Research On Binary Trade Mark Image Retrieval Method Based On NMI And Entropy Feature

Along with the arrival of multimedia time, people can get more and more image information. How to provide an efficient approach to query these images, which have rich content, therefore becomes an active hot point in the area of retrieval research. Content based image retrieval (CBIR) is just one of key technologies for such a problem.In this thesis, the concept of CBIR is simply introduced and the development of content-based image retrieval is described. Then, the key research techniques of content-based image retrieval are discussed extensively and comprehensively. Extracting image' s content in the level of feature is the base of CBIR . So; the technology of image retrieval based on the visual feature is researched in this thesis.A new image retrieval method based on NMI (Normalized Moment of Inertia) feature of binary image is presented, Which is simple with high efficiency, and has the ability of anti-geometrical distortion while robust to some image processing such as filtering and smoothing, etc, to some extent. Experimental results show that it can be used in trademarks image retrieval.Trademark plays an important role in the market economy, reflecting the merchandise quality and the manufacturer credit standing. According to the feature of the trademark, this thesis proposes a two-grade retrieval algorithm, which is combined the NMI feature and the entropy feature of the trademark image. We also experimented with the methods mentioned above the trademark image, and the results are satisfied.Now, relevance feedback is a focus of research on content-based image retrieval, it realizes through human-machine interaction in the during of image retrieval. First, receive the user' s feedback of the current search result. Then adjust the search automatically according to the feedback information. Finally make use of the optimized query retrieval the result again. This thesis simply introduced the background and classification of relevance feedback. Twokeys step, which is normalization and the weight adjustment of relevance feedback is detailed introduced. Finally gives the relevance feedback algorithm of based NMI and entropy feature according to the trademark library. Experimental results show that it can improve retrieval accuracy.