Content-based Image Retrieval Method And The Experimental System Design

Content-Based Image Retrieval (CBIR) has been one of most active researches in computer vision, image database, knowledge discovery and so on. The present and the future of CBIR and the key issues are discussed in this thesis, and some methods of extracting image content are analyzed detailedly. An actual CBIR system is designed aiming at concrete image database and some researches of CBIR are introduced as follows:Firstly, the application of fuzzy theory in CBIR is researched elementarily, a method based on fuzzy color histogram of the image is carried out, and some analysis and discussion are done about the experiment results.Secondly, the retrieval method based on color feature is analyzed detailedly, and two improved methods based on the object in the image are presented. By comparing with general color method, the improvement of retrieval efficiency is proved. A method using combined color and texture features is also given and compared with which using single feature.Thirdly, an emphasized research is focused on the appliance of relevance feedback in CBIR, and an improved method based on the relevance of the users is given. With plentiful experiments it is proved that the retrieval efficiency can be enhanced greatly by using this method.