Abstract: In order to solve that the existing color histogram retrieval technology exists less robust, feature dimension is too high and so on, a binary tree at all levels based on image pixel sub-image color histogram of image retrieval methods is proposed. First of all, the method combines the characteristics of noise attack, extracts important image pixels of the binary tree at all levels to quantify sub-image color histogram and the right to determine their visual value; then constructs an important tree image pixel color sub-image histogram; finally, important sub-binary tree image pixel color histogram for the characteristics of images to calculate the similarity between images and content retrieval. Simulation results show that this method is not only able to accurately retrieve the user desired image, but also for light, sharpening, fuzzy and other attacks has a better noise robustness.

Key words: image retrieval, pixel tree, quantification subgraph, color histogram, noise