3Qs: Facial recognition is the new fingerprint

EarÂlier this month, the FBI began rolling out a $1 bilÂlion update to the national finÂgerÂprinting dataÂbase. Facial-​​recognition sysÂtems, DNA analysis, voice idenÂtiÂfiÂcaÂtion and iris scanÂning will all conÂtribute to the government's arsenal of Next GenÂerÂaÂtion IdenÂtiÂfiÂcaÂtion (NGI) data. We asked RayÂmond Fu, a new assisÂtant proÂfessor with joint appointÂments in the ColÂlege of EngiÂneering and the ColÂlege of ComÂputer and InforÂmaÂtion SciÂence, to explain the sciÂence behind one of these new techÂnoloÂgies: facial-​​recognition software.
How does facial recognition work, and where is the state of the art now?
Face-​​recognition research has been popÂular for more than two decades. Great advances have been made from researchers from a broad comÂmuÂnity, such as bioÂmetÂrics, comÂputer vision and machine learning. The state-​​of-​​the-​​art techÂniques have been applied to real-​​world sysÂtems for appliÂcaÂtions in surÂveilÂlance, secuÂrity and forenÂsics. Face recogÂniÂtion is a techÂnology that requires high accuÂracy, espeÂcially when secuÂrity and forenÂsics facÂtors are conÂsidÂered. The curÂrent chalÂlenges are scalÂaÂbility of dataÂbases; large variÂaÂtion facÂtors in difÂferent enviÂronÂments; aging, makeup and pose facÂtors of faces; and faces in social-​​media spaces.
Face-​​recognition sysÂtems start with face detecÂtion and tracking. ComÂpuÂtaÂtional algoÂrithms detect face posiÂtions and poses in an image and then extract them for proÂcessing and analysis. During this pipeline, a couple of major chalÂlenges create botÂtleÂnecks for the perÂforÂmance of real-​​world sysÂtems. Facial expresÂsions, aging and makeup are key variÂaÂtions that cannot be easily removed. TechÂniques of 3-​​D morÂphable modÂeling and local feaÂtures have been develÂoped to mitÂiÂgate such variÂaÂtions. Lighting variÂaÂtions can sigÂnifÂiÂcantly affect the recogÂniÂtion accuÂracy espeÂcially when a system is used outÂside. BenchÂmark dataÂbases have been colÂlected from well-​​controlled lighting sources for develÂoping lighting insenÂsiÂtive feaÂture extracÂtion and anaÂlytÂical modÂeling for such purposes.
The increasing accesÂsiÂbility of the social-​​media space presents yet a new chalÂlenge to develÂoping a large-​​scale idenÂtity dataÂbase. ConÂfuÂsion of simÂilar appearÂances, overÂload comÂpuÂtaÂtions and mulÂtiple data sources bring up uncerÂtainÂties in modern face recogÂniÂtion. AddiÂtionÂally, new trends of soft-​​biometrics, big data and mulÂtiÂmodality face recogÂniÂtion have opened up new research thrusts.
What are the challenges and differences between identifying a single presented face and picking faces out of a crowd?
Face recogÂniÂtion and idenÂtiÂfiÂcaÂtion are two difÂferent probÂlems. Face recogÂniÂtion is to match a person's face against a set of known faces and idenÂtify who he or she is. For example, in a crimÂinal invesÂtiÂgaÂtion, a detecÂtive may want to ID a susÂpect from a face image capÂtured on a surÂveilÂlance camera.
IdenÂtiÂfiÂcaÂtion is to valÂiÂdate the match of a given face and the claimed ID. For example, if an employee wants to access a secured area in a clasÂsiÂfied departÂment, she shows her ID card to the sensor while a camera capÂtures her face to match it with the record retrieved from the ID card input. If the match passes, the door will open automatically.
How would you address concerns raised by privacy advocates?
Face recogÂniÂtion can be either pasÂsive or active. In the airÂport, for example, the surÂveilÂlance camÂeras are taking videos in real time. PasÂsenÂgers' faces are capÂtured in a pasÂsive way. Online social-​​media spaces, like FaceÂbook, proÂvide public domains for users to share their photos in an active way. Both may involve priÂvacy issues. How to balÂance the priÂvacy issues and the public needs of secuÂrity and human-​​computer interÂacÂtion are new research topics in this era.
In my research group, we have been funded by Air Force Office of SciÂenÂtific Research, IC Postdoc FelÂlowÂship and Google Research on these issues. Our research is mainly focused on underÂstanding social status and netÂworking of social-​​media users and their priÂvacy conÂcerns. We are working on new comÂpuÂtaÂtional methodÂoloÂgies that could well anaÂlyze the visual conÂtent of social media and proÂvide autoÂmatic soluÂtions for human-​​computer interÂacÂtion that could advance future social-​​network ecosystems.
Provided by Northeastern University