Masoumeh Bourjandi - Department of Computer, Aliabad katoul Branch, Islamic Azad University, Aliabad katoul, Iran

Feature selection is an important step in image processing, especially for applications such as image retrieval based on content. In a large database, it is not possible to search the entire database of images to identify the images similar to the query image.Thus, techniques that can provide desired characteristics of the user at eachstage of well-timed feedback can be useful to reduce semantic gap and computing the volume. Since the Roughtheoryand Fuzzylogic are two techniques for resolving data ambiguity and uncertainty in image retrieval systems, including content of the images, and the images indexed by user characterize the query image, in this paper, we have proposed a method based on Rough fuzzy theory to reduce the feature vector, update queryimage and provide feedback at each step to determine the classification rules. In order to determine theefficiency of the proposed method, a comparison between the method using a competitive fuzzy edge detection, fuzzy color histogram and color histogram method combined withcolor histogram and fuzzy edges is done. Experiments on the COREL image database with 4000 images has been done. Test results show that the proposed method has higher accuracy than other methods based on the theory of Fuzzy Rough retrieval of images

fuzzy rough theory, image retrieval, relevance feedback