Research On The Key Technologies Of Semantic-based Image Retrieval Using Fuzzy Domain Ontology

With the development of network and multimedia,the number of digital image keeps increasing.It becomes an urgent problem that how to find needed image efficiently in large-scale image database.However,a great semantic gap breaks the relation from the low-level visual features of the image to its high-level semantic expression,so an efficient image retrieval system should be one that can make full use of the rich semantic information of the image to solve this problem.Therefore,founded on Content-based Image Retrieval(CBIR),Semantic-based Image Retrieval has developed,and it extracts the semantic information of images in a multi-channel way and searches a certain image according to the semantic information to meets people's demands further.Grounded on the analysis of the problems existing in the semantic model of image and the methods of image semantic information extraction,the paper researches the key technologies of SBIR based on Ontology and Fuzzy Set Theory,and designs an efficient semantic-based image retrieval prototype system based on FDO(SBIRFDO).The main work includes:(1)According to the researches on the methods by which low-level features(such as color,texture,shapes) are extracted,we proposed a modified K-means algorithm based on color and texture features to do image segmentation,then we extracted the region-of-interest(ROI) of an image,finally,we extracted the shape and spatial position feature of the regions segmented.(2)Constructed a Semantic Feature Description Model using Fuzzy Domain Ontology(SFDMFDO).This model combined ontology technology and fuzzy set theory and applied them to semantic feature descriptions of an image.Ontology is a kind of model that is used to describe the concepts and the relations of them,and fuzzy set theory can make image retrieval apart from precision of calculating.By adding fuzzy membership to the concepts and the relations of them in the domain ontology,we got a Fuzzy Domain Ontology(FDO) which could be used to describe the semantic features of an image in a way catering for human's fuzzy thoughts.(3)Proposed a new method for image semantic classification based on Vague set and FSVMs(V-FSVMs).We used the new method to map the low-level visual features of an image to the high-level concept feature in ontology,thus the system could automatically acquire the high-level semantic information from the low-level visual features.(4)Presented a relevance feedback algorithm based on semantic web and vague (RFSV).RFSV algorithm added the truth-membership and false-membership to traditional relevance feedback algorithm,and it could more naturally describe the importance of the corresponding features.By updating the semantic structure according to the different understandings of different users,the mapping from low-level visual features to high-level semantic features realized.(5)Designed and realized a semantic-based image retrieval prototype system based on FDO(SBIRFDO).We did some contrast experiments on standard image library comparing with the SBIR based on keywords.The results proved that the method in this paper was superior to the SBIR based on keywords in precision and recall performance. Furthermore,after some times relevance feedback,the performance of the system improved gradually and tended to be stable.