An Ontology-based Semantic Image Retrieval Technology Research

The steep development of computing, network, semantic network and multimedia technology leads to an explosive growth of multimedia data, thus the representation of retrieval of multimedia information has become a research work focus and a frontier research topic.In this thesis we mainly discuss the ontology modeling of image field based on semantic features of digital images resources. And, once the ontology is constructed, the instantiation of local image resources semantic databases could be accomplished by referring to the model and SVM technique. Moreover, by employing the ontology model, the semantic based deduction and indexing could be implemented. Compared against classic indexing methods, semantic based information retrieval could reduce the "semantic gap" between human beings and computers, and increase the ratio of correctness and completeness.The main contributions of the paper are as followings:Semantic based image indexing technique is discussed. And an overall introduction of semantic Web techniques, principles and deduction mechanism of ontology, parsing and deduction tool Jena, and RDF query language SPARQL is summed up.According to the principles of ontology theory, an ontology model of image field is built, referring CIDOC CRM field ontology model, MPEG-7 standard and images high-level semantic resources description tree. In addition, we apply UML meta- modeling and bring forward an ontology UML profile meta model of digital image field.A semi-automatic labeling method, which remedies some defects of current semantic indexing techniques and makes use of SVM, is put forward to construct image knowledge database.An intelligent image semantic indexing system is implemented and put to the test. According to final statistics of the prototype system, the ontology technique proposed in thesis surpasses traditional indexing operations. And it can be seen that the proposed framework is advisable.