Researches On Semantics-based Image Retrieval

With the increasingly widespread application of Internet and the increasingly richness of image resources, effective image retrieval in a large-scale has become an important topic now. In this background, content-based image retrieval(CBIR) was proposed. Images can be retrieved by extracting low-level features such as color, texture and shape. However, because of the semantic gap exists between the low-level features and high-level semantics features, the low-level features-based image retrieval still has weak efficiency. So how to establish the association between low-level features and high-level semantic has become one of the important problems to be resolved.In this thesis, the following works have been done based on the review of the status of image retrieval researches:(1) This thesis provides a survey of extraction algorithms of various low-level features, such as color, texture and shape, as well as the measure of similarity and the evaluation of retrieval performance. Semantic-based image retrieval is then studied and the representation and extraction of semantics is summarized. Combined with fuzzy theory, this thesis also studies the application of fuzzy relation in image semantics classification.(2) Association-ship exists between different image semantics in images database. By mining association rules exists in database, the recall rate of the retrieval can be improved. The thesis has made the experiments to show the effectiveness of this method.(3) Relevance feedback(RF) is studied. Based on the researches on images in the Internet, a centre-point-movement-based RF method is proposed. The experiments show that this method can improve the accuracy of semantic classification.(6) Finally, a semantics-based image retrieval system is designed on the basis of previous researches, and the experiments show the effectiveness and efficiency of the system.