Image Retrieval Technology Researches Based On Color Comentropy&Marginal Information Entropy

With the rapid development of modern electronic technology, network communication technology and multi-media technology, numerous picture information databases with diversified contents appear constantly. People are overwhelmed when facing so much information; hence, it requires urgently the good performance of retrieval tools to improve the efficiency of application; however, the traditional query method based on keywords information can not effectively support multi-media information query and retrieval. Therefore, how to effectively organize, manage and fully utilize the resources of picture information databases has always been the focus of concern for scientific researchers at home or abroad. An image retrieval technology emerges in this background, in which the content-based image retrieval technology (CBIR) is the key to solve this problem.Content-based image retrieval technology has wide applications in industrial and scientific research, such as biometrics, fingerprint identification and face recognition, all belonging to the category of image retrieval; in intellectual property protection, the content retrieval of trademarks database brings about convenience in management and supervision; the realization of medical image retrieval improves the efficiency and quality in medical work; satellite image retrieval can result in more effective management of the images and more effective locating of needed images for TV editing.Locating images needed according to the picture contents is a complicated working process. Currently content-based image retrieval resorts mainly to image characteristics, whose basic working principle is to analyze and extract the characteristics of an image as eigenvector and the characteristics is to be stored in a database. The image retrieval is implemented by providing relevant image features or characteristics and matching them with the eigenvector in the database, and the corresponding image is tracked in accordance with the matching results.This thesis focuses on the conventional technology analyses in content-based image retrieval technology and the optimization of color information entropy algorithm. The author of this thesis puts forward the concept of marginal information entropy and tries to design a more comprehensive retrieval system by combining the two methods to improve the efficiency of searching. Chapter one is an introduction to the research background of CBIR, the concept and general retrieval process, the main applicational fields and future development trends; in Chapter two the author analyzes the technical principles of CBIR and gives emphatic analyses of the key technology and basic methods of CBIR, including the extraction and presentation of various common visual features such as colors, texture and shapes; Chapter three gives an introduction to color information entropy and a detailed description of the retrieval approaches and experiment procedures based on color information entropy; in Chapter four the author of this thesis introduces the marginal features extraction methods and marginal information entropy, simultaneously giving a detailed instruction of the retrieval methods and experiment processes based on marginal information entropy; in Chapter five the author illustrates the scientific property of multi-feature retrieval methods by combining the two methods mentioned above. In Chapter six the author tries to design a multi-feature retrieval system, and the design has been implemented; Chapter seven is the summary of the whole thesis in which the author discusses and predicts the development trends of CBIR.