【作者】 申家振；

【导师】 张艳宁；

【作者基本信息】 西北工业大学 ， 计算机应用技术， 2005， 硕士

【摘要】 本论文的主要研究内容是基于内容的图像检索。作为一个有着广泛应用背景的研究领域,基于内容的图像检索通常具有较强的任务依赖性。在实际应用中,往往针对特定的应用,采用特定的算法。面对这种研究现状,本文详细分析了基于内容的图像检索的特征提取方法、相似性度量方法以及语义检索方法,在此基础上,构建了一个实用的图像检索系统原型,并进行了相应实验。 在图像特征提取方面,分别从颜色特征、纹理特征和形状特征三个方面进行了分析和研究,并实现了常见的特征提取方法。在形状特征提取方法中,提出了一种基于形状上下文的新方法,即“自适应边界点选取的形状上下文”,实验证明,相比于形状上下文,该方法具有较好的效果。 在相似性度量方面,在分析相似性度量概念的基础之上,实现了常见的度量方法,并通过实验比较了各种度量方法的检索效果。 在语义检索方面,在分析常用的语义检索方法的基础之上,实现了相关反馈中的修改查询向量方法。通过分析该方法的不足,提出了一种新的方法,即“加权查询向量的修改”,并通过实验验证了其有效性。 在此基础上,设计并实现了一个基于内容的图像检索系统原型。通过分析系统功能,将其划分为检索参数设置、检索结果浏览和相关反馈模块三个功能模块,并对三个模块进行了详细介绍。 最后,对本文的研究工作进行总结,并对下一步的研究工作进行了探讨。更多还原

【Abstract】 The thesis is focused on content-based image retrieval (CBIR). As a widely used technology, CBIR seems to be highly task-dependent. In practice, special algorithms are taken according to special application. The researches of this thesis include feature extraction, similarity measure and semantic retrieval. On the basis of it, a system prototype of CBIR is designed and some relevant experiments are carried out.On the aspect of feature extraction, color feature extraction, texture feature extraction and shape feature extraction are separately analyzed and frequently used methods are realized. On the basis of shape context, a new shape feature extraction method called Shape context by choosing the edge points self-adaptively is proposed. Experiments show that the method acquired better effect compared with shape context.On the aspect of similarity measure, the frequently used methods are realized on the basis of analyzing the concept of similarity measure. The retrieval effects of these methods are analyzed on experiments.On the aspect of semantic retrieval, the method of query vector modified is realized on the basis of analyzing frequently used semantic retrieval methods. To overcome the drawback of the above method, a new method named weighted query vector modified is proposed. Experiments show the new method is satisfactory.A system prototype of content-based image retrieval is designed and realized on the basis of it. Through the analysis of system functions, the system is divided into three modules as follows: Retrieval Parameters Setting Module, Retrieval Result Browsing Module and Feedback Module, and further introductions to these modules are made.Finally, a summary of our work are made and a prospect of the following research work are given.更多还原