboodiri, “A Unified Model for Fingerprint Authentication and Presentation Attack Detection”, International Joint Conference on Biometrics (IJCB), 2021.

Evaluation results of matching performance

To learn more, click on the link below

Skip to content