040521基于文本的web图片搜索引擎的研究 - 北京大学网络与信息 .doc

硕士研究生学位论文 题目：基于文本的Web图片搜索引擎的研究 姓 名： 王鹏云 学 号： 系 别： 计算机科学技术系 专 业： 计算机应用 研究方向： 网络与通讯 导 师： 宋再生 教授 二零零四年五月 版 权 声 明 任何收存和保管本论文各种版本的单位和个人，未经本论文作者同意，不得将本论文转借他人，亦不得随意复制、抄录、拍照或以任何方式传播。否则，引起有碍作者著作权之问题，将可能承担法律责任。 摘 要 本文研究工作是针对Web图片搜索引擎应用背景，以构建Web图片搜索引擎为目标，地研究如何从HTML文档中提取图片相关信息，保证高效和准确的实现图片检索在真实数据实验和分析的基础上提出了若干关键技术，， 本文提出的方法通过分析验证与图片相关的信息提高图片的通过统计总结HTML文件中表现出的一些潜在规律LSI算法应用于图片搜索引擎来整合文字和内容信息的方法，并通过简单实验进行了效果验证。设计了一个Web图片搜索引擎。Abstract In the thesis, we form a scheme to design a large-scale Web image search engine system using mainly text-based technology. We introduce and research a series of techniques related to Web image search engine, such as crawling, relevance ranking (VSM and LSI), information extraction and indexing. Those techniques will be used in our system design. We concentrate on how to extract information relevant to images from HTML documents more effectively and precisely. According to experiments and analysis on real data, we propose several key techniques as below for designing the system: We analyze carefully the structure of HTML components including img tag, a tag, title of web page, anchor text of web page, URL of image, meta tag, table tag, surrounding text of img tag etc. And sum up nine extraction patterns to fetch information relevant to images. We also research three extracting methods: DOM based method, String based method and Wrapper based method. We propose some methods to filter useless images according to file size, width and height of images and referred count of images by img tags. Through statistics of mass of HTML documents, We conclude some latent rules, such as the difference between JPG and GIF, the difference between a tag and img tag, the difference between different referred count of images. We Simply research the application method of LSI to integrate high-level and low-level information of images. We design and implement a text-based Web