In this paper

we propose a novel image clustering algorithm for effective image retrieval in Web2.0 tag-space. Different users may use different tags to describe the same object

causing inconsistency in tagging. Our algorithm capture the semantically similar tags to perform query expansion

and retrieve the candidate images which are possibly relevant to the query. The candidate tags can be shortlisted according to their tag relevances to the query tags. The shortlisted tags are then clustered on-the-fly using a graph partitioning algorithm. The candidate images are clustered based on the tag cluster results. The proposed algorithm is implemented in a prototype system called PivotBrowser. Experiment results performed on a large scale images that random downloaded from Flickr reveal that our proposal effectively address the inconsistency and ambiguity problems in tag-space image retrieval

and provide improved user satisfactory.