Image retrieval based on image decomposition and weight assignments

In view of the lack of global color histograms, the paper proposed an image retrieval method based on image decomposition and weight assignment. This method not only utilized effectively color feature of the image, but also reflected spatial information of the image. Further more, the robustness and adaptability of algorithm were improved. In this method, the image was firstly divided into several sub-blocks using color histogram to express the characteristics of each image block. Then, each block was allocated different weights, which indicated the importance of each block. Finally, the similarity of target image and querying image was calculated according to histogram intersection and the weight of each block. The experiments in the image library which had more than 400 images showed that the method presented in this paper was superior to global color histogram and blocks average query methods. Moreover, this method has high retrieval efficiency.