摘要：The popularity of photo-sharing websites like Flickr give us a chance to observe what ordinary users do in their daily life. Particularly, Flickr allows the users to provide personalized tags when uploading photos, and then we can annotate Flickr photos using user-supplied tags. This paper proposes an approach to automatically annotate Flickr photos by exploiting user-supplied tags. User-supplied tags are submitted to Wikipedia to prune noisy tags, and then the reserved tags are denoted as initial tags. Afterwards, the initial tags are ranked using manifold-ranking algorithm, by which regions of the photo to be annotated are served as queries to launch the manifold-ranking algorithm which ranks the initial tags according to their relevance to the queries. Next, using Flickr API, top ranked initial annotations are expanded by a weighted voting scheme. Finally, we combine top ranked initial tags with expanding tags to construct final annotations. Experiments conducted on Flickr photos show the effectiveness of the proposed approach.
