Identity Annotation in Photo Collections: A Survey

This paper reviews the problem of determining the identity of people appearing in photo collections. The applications and social impact of problem solutions are also given. Finally, some of the main open challenges and other future outlooks are provided. The basic approach to the person identification problem uses machine vision techniques to perform face recognition, however such methods are ineffective, especially for natural photographs that may be captured using cameraphones and other such devices. Context-based methods that leverage metadata are surprisingly effective and may be combined with face recognition to further improve results. Photographs that are tagged with the identities of people appearing in them are useful for browsing, searching, and other social activities. Some speculation on future algorithmic developments based on learning social network structure and the constraints of social life are made at the end of the paper. “Though we were the closest thing Pranab Kaku had to a family that day, we were not included in the group photographs” — Jhumpa Lahiri [2004] “There is a photograph of the dinner, taken by my father, the only picture, to my knowledge, in which my mother and Pranab Kaku appear together” — Jhumpa Lahiri [2004]