CASIA OpenIR  > 毕业生  > 硕士学位论文
基于语义分析的网络图像检索
其他题名Semantic Analysis Based Web Image Retrieval
桂创华
学位类型工学硕士
导师周志鑫 ; 卢汉清
2010-05-23
学位授予单位中国科学院研究生院
学位授予地点中国科学院自动化研究所
学位专业模式识别与智能系统
关键词网络图像检索 语义分析 层次聚类 Web Image Retrieval Semantic Analysis Hierarchical Cluster
摘要随着数字影像技术和互联网技术的发展,图像作为传递信息的一种重要载体,要比文本形式更加直观逼真、形象生动,因此在人们日常生活中发挥着越来越重要的作用。每天会有百万级甚至千万级的图像数据传入互联网,如何对这些数据进行有效的管理,让用户方便快捷的找到自己所需要的信息,已成为当前迫切要解决难题之一。 在图像检索过程中,用户关心的是在概念层次上的图像内容,即图像所表达的语义。为了使图像搜索系统更好的满足用户需求,如何实现基于语义的网络图像检索变得十分必要。本文将综合考虑传统的基于文本的图像检索 (Text Based Image Retrieval, TBIR) 和基于内容的图像检索(Content Based Image Retrieval, CBIR) 的各自优势,提出了基于语义分析的网络图像检索框架。该框架可以有效克服文本检索和内容检索的不足,通过将语义分析融入检索过程,有效改善了图像检索的准确率和结果呈现方式。 本文的主要工作和贡献有: (1) 提出了一种基于层次聚类的网络图像检索方法。对于用户输入的具有多义性的关键字,传统的基于文本的图像检索技术得到的结果往往包含多个主题,它们交互混叠在一起,不利于用户找到自己需要的信息,我们分析查询关键字的语义特性,并对结果进行语义和视觉层面上的聚类,将结果分类后呈现给用户,让用户方便地就定位到自己需要的信息上。 (2) 提出了一种基于查询图像语义分析的网络图像检索方法。传统的基于内容的图像检索技术受困于“语义鸿沟”,无法取得很好的效果。我们通过对查询图像进行语义分析,并使用学习到的语义和图像的底层特征联合检索,期望返回给用户语义和视觉层面上都相似的图像。同时,考虑到不同特征在不同环境下的描述能力不一样,我们使用语义一致性来衡量特征的描述能力,并给描述能力强的特征更高的置信度。 (3) 搭建了基于层次聚类的网络图像检索系统和基于查询图像语义分析的网络图像检索系统。目前,系统共收录了800万幅网络图像。
其他摘要Digital techniques and Internet techniques have been experiencing a booming period these years. As a main carrier to transmit information, digital images are more vivid, expressive and direct than textual information. So images are playing more and more important roles in our daily life. The explosive growth of images on Internet brings an increasing need for effectively indexing and searching techniques. Usually, during a image searching process, users focus on the semantic meanings of image content. Then how to conduct semantic-level image retrieval becomes to be necessitate and important. To address this problem, we propose a new semantic analysis based image retrieval framework by integrating the two typical techniques of TBIR (Text Based Image Retrieval) and CBIR (Content Based Image Retrieval), so as to promote their strengths and avoid their weakness. In the proposed framework, we associate semantic analysis of relevant textual information of images into the image searching process and further improve the retrieval performance. Besides, a more friendly and informative display of the searching results is presented in our work. The main contributions of this thesis include the following issues: (1) A hierarchical cluster based image retrieval method is proposed. The traditional image search engine returns a ranked list of search results according to their relevance to a given query. However, due to the query’s polysemy, the results always contain multiple topics and they are mixed together. It is inconvenient for users to select what they really want from the topic-mixed results. We first analyze and find the semantic property of the given query, and then cluster the results in the semantic level and visual level. Finally, the cluster results are returned to users with a navigation sidebar. The hierarchical cluster framework can help users find useful images quickly and efficiently. (2) A web image retrieval method via learning semantics of query image is proposed. The performance of traditional content based image retrieval approaches remains unsatisfactory, as they are restricted by the well-known semantic gap. We first analyze and learn the semantic representation of the query image, and then combine the learned semantic representation and the visual features to find relevant images to the query image on both visual and semantic levels. In addition, considering that different visual features have varying discriminative powe...
馆藏号XWLW1529
其他标识符200728014628021
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/7509
专题毕业生_硕士学位论文
推荐引用方式
GB/T 7714
桂创华. 基于语义分析的网络图像检索[D]. 中国科学院自动化研究所. 中国科学院研究生院,2010.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
CASIA_20072801462802(1682KB) 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[桂创华]的文章
百度学术
百度学术中相似的文章
[桂创华]的文章
必应学术
必应学术中相似的文章
[桂创华]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。