BiometricsU.S. intelligence community seeking better face recognition biometrics
Intelligence analysts often rely on facial images to assist in establishing the identity of an individual, but too often, just examining the sheer volume of possibly relevant images and videos can be daunting. While biometric tools like automated face recognition could assist analysts in this task, current tools perform best on the well-posed, frontal facial photos taken for identification purposes. The Intelligence Advanced Research Projects Activity (IARPA), the research arm of the U.S. intelligence community, is seeking significantly to improve the current performance of face recognition tools by fusing the rich spatial, temporal, and contextual information available from the multiple views captured by today’s media.
The Intelligence Advanced Research Projects Activity (IARPA), the research arm of the U.S. intelligence community, invests in high-risk, high-payoff research programs that have the potential to provide the United States with an intelligence advantage over future adversaries. IARPA works closely with the various members of the intelligence community in order to make sure that its programs address relevant future needs and to facilitate the transition of demonstrated capabilities.
IARPA stresses that it is not an operational organization, and it neither collects raw intelligence nor produces and disseminates intelligence analyses.
IARPA was created in 2006 to conduct cross-community (intelligence community, that is) research, target new opportunities and innovations, and generate new capabilities, while drawing upon the technical and operational expertise that resides within the intelligence agencies. IARPA’s structure was modeled on that of the Defense Advanced Research Projects Agency (DARPA).
IARPA’s programs are designed to anticipate the long-term needs of, and provide research and technical capabilities for, the intelligence community. IARPA says that agility is key in the intelligence community, and IARPA’s approach is to look for, and seek out, new innovative ideas and perspectives.
To ensure organizational agility, IARPA focuses on long-term, 3-5 year programs rather than the short-term time horizons.
IARPA says it tackles some of the more difficult challenges across the intelligence agencies and disciplines, and results from its programs are expected to transition to its intelligence community customers.
Improving face recognition biometrics for the benefit of the intelligence community is one example of how IARPA operates.
Intelligence analysts often rely on facial images to assist in establishing the identity of an individual, but too often, just examining the sheer volume of possibly relevant images and videos can be daunting. While biometric tools like automated face recognition could assist analysts in this task, current tools perform best on the well-posed, frontal facial photos taken for identification purposes.
IARPA’s Janus program aims significantly to improve the current performance of face recognition tools by fusing the rich spatial, temporal, and contextual information available from the multiple views captured by today’s “media in the wild.” The program will move beyond largely two-dimensional image matching methods used currently into more model-based matching that fuses all views from whatever video and stills are available. Data volume now becomes an integral part of the solution instead of an oppressive burden.
IARPA says that the program is seeking to fund rigorous, high-quality research which uses innovative and promising approaches drawn from a variety of fields to develop novel representational models capable of encoding the shape, texture, and dynamics of a face. Instead of relying on a “single best frame approach,” these representations must address the challenges of Aging, Pose, Illumination, and Expression (A-PIE) by exploiting all available imagery.
The technologies IARPA would fund must support analysts working with partial information by addressing the uncertainties which arise when working with possibly incomplete, erroneous, and ambiguous data. The goal of the program is to test and validate techniques which have the potential materially to improve the performance of biometric recognition in unconstrained imagery. To that end, the program will involve empirical testing of recognition performance across unconstrained videos, camera stills, and scanned photos exhibiting a broad range of real-world imaging conditions.
IARPA says it is anticipated that successful teams will transcend conventional approaches to biometric recognition by drawing on the multidisciplinary expertise of researchers from the fields of pattern recognition and machine learning; computer vision and image processing; computer graphics and animation; mathematical statistics and modeling; and data visualization and analytics.
