A SURVEY ON NAMING FACES USING ANNOTATIONS BASED ON EXTERNAL KNOWLEDGE FROM VIDEOS

Information Technology

This paper examines the difficulty of celebrity face naming in unrestricted videos with user-provided metadata. Instead of relying on precise face labels for supervised knowledge, a rich set of associations robotically derived from video content and information from image domain and social cues is leveraged for unverified face labelling. The relations refer to the appearances of faces under dissimilar spatio-temporal contexts and their visual similarities. The knowledge includes Web images weakly tagged with celebrity names and the celebrity social networks. The associations and information are elegantly encoded using conditional random field (CRF) for label inference. Two versions of face annotation are considered: within-video and between-video face labelling. The previous addresses the problem of incomplete and noisy labels in metadata, where null assignment of names is allowed—a difficulty seldom been measured in the literature. The concluding further rectifies the errors in metadata, specifically to correct false labels and annotate faces with missing names in the metadata of a video, by allowing for a group of socially connected videos for joint label inference.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.