Research On Image Retrieval Using Trademark Information Of Significant Region

Content-based image retrieval technology is one of the popular researches in the fieldof intelligent information retrieval, and is product of image processing and artificialintelligence. It plays an active role in promoting the development of digital informationand even people’s lifestyles. In the perception of image content, human will pay moreattention on the local significant region. So research based on significant region of imagesis a useful attempt to improve the image retrieval technology. The thesis focuses on themethod of image trademark positioning research, the vocabulary tree hierarchy semanticmodel and the image retrieval system based on the model. All aim at providing a way ofeasy understanding retrieval technology for human. Main research works are as follows:(1) It discusses the problem of target image detection technology and a typical methodof significant region detection. To achieve the structure of the focus area of interest, weproduce the multi-level window matching strategy based on support vector machineclassification model. The method can accurately locate the area of trademarks and providesconditions to extract trademark semantic information.(2) Analysis the text search tree conceptual model, and apply it to the image featuredescription and SIFT vocabulary tree. Study semantic knowledge in image retrieval, for theimage object semantics extraction, produced the vocabulary tree level semantic mappingmodel combined with Bayesian statistical decision theory. This model can not onlyquantify the set of low-level features, but also achieve the description of the trademarkssemantic features.(3) Analysis the basic framework of content-based image retrieval system; describe indetail on image retrieval prototype system using trademark information of significant region. The prototype system fuses low-level features and semantic features of trademarkimages, greatly enhancing the retrieval results.