【作者】 阎冬明；

【导师】 赵杰；

【作者基本信息】 河北大学 ， 通信与信息系统， 2008， 硕士

【摘要】 数字图像作为重要的多媒体信息载体之一,由于其包含的信息量大,具有文字所不能替代的优势,得到越来越广泛的应用。对庞大的数字图像集合进行有效的利用和管理成为近年来的一个研究热点。基于内容的图像检索(CBIR)就是利用图像本身具有的内容,如颜色,形状,纹理等特征完成图像匹配,快速有效地找到目标图像的技术。本文对颜色特征,形状特征及相关反馈几个方面进行了研究,实现了基于内容的图像检索。本文提出一种基于颜色特征及形状特征的两层图像检索算法。论文的主要工作和成果如下：1、对图像信息特征提取方法进行了研究。分别介绍了颜色、纹理、形状特征表述方法以及相似性度量匹配等。本文算法选取全局及局部颜色直方图来表述图像颜色特征。既利用了全局颜色直方图便于提取、计算的优点,又应用了局部颜色直方图包含图像空间分布信息的优点。2、利用小波变换具有的良好局部分析特性用于获取图像信息。本文应用二维小波的多尺度分析以及小波变换的模极大值提取了图像目标的边缘形状特征,并利用七阶不变矩得到形状边缘特征的描述。3、针对图像检索相关反馈技术进行研究。将支持向量机(SVM)理论应用在图像检索相关反馈中,并对SVM参数选择对反馈结果的影响进行了讨论。实验结果表明,两层图像检索算法结合相关反馈技术的应用能取得良好的检索效果。更多还原

【Abstract】 As a kind of important multimedia information carriers, the digital image is more and more widespread for it contains the huge information and has the superiority which the text can not substitute.How to use and manage the huge digital image set effectively has become a research hot spot in the recent years. The content-based image retrieval (CBIR) which is a technology matches the similar image effectively and rapidly from the image target set according to the intrinsic feature or content such as color, sharp, texture etc. This thesis researches the image color features, shape features and relevance feedback technology, and achieves the content-based image retrieval.This thesis proposes a kind of two-level image retrieval algorithm based on the color features and the shape features. The main research work and production are as follows:1、Research on the methods of image feature extraction. We introduce several kinds of feature expressions of color, texture and shape. The proposed algorithm of the thesis uses the global and local spatial color histograms to describe the image color features. It is not only using simple calculation and fast retrieval advantage of the global color histograms, but including the image spatial distribution information in the local color histograms.2、Research on wavelet transformation for extracting the image features. The wavelet transformation has the good partial analysis feature. This thesis introduces the method to extract the edge shape feature of the image object, which uses two-dimensional multi-scale wavelet analysis and the wavelet transform modulus maxima value and describes the image shape feature by the seven invariant moments.3、Research on the image retrieval relevance feedback. Apply support vector machines (SVM) theory to image retrieval relevance feedback and discuss the influence of the choice of different kernel functions.The experimental results show that the two-level image retrieval algorithm combined with the relevance feedback technology can get a better retrieval results.更多还原