Research On Method Of Content-Based Interactive Image Retrieval

In recent years, with the rapid development and wide use of multimedia and networktechnology, more and more multimedia data has been generated. In order to organize, express,store, manage, and retrieve vast amounts of multimedia data effectively, The technology ofcontent-based image retrieval which is becoming a hot topic of research at home and abroadhas become a key technology in the major projects of digital earth, digital city, digitalhome. To construct interactive content-based image retrieval system, this dissertation focuseson the research of low-level feature extraction of images, relevance feedback technique.Color is a widely used low-level visual feature in image retrieval. Traditional method ofcolor image retrieval based on color histogram has disadvantages: the spatial information ofimage is lost and the image feature dimension is high. In order to overcome the defects, thisdissertation presents a new method of image retrieval based on the difference degree of spatialdistribution. First, the image is split into blocks and the similarity of these blocks is formed;Second, the difference degree of spatial distribution is calculated, then the weight values ofeach partition are defined; Finally, the similarity in different branch module will beaccumulated by weight to get the whole image similarity. Experiments show that thisalgorithm can overcome the disadvantages of the traditional methods, and has better retrievalperformance.In order to integrate relevance feedback into image retrieval, this dissertation presents theconcept of multi-resolution image retrieval and a new method of interactive image retrievalwhich based on multi-resolution image retrieval. First, the algorithm divides the images intosub-blocks by three different resolution modes, and calculates the similarity between theimages under each resolution mode; Second, the similarity gained in different resolution modewill be accumulated by weight, and the similarity scores of the relevance result are given bythe user; Finally, the value of each resolution modes will be adjusted automatically byrelevance feedback. The algorithm can improve the efficiency of retrieval.Finally, on the basis of the above algorithm, an interactive content-based image retrievalprototype of the system which is named Magic-Orb was designed and realized. The prototypeof the system can provide users with a multi-modality image retrieval service, and has thefunction of human-computer interaction. The user can participate in the retrieval process bythe similarity score of the retrieval result.