A novel approach for improving tag ranking quality

Social media webs allow users to describe the media content with tags. However, tagging all images in a photo album is a time-consuming task and assigning unified tags to the whole album will greatly degrade the tagging accuracy. In this paper, a novel scheme to automatically tagging photo albums is proposed. For a photo album, an affinity propagation algorithm is first adopted to obtain a set of representative images and the number of representative ones depends on the content. Then, both visual information and semantic information are exploited to estimate the relevance scores of tags for representative images, and a random walk algorithm is adopted to refine the obtained relevance scores. Finally, tags of the rest images are automatically obtained based on a graph-based semi-supervise method. The experimental results on Flickr photo collection demonstrate the effectiveness of the proposed scheme.