National
Security
Facility
Surveillance with Face Recognition
Facial
Recognition in Unconstrained Environments
It
has recently become clear that currently available face recognition
algorithms are not robust in unconstrained environments. The presence
of confounding factors, such as variable illumination, facial expression,
and/or decoration (e.g., eyeglasses or facial hair), can reduce
the performance of commercially available systems to unacceptable
levels. Researchers at ORNL have developed a new face recognition
algorithm that is significantly more robust to such factors and
is suitable for personnel monitoring in secure facilities.
Base
Technology
The
ORNL face recognition method employs modular subsystems –
each frontal face image is processed by three different observers,
each trained to recognize different regions of the image (see Fig.
1). Each of these observers is a template-based classifier that
employs a dimensionality reduction approach recently developed at
ORNL called direct, weighted linear discriminant analysis (DW-LDA).
The
output of each of the three observers is then integrated using classifier
combination to produce an identification result with a confidence
measure. Experimental results have demonstrated that the ORNL system
is robust in the presence of confounding factors.
Applications
The
target application for the ORNL face recognition technology is passive,
automated personnel monitoring in secure facilities. The system
can be combined with existing security measures, such as radio frequency
personnel and material tags, to enhance facility security.
Specifications
and Features
- Robust
to illumination, expression, and decoration
- Easily
extendible to multiple modalities (thermal, 3D) and multiple poses
- Provides
confidence measure and can indicate multiple potential matches
- Intended
for video-based personnel monitoring in secure facilities
Fact
Sheet available here in PDF format.
Point
of Contact:
Jeffery R. Price, Ph.D.
Image Science & Machine Vision
Engineering Science and Technology Division
Oak Ridge National Laboratory
P.O. Box 2008, MS-6010
Oak Ridge, Tennessee 37831-6010
Office: (865) 574-5743
E-mail: pricejr@ornl.gov
|