Skip to Main Content
An approach to image categorization and retrieval based on the combination of visual and semantic features using rough set theory is presented in this paper. We adopt relevance feedback theory to extract the semantic features of images. The decision table is made with the semantic features (keywords) as the condition attributes and the classes of images as the decision attributes. The optimal features can be selected by attributes reduction. For desired images and sample images, we also introduce a corresponding computing technology of similarity. The minimal keyword set for differentiating image categorization is acquired. It shows in the experiment that the dimension of vector space and the scale of the problem are reduced and both the accuracy and the speed of retrieval system are high.