Abstract: The semantic gap between low-level visual feature and high-level semantic has become a primary problem．For improving the efficiency of semantic-based image retrieval system,this paper based on chunked weight and a visual vocabulary proposed a semantic image model which utilizes the correlation of low-level feature and high-level semantic．The model is used to interpret the image visual characteristic and detecte high-level semantic,which improves the efficiency of the semantic image retrieval．The experimental results show that this method improves the precision and recall of semantic image retrieval system.

Key words: Bags-of-word,Image retrieval,Sub-block,Semantic