Abstract: A new image retrieval method using ontology is presented in this paper.First，the proposed approach employs a improved K-mean and fully unsupervised segmentation algorithm to divide images into regions，and low-level descriptors for the color，shape，position，and texture of each region are subsequently extracted.And then define a object ontology using these features.In order to improve the precision of image retrieval，a relevance feedback mechanism，based on support vector machines is invoked.The experiment results show that the proposed approach not only has an excellent precision but also has strong significance for reducing the “semantic gap” between the visual feature and semantic visual.