Identifying people from video streams or boatloads of images can be a daunting task for humans and computers.
But a 4-year development program set to start in April 2014 known as Janus aims to develop software and algorithms that erase those problems and could radically alter the facial recognition world as we know it.
Funded by the Office of the Director of National Intelligence's "high-risk, high-payoff research" group, Intelligence Advanced Research Projects Activity (IARPA) Janus "seeks to improve face recognition performance using representations developed from real-world video and images instead of from calibrated and constrained collections."
During daily activities, people laugh, smile, frown, yawn, and morph their faces into a broad variety of expressions. For each face, these expressions are formed from unique skeletal and musculature features that are similar through one's lifetime. Janus representations will exploit the full morphological dynamics of the face to enable better matching and faster retrieval, IARPA stated.
IARPA says Janus is not focused on furthering generic object recognition, or on the development of advanced interfaces for facial analysis but rather wants new technology that can make use of use new image representations where additional information such as novel poses or lighting variations to improve recognition performance.
IARPA envisions experts from quite a variety of technical fields could come together to develop a Janus system: biometrics, pattern recognition and machine learning, computer vision and image processing, computer graphics and animation, mathematical statistics and modeling, physiology and anatomy, high performance computing, and software development. Development teams might also include detection experts from other fields in which signal processing involves multimodal, noisy, incomplete, and contradictory data.
IARPA noted a number of challenges potential vendors will have address to build a Janus system, including:
- Demonstrate that new algorithms can work robustly on the full range of media, of various resolutions, quality, and quantity to include full video processing through still photographs.
- Demonstrate that the representation is discriminative and can be algorithmically created using multiple views of a subject.
- Demonstrate an off-line repository of all subject representations and perform matching using this representation repository in lieu of the original media.
- Demonstrate that the representation can scale to support an arbitrarily large number of subjects.
- Exploit all available video and still images of a subject to address the challenges of aging, pose, illumination, and expression.
- Show that the size of a subject's representation is independent of the amount of subject imagery processed.
- Show that while processing time to build representations can be a function of the amount of imagery per subject, search time shall be dependent only on the number of subjects in the repository and not the amount of imagery used to build the repository.
- Support merging representations known to come from the same individual into one representation and conversely separating representations upon determining they identify different individuals.
- Show how individual subjects will be represented across a wide variation of subject age.
- Support robust methods for working with incomplete, erroneous, and ambiguous data during both acquisition and query time.
- Enable analysts to understand the impact of partial and uncertain information on recognition decisions.
- By the final phase of the Janus project, developers "should expect thousands of hours of video of over 10,000 subjects with a wide range of ages. Multiple video clips per subject, with little or no ancillary information, will be provided" for identification.
Sign up for CIO Asia eNewsletters.