Toward Large-Scale Face Recognition Using Social Network Context

Zak Stone, Todd Zickler, Trevor Darrell
2010 Proceedings of the IEEE  
The authors of this paper believe that social incentives can be used to obtain numerous facial images of faces and they propose a computational method for using these images. ABSTRACT | Personal photographs are being captured in digital form at an accelerating rate, and our computational tools for searching, browsing, and sharing these photos are struggling to keep pace. One promising approach is automatic face recognition, which would allow photos to be organized by the identities of the
more » ... duals they contain. However, achieving accurate recognition at the scale of the Web requires discriminating among hundreds of millions of individuals and would seem to be a daunting task. This paper argues that social network context may be the key for large-scale face recognition to succeed. Many personal photographs are shared on the Web through online social network sites, and we can leverage the resources and structure of such social networks to improve face recognition rates on the images shared. Drawing upon real photo collections from volunteers who are members of a popular online social network, we asses the availability of resources to improve face recognition and discuss techniques for applying these resources. automatically parse images so that they can be effectively indexed, browsed, searched, and shared. One useful way to index photographsVespecially personal photographsVis through the identities of the individuals they contain, and, in theory, this can be executed at scale using automatic face recognition. However, recognizing individuals from facial images is a hard problem, particularly when the images are like those in Figs. 1 and 2: collected Bin the wild[ with uncontrolled variations in pose, lighting, and expression. This difficulty is exacerbated in large online photo collections in which hundreds of millions of individuals might appear; the difference in appearance between individuals becomes very small relative to the appearance variation of any particular individual. Furthermore, even the preparation of training data (by manually labeling images, for example) to enroll people in an automatic recognition system becomes burdensome. This paper argues that online social networks can provide the keys to successful face recognition in large photo collections on the Web. This argument is based on two observations. First, online communities induce social incentives for members to manually attach identity labels to facial images. The resulting practice of users voluntarily Btagging[ themselves and their friends in photos can produce extraordinary quantities of labeled facial images, which reduces or eliminates the traditional enrollment burden. The second observation is that the social network graph of an online community, which is often available in machine-readable form, provides powerful contextual information that improves both performance and computational efficiency. By drawing on photos embedded in the online social network Facebook, we assess the availability of labeled face data, and we build on our earlier study [2] to show how social network context can be leveraged to improve recognition. While these results are preliminary, they suggest that Bsocially aware[ face recognition is a problem that deserves research attention.
doi:10.1109/jproc.2010.2044551 fatcat:npl7bpja25cxddteph3dku7du4