A effective clustering indexing based on EMD for Web image retrieval

摘要

In this paper, we develop an effective clustering indexing scheme based on Earth Moverspsila Distance (EMD) for Web image retrieval. By the proposed clustering method, the collected web images can be automatically classified into some image subclasses with the number of the elements less than a given threshold. Then the resulted clustering structure is hierarchical tree, which extremely facilitate to quickly locate the similar image subclass nodes. Additionally, a triple tree structure is presented to effectively represent hierarchical clustering tree, and a query scheme based on approximative EMD measure is developed to achieve fast image retrieval based on the sample image. The experimental results also demonstrate the excellent performance of the proposed method.