Iris recognition technology that works even with walking subjectsFeatured Technologies
November 6, 2019
NEC developed a new technology that enables high-precision identity verification based on iris recognition, which works even with walking subjects without requiring people to stand still in front of the camera. We spoke with the researchers about the details of this technology.
A walkthrough iris recognition that creates synergy with face recognition
― What are the advantages of being able to use iris recognition with walking subjects?
Authentication can be done smoothly without requiring users to stand still in front of the camera. We believe that this will have a huge effect on alleviating congestion and crowding at the security gates at airports and other facilities by streamlining the process. Walk-through face recognition has already been established and put into practical use, but iris recognition has yet to see practical applications.
Face recognition and iris recognition are extremely compatible technologies. If the face is captured, the eyes are also captured at the same time, so synergy can be efficiently produced by drawing on each other’s strengths. For example, iris recognition can be used as long as the eyes are exposed, which means that it is a great complement in cases where face recognition alone is insufficient. Additionally, the combination of face recognition and iris recognition can greatly improve the precision of large-scale 1-to-N authentication of tens of millions of people. In the sense that iris recognition can create synergy with face recognition, there is significance in the development of a walkthrough iris recognition.
― Why was walkthrough iris recognition not developed until now?
Looking at iris recognition from a purely biometrics point of view, it already has high precision and is being used at airports and national ID systems in many countries today. In particular, NEC’s iris recognition showcases the World No. 1 matching accuracy in the Accuracy Evaluation Tests performed by U.S. National Institute of Standards and Technology (NIST) in 2018. Nevertheless, iris recognition for walking subjects has additional challenges. There are broadly two barriers to overcome in order to make this technology a reality.
One is that iris recognition needs an exceedingly large amount of data. The diameter of a human iris is about 1 cm, and for accurate authentication, around 200 pixels worth of information is needed in that space. So what if we captured an area of about 50 cm wide and 100 cm high―this covers people with different body types and height―with this resolution? We will need the pixel information for as many as 200 million pixels. If we tried to capture images and do authentication with this amount of pixel information, naturally the processing will not make it in time.
The other barrier is the narrowness of focus. To actually use iris recognition at gates, the authentication must be completed at a place away from the camera and authentication system. This is because the gate needs to be opened or closed according to the result of the authentication. Therefore, we need to capture the user’s eyes at a place several meters away from the gate and system, but this limits the range where the camera can come into focus. The time span of a user walking at an average speed of 1.5 m/s passing through the focal range is only a split second. A high frame rate is essential for capturing this moment, and this massive image group must be processed in real time. To add to this, ISO/IEC defines an image format for iris recognition, which is a width of 640 pixels and a height of 480 pixels per eye (Note 1). In short, a walking user must be captured at a high frame rate, the images that have the iris in focus must be quickly and accurately detected from among the voluminous group of images, and the extracted images must be cut out into the regulated format. This series of processing tasks must be done almost instantaneously.
- Note 1: ISO/IEC 19794-6
Achieving high-speed, high-precision authentication with positional prediction and automatic detection of optimal images
― How did you solve those issues?
First, to compose an amount of information that can be read in real-time by a camera system, we worked on scanning only the area of both eyes instead of reading all information from the camera sensor.
However, obviously the image capture of just the area of both eyes cannot be started until the scanning position is established, not to mention that the time that can be spared for focusing on a walking user’s eyes are very limited. It is too late if we were detecting the position of the eyes after the camera comes into focus. Therefore, we developed a technology that accurately predicts the position of the user’s eyes at the exact spot the image is captured. We created a design, including the camera system, that captures the user walking from a distance and accurately predicts the position of the eyes when they come into the focal range. When people walk, the head inevitably moves up and down due to reflex action. The key point is that the prediction also takes this into account.
By limiting the scanning area, we were able to capture high-resolution images at high frame rates and extract walking people’s eyes clearly with high definition.
Furthermore, to support real-time authentication of walking people, we made improvements to the processing by broadly extracting images that are in focus from the massive amount of images taken at a high frame rate, detecting the optimal images that have the iris in focus from that extracted group of images, and cutting them out into the specified iris recognition image format.
The extraction of optimal images is very difficult, but thanks to experienced members who are skilled in image quality index design and signal processing, we were able to design our own image metrics for high-speed, precision calculation in short time. The iris recognition itself also uses NEC’s in-house engine, the World No. 1 recognition engine, as I mentioned before. It has also been a great help that we have known the engine well and could predict what characteristics in an image make for successful results.
All of these technologies have contributed in realizing the walkthrough iris recognition.
Aiming for practical use in 2021 by supporting authentication of glasses-wearers
― Tell us about the future prospects of this technology.
First, we want to improve recognition precision for people wearing glasses as one of our challenges for realistic application. We have an idea about the solution, and we are expecting to reach a high level of robustness that can support practical applications by the end of 2020. We are working on putting this technology to practical use to that end.
Also, as I said in the beginning, we believe that iris recognition can be very meaningful in combination with face recognition. For example, as seen in the adoption of biometrics-based citizen IDs in India and other countries, multiple biometrics such as iris recognition and fingerprint recognition have been combined with face recognition to ensure high security at the national level. When walkthrough iris recognition becomes reality, higher-precision services can be provided by combining with walkthrough face recognition. This technology can also cater to improving security and alleviating crowding at airport security gates as well as improving convenience and security at corporate buildings, stadiums, and concert halls―a wide area of application is waiting for this technology.
Needless to say, it is also very important to have a firm consensus with society in the context of the use of personal information. On that basis, we want to continue our research with a vision of a society where walkthrough iris recognition is widely used as something that improves the users’ convenience, safety, and security.