Image Retrieval Using Features in Spatial and Frequency Domains Based on Block-Division

In recent years, the demand for image retrieval is increasing due to the development of the information society. There are mainly two categories of image retrieval approaches: TBIR and CBIR. The local binary pattern (LBP) texture method has been reported as a highly robust approach in CBIR studies. In our previous paper [9], we have proposed a novel image retrieval algorithm to improve retrieval accuracy, which using both features obtained from spatial and frequency domains. We added the DCT features as frequency characteristics. In this paper, we propose a block division method in order to improve the retrieval accuracy by adding geometric information. We divide the image area into separate regions relating to the image blocks. Retrieval results with different blocks are first obtained separately and then combined by weighted. Corel database is used for the evaluation of our proposed algorithm. Experimental results show a significantly higher retrieval accuracy compared with previous research.