This is the top of the page.
Displaying present location in the site.
  1. Home
  2. About NEC
  3. Research & Development
  4. Laboratories
  5. Data Science Research Laboratories
  6. Technology Fields
  7. Video Understanding
  8. Face Recognition Technology
Main content starts here.

NEC's World-renowned Face Recognition Technology

Consistent No.1 in a U.S. vendor evaluation project

Image of Face Recognition

Researcher Hitoshi ImaokaFace Recognition Technology Team Leader Hitoshi Imaoka (Research Fellow)

From the researcher
We have been conducting research and development to come up with the "world's fastest and most accurate recognition algorithm." We have participated in benchmark tests conducted by the National Institute of Standards and Technology (NIST) and have consistently garnered first place in the past four tests, namely the Multiple Biometric Grand Challenge (MBGC) in 2009, the Multiple Biometric Evaluation (MBE) in 2010, the Face Recognition Vendor Test (FRVT) in 2013, and the Face In Video Evaluation (FIVE) in 2017.
Products based on NEC's face recognition technologies are now used in more than 40 countries around the world. Going forward, we will continue to conduct research and development on the core technologies to realize a secure and smart society.

Cutting-edge Technologies

Face recognition technology is comprised of three basic technologies, namely, face detection technology, feature point extraction technology, and face matching technology.
Face detection technology detects the "position of the face" within an image, while feature extraction technology finds the "position of facial feature points," such as the pupil, subnasal point, mouth corners, etc. Face matching technology determines the "identity" of the detected face.

Results of NIST facial recognition benchmark tests: prepared from NIST Interagency Report 8173
(https://www.nist.gov/programs-projects/face-video-evaluation-five)

Research initiatives

NEC Laboratories are conducting research and development on three key technologies that comprises our face recognition technology: 1) face detection technologies for quick and accurate detection of the positions of the faces for recognition within an image, 2) facial feature point extraction technologies for stable analysis of facial features without being influenced by aging and facial expressions, and 3) high-precision face matching technologies for elimination of false matches for any possible scenario.

Evaluation Result Example 1: Passenger Gate

Evaluation Result Example 2: Sports Arena

Three key technologies

Facial recognition technology is comprised of three basic technologies, namely, face detection technology, feature point extraction technology, and face matching technology.
Face detection technology detects the "position of the face" within an image, while feature extraction technology finds the "positions of facial feature points," such as the pupil, subnasal point, mouth corners, etc. Face matching technology determines the "identity" of the detected face.

Detects the "position of the face"

Generalized Learning Vector Quantization (GLVQ)
Rectangular areas that match the face are extracted by sequentially searching face areas starting from the edge of the image. The Generalized Learning Vector Quantization (GLVQ) algorithm, which is based on the Minimum Classification Error criterion, is used to recognize whether areas are face areas or not, enabling fast and accurate face detection functions.

Finds the "position of facial feature points"

Multiple feature point detection method
This method is used to find the position of feature points, such as the pupil, subnasal point, and corners of the mouth. Brightness patterns around the feature points are used to find the most optimum position, while the facial shape model is used to constrain the alignment of feature points, enabling precise estimation of their positions.
To recognize faces in a video, robustness was enhanced against change in face angle and crowded environment (partial face occlusion).

"Identity" of the detected face

Multi-dimensional feature recognition method
The most appropriate feature to recognize an individual is chosen after extracting features from the face, such as facial contours and tilt. This enables robust personal identification that is unaffected by changes due to aging and other factors.
Deep Learning technology was used to enhance robustness against change in face angles and far faces from a camera (low resolution).

Deployed in more than 40 countries

NEC's biometric face recognition technology is used worldwide for fighting crime, preventing fraud and improving public safety. By applying experience in biometric identification solutions used in 40 countries worldwide over the past decade, NEC has concentrated on developing stronger face recognition methods within the framework of biometric security systems and is now applying face recognition technology to law enforcement and other markets.

Results shown from the MBGC2009, MBE2010, FRVT2013, and FIVE2017 (Face In Video Evaluation 2017) do not constitute endorsement of any particular product by the U.S. Government.

Top of this page