Face Recognition via Social Network
http://www.eecs.harvard.edu/~zickler/papers/Autotag_IVW2008.pdf
The linked research paper provides an interesting and practicle method of face recognition in personal photos. Traditional face recognition method simply compares the face in a photo to some known faces, and then generates a “similarity score” showing how similar the faces are. If the similarity score reaches some certain value, then the system will automatically recognize the new face. This method might not be accurate sometimes because it is very likely that the new face looks like multiple known faces in the system and the system will generate several high scores, which is confusing. The new method, however, provides a solution to this problem by using social network. Since in most of the personal photos there are more than one faces, the face recognition system can take the relationships of all the figures in the photo into consideration. If one of the faces is known by the system, then it is very likely that the other faces belong to people who are related in the social network to the person whose identity is known. Formally, the system will combine the face similarity score with the social network context into a conditional random field (CRF) and recalculate a more accurate weighted score.
The idea of using social network to enhance face recognition is an excellent demonstration of the importance of social network. In large social networks like Facebook, there are millions of users and billions of pictures. In such huge social networks, the relationships between users can be really complex, but if we can handle the relationships correctly using the knowledge of Graph Theory and social network models, it will help a lot. It will not only make face recognition more accurate, but also bring positive changes to other fields. Social network is developing at an astonishingly fast pace, and making use of it will definitely surprise us in the future.