Annotating Web Image using Parallel Graph Bipartition and Word Clustering

A novel web image annotation method by candidate annotations clustering and parallel graph bipartition is proposed in this paper. Firstly, surrounding texts and other textual information in the hosting pages are extracted as the candidate annotations. For Web images, the candidate annotation sets of which are usually fairly large. Therefore, we cluster candidate annotations to reduce computation complexity. Next, centroids of clustering results and the distance between them are used to construct a graph. Then a parallel 0.87856 heuristics MAX-CUT algorithm is applied to partition the graph. Finally, one part of the graph partition results is selected as final annotation results. Experimental results show that our method works more effectively than existing methods.