【作者】 张芳芳；

【导师】 周宁；

【作者基本信息】 武汉大学 ， 情报学， 2004， 硕士

【摘要】 随着网络技术、多媒体技术、数据库技术、海量存储技术等技术的发展，数字图像的数量不断增加，使用日益广泛，并成为信息社会中的主要信息资源之一。如果没有对图像及视频数据的自动和有效的描述，大量信息将淹没在信息的海洋之中，无法在需要时被检索出来。因此，如何组织、表达、存储、管理、查询和检索这些海量的数字图像，自90年代以来一直是一个非常活跃的研究领域。 然而，由于图像往往具有丰富的语义信息和复杂的视觉特征，目前的计算机视觉技术还不能稳定地建立起多媒体对象的语义信息与其视觉特征间的对应关系，使得基于内容的图像检索(CBIR Content-Based Image Retrieval)在检索的准确性上还难以满足实际应用的要求。为了弥补基于内容的图像检索系统性能上的不足，本文对CBIR系统性能优化方法进行了探讨，并通过实验对其优化效果进行验证和评价。 全文分为五个部分： 第一部分阐述了图像的检索模式及其特点，介绍了基于内容的图像检索系统的应用领域和典型的图像检索系统。 第二部分论述了CBIR的关键技术，主要包括图像特征的提取与表达、图像特征的相似性度量技术、图像高维特征的约减和索引技术，以及CBIR系统性能评价等。 第三部分介绍了CBIR系统的优化方法，主要包括聚类技术对图像数据库的优化和多特征组合检索对图像检索效果的改善，并介绍了相关反馈技术的基本理论和引入图像检索系统的意义。 第四部分探讨了基于相关反馈的CBIR系统的实现，介绍了相关反馈算法和基于相关反馈的CBIR系统的构建等。 第五部分通过对实验结果的分析和总结，证明引入相关反馈机制对CBIR系统性能的优化和改善。更多还原

【Abstract】 With the development of the technology of the network, multimedia, database and mass storage, the quantity of digital image is increasing continuously, used extensively day by day, and digital images become one of the main information resources in the information society. Without the automatic and effective description about the image and the vision, a large amount of information would be flooded in information ocean, and unable to be searched out while needing. So, how to organize, express, store, manage, search and retrieval the giant digital images has been a very active research field all the time since the 1990’s.However, because images possess abundant semantic information and complicated vision features, it is difficult to set up relations between semantic information and corresponding vision features of the multimedia target in present computer vision technology, which leads to the Content-Based Image Retrieval (CBIR) to be difficult to fulfill the request for practical application at accuracy of retrieval.In order to overcome the defect in CBIR system, this dissertation studies the methods to improve CBIR system performance, and then gives an experiment to prove and appraise the conclusions through an experiment.The dissertation is divided into 5 parts:The first part expatiates on the image retrieval mode and its characteristics, introduces the typical system of the image retrieval and its application field.The second part describes the key technology of CBIR, mainly including the extraction and expression of the image features, the image similarity measure technology, the decreasing and indexing of image high-dimensional-feature, and evaluating of CBIR’ performance.The third part introduces the improving methods of the CBIR system, mainly including clustering technology., multi-feature retrieval, and the relevance feedback technology .The fourth part discusses about realizing of CBIR system based on relevance feedback, introducing algorithms of relevance feedback and the construction of CBIR system on the basis of relevance feedback.The last part proved effectiveness of CBIR system based on relevance feedback technology through an experiment.更多还原