Social image reranking with diversity, representative and relevance

In view of the problem in the image retrieval that the returned images don´t match the query tag in the current fast-growing social media websites, i.e., the returned images from the web are irrelevant to the query, thus in this paper we improve the existing image ranking method which based on relevance of Bayesian and diversity of semantic, and take the representative into account in order to get better images results that we want. The final returned images results are generated by the algorithm proposed in this paper which takes relevance, diversity and representative into account and calculates the diverse score based on both the feature similarity and semantic similarity. The experimental results show its superior performance.