NEC 2010 Annual Report Download - page 25

Image recognition technology is to analyze a variety of
images using a computer to extract the needed information,
such as fingerprints or facial images, according to the
application, and identify the image. By cross-checking the
fingerprints or facial images against registered images in a
database, the identity of an individual can be authenticated
through fingerprint matching or by face recognition.
NEC has long conducted R&D on image recognition
technology. NEC’s fingerprint identification technology
dates back approximately 40 years, while its face
recognition technology has a roughly 20-year history.
NEC commercialized fingerprint identification technology
in 1982. Since then, NEC has deployed more than 200
systems, including automated fingerprint identification
systems (AFIS) for law enforcement and national ID
systems, primarily to government users in 30 countries
around the world. Additionally, NEC’s fingerprint identifica-
tion technology is widely used both in Japan and overseas
for such applications as room access control systems and
personal authentication systems.
Since being commercialized in 2002, NEC’s face
recognition technology has steadily penetrated the market,
with applications in immigration control systems using
passport images, as well as facility entrance systems
using facial images for entry passes. Going forward, NEC
expects growth in the applications for its face recognition
technology, including digital signage (electronic advertise-
ments), as well as user interface applications for mobile
phones, home appliances, ATMs, automobiles and so forth.
Eyeing further growth in opportunities to use these face
recognition technologies, Central Research Laboratories
are directing research at increasing recognition accuracy
while making the technology more resilient to variations in
photo-shooting environments. Two methods are currently
being developed. One is the Perturbation Space Method.
This method generates numerous images from various
facial angles and lighting conditions from a single registered
image. The generated images are then added to the
registered images, in order to realize robust matching
against changes in photo-shooting conditions and
enhance recognition accuracy. The second method is the
Discriminative Multi-feature Fusion Method. Under this
method, only the effective features for identifying the
individual are selected adaptively from the many image
features extracted from the facial image. Using these
features, the system is robust with respect to facial
changes caused by aging or the effect of wearing or
taking off glasses, among other factors.
These technologies, cultivated by NEC through half a
century of R&D activities, have received high marks
throughout the world. Since 2004, NEC’s fingerprint
identification technology has ranked first in the world each
year in the benchmark test conducted by the National
Institute of Standards and Technology (NIST) of the U.S.
Furthermore, NEC’s face recognition technology placed
first in the NIST’s biometrics technology benchmark test
(still face image recognition) in 2010.
NEC will leverage its globally competitive fingerprint
identification and face recognition technologies to
proactively expand its public safety business.
IMAGE RECOGNITION TECHNOLOGY IN SUPPORT OF PUBLIC SAFETY BUSINESS
Developed method: Only one frontal facial image is required.
Conventional method: Each individual must take numerous and accurate facial images under various
positions and illumination conditions
PERTURBATION SPACE METHOD
Generate and register various position and illumination variation images from one registered image.
Registered
image
3D facial shape
Illumination base
Position variation image
Illumination variation
image
Registration Lessen the degra-
dation in recognition
accuracy caused by
changes of facial
directions and illu-
mination variation.
24
NEC Corporation
Annual Report 2010