NEC achieves the highest performance evaluation in the Face in Video Evaluation performed by America’s NIST

In the Face in Video Evaluation (FIVE), a benchmark test conducted by the U.S. National Institute of Standards and Technology, an internationally respected authority, NEC's face recognition technology won first place with a matching accuracy of 99.2%, leaving the rest of the competition far behind. NEC was ranked first for the fourth consecutive time following still image face recognition tests conducted thus far. We encourage you to read the following overview of the press conference held on March 16 (Thu.).

Senior Vice President Masakazu Yamashina explaining the details of the announcement

Research Fellow Hitoshi Imaoka of Data Science Research Laboratories providing the technical explanation

Press conference

Overview of the press conference presentation

NEC achieves the highest performance evaluation in the Face in Video Evaluation performed by NIST

NEC recently achieved the highest performance evaluation in the Face in Video Evaluation testing performed by the U.S. National Institute of Standards and Technology (NIST), an internationally respected authority. NEC had thus far placed first three consecutive times in still image face recognition tests, making this the fourth consecutive time it has captured the top spot for face recognition. Going forward, NEC will make further technological enhancements by providing feedback to those involved in research and development based on needs in actual operational environments in the business field.

Differences with conventional still image recognition

The following is an explanation of differences with conventional still image face recognition. In still image face recognition, subjects are willingly subjected to recognition, thereby making it “cooperative recognition.” In this case, subjects stand still, face the camera, and stand in the correct place to undergo recognition.

In comparison, video face recognition, for which NEC has been ranked No.1, is known to have the following difficulties.

・High-speed recognition in real time is required
・Simultaneous recognition of multiple people captured by the camera is necessary
・Facial resolution may be low as the subjects are further away from the camera
・It is necessary to deal with various environmental conditions such as the lighting conditions and different face angles of non-cooperative subjects

This video face recognition technology can be applied to both non-cooperative recognition, in which the subject is unaware s/he is undergoing video face recognition, and cooperative recognition, in which the person is aware s/he is undergoing video face recognition. Moreover, it can accommodate various environmental conditions.

New value provided with video face recognition

The graph on the right shows the new value that can be provided with video face recognition.

The vertical axis indicates expansion of the scope of application and improvement of convenience while the horizontal axis indicates the difficulty of the matching process. If video face recognition becomes more accurate, the scope of application will expand even further and convenience will improve.

More specifically, this will enable not only quick detection of subjects from videos but also detection of the same individual from multiple videos.

Expansion of customer segments through the application of video face recognition

NEC is currently deploying video face recognition technology in the Public Safety domain, but also plans to expand the scope of its application in the future to a broad range of industries including manufacturing and processing, media, education and science, financial institutions, wholesale and retail, and logistics, transport, and service.

Through the application of this No.1-ranking video face recognition technology, NEC will accelerate the realization of a safe, secure, efficient and equal society.

NEC’s advanced technologies enabling video face recognition

Now we will move on to an explanation of NEC’s advanced technologies enabling video face recognition.

The first is enhanced robustness against partial occlusion. As you can see in the photo on the bottom left side of the slide, distinctive facial features can be identified and recognized even when part of the subject’s face is partially hidden behind the shoulders of the person in front of them.

The other is enhanced deep learning. This enables accurate identification of individuals even with low-resolution facial images of subjects far from the camera, and images with varying face angles.

Overview of Face In Video Evaluation (FIVE)

The benchmark test recently performed by NIST is referred to as the Face in Video Evaluation (FIVE).

FIVE is a benchmark test in which video face recognition is carried out to evaluate the accuracy of face recognition with non-cooperative subjects. Sixteen leading vendors from around the world participated in this test for which evaluations began in February 2015 and evaluation results were unveiled in March 2017.

The test was sponsored in part by the U.S. Department of Homeland Security. Examples of the evaluations conducted include entry-exit management at an airport passenger gate and the detection of suspicious individuals at a sports arena.


We will now conclude this overview with a summary of the presentation made at the press conference. In the benchmark test conducted by the U.S. National Institute of Standards and Technology (NIST), NEC ranked No.1 for not only still image recognition but also video face recognition.

In addition to being the only vendor to achieve a matching accuracy of over 99% in the passenger gate application, NEC was able to realize high-accuracy recognition for videos captured in adverse environments such as a crowd in a sports arena. Going forward, NEC will aim to develop a system capable of detecting, identifying, and tracking suspicious individuals instantaneously even in crowded places toward the realization of a safe society on a global scale.


Following the end of the press conference, a demonstration of real-time, high-speed face recognition of multiple pedestrians was carried out.