An exciting new research project - called Janus - has been announced by US intelligence agency, IARPA, that aims to achieve a dramatic improvement in the performance of facial recognition technology in unconstrained videos and photos.
IARPA (Intelligence Advanced Research Projects Activity) is a government agency that invests in high-risk, high-payoff research programs that have the potential to provide the United States with an overwhelming intelligence advantage over future adversaries.
IARPA says that intelligence analysts often rely on facial images to assist in establishing the identity of an individual, but the sheer volume of possibly relevant video and photographs can be daunting. While automated face recognition tools can assist analysts in this task, current tools perform best on well-posed, frontal facial photos taken for identification purposes. Janus aims to fill this gap by developing tools and techniques to significantly improve the performance of face recognition on unconstrained video and photos.
IARPA hopes the dramatic improvements will be achieved by funding rigorous, high-quality research drawing on a variety of fields to develop novel representations to encode the shape, texture, and dynamics of a face, and new ways to use these techniques for faster and more accurate search and retrieval.
IARPA says that instead of relying on a "single best frame approach" these representations shall make use of all available imagery. The program will conduct empirical testing of recognition performance across datasets. The program will also deal with ambiguities and uncertainties due to incomplete data and partial representations.
Janus will consist of three phases over four years beginning in April 2014.